By Ed Dombrowski | Freedom Grid | April 2026

“This is the thing nobody is talking about.”

I see that sentence on LinkedIn about 400 times a week. It is always, without exception, something that absolutely everyone is talking about. AI. Bitcoin. Return-to-office mandates. The latest thing some CEO said on a podcast. The bar for “nobody is talking about this” has never been lower.

So I need you to understand: I post about energy infrastructure every single day. Nuclear fuel supply chains. Transformer bottlenecks. Grid interconnection queues. Cooling water constraints. My feed, and probably yours if you’re reading this publication, is saturated with energy content. I scroll through it at 5 AM with coffee and mild regret.

You know what I have never seen? Not once? A single LinkedIn post about geothermal energy. (Im not sure I have seen any on Substack either.)

Not. One.

And here’s where it gets genuinely strange: geothermal is further along than nuclear. By years. Not in theory, not in some DOE whitepaper projection, not in a pitch deck with hockey stick charts. In actual megawatts connecting to actual grids in the actual year 2026. Three companies are proving it through three completely different technical approaches, using workforce and equipment that already exist, with bipartisan political support. And the market is barely paying attention.

This is the rare case where the cliche is earned. Let me show you why.

Subscribe now

The Signal

Geothermal energy is undergoing the same technological transformation that turned shale gas from a geological curiosity into the dominant force in American energy. The companies leading it are applying horizontal drilling, hydraulic fracturing, and subsurface engineering techniques perfected by the oil and gas industry over decades. They are pointing those tools at hot rock instead of hydrocarbons. And the results are arriving faster than the nuclear industry’s best-case deployment timelines.

Three companies illustrate the three distinct paths this transformation is taking.

Ormat Technologies: The Business That Already Exists

Before we get to the startups, start with the company that’s been doing this for sixty years and still somehow flies under the radar. Ormat Technologies (NYSE: ORA) operates approximately 1,340 MW of geothermal capacity across six countries. Trailing twelve-month revenue sits around $944 million. Product segment backlog hit $295 million as of Q3 2025. Full-year net income came in at $123.9 million.

Those are not speculative numbers. That’s an operating business with decades of cash flow behind it.

What makes Ormat particularly interesting right now is the company’s positioning at the intersection of legacy geothermal and next-generation development. In February 2026, Ormat signed a portfolio PPA of up to 150 MW with Google through NV Energy to support data center operations. In January 2026, Ormat co-led a $97 million Series B round in Sage Geosystems. The company also partnered with SLB to accelerate enhanced geothermal system development. Ormat is negotiating approximately 250 MW of additional hyperscaler contracts at rates exceeding $100 per MWh.

That last detail matters. When your existing geothermal fleet starts commanding data-center pricing, the revenue trajectory changes. Ormat has guided toward 2.6 to 2.8 GW of generating capacity by end of 2028, roughly doubling the current portfolio. This isn’t a company waiting for the future. It’s a company whose existing business just became dramatically more valuable because AI compute drove electricity demand through the ceiling.

Fervo Energy: Green Fracking Is Not a Typo

Fervo Energy is building what will be the world’s largest enhanced geothermal project at Cape Station in Beaver County, Utah. Phase I, rated at 100 MW, is in advanced commissioning right now and will begin delivering power to the grid in 2026. The full 500 MW buildout is contracted and scheduled for completion by 2028. All 400 MW of remaining capacity is sold to investment-grade buyers including Southern California Edison and Shell Energy.

The technology is, at its core, fracking. Fervo drills horizontal wells into hot, dry rock formations at depths around 15,000 feet where temperatures reach roughly 500 degrees Fahrenheit. They fracture the rock in controlled stages to create fluid pathways, circulate water through the engineered reservoir, and bring the heated water to the surface to generate power through Organic Rankine Cycle turbines supplied by Turboden, a Mitsubishi Heavy Industries subsidiary.

Two days ago, Fervo signed a three-year framework agreement with Turboden for ORC units supporting over 1.7 GW of geothermal development. That is not a pilot program. That is supply chain infrastructure for a multi-gigawatt buildout.

The cost curve tells the story. Fervo’s first wells took about a month to drill. Current drilling times are averaging mid-teens in days. Per-well costs dropped from $9.4 million to $4.8 million. The company has raised approximately $1.5 billion since 2017, including a $462 million Series E in December 2025 and a $421 million non-recourse project finance loan in March 2026. Google, Baker Hughes, Mitsui, and AllianceBernstein are investors. The Cape Station project area alone can support over 5 GW of development.

Read that progression again. Drilling times cut by more than half. Well costs cut by roughly half. A project finance loan from institutional lenders. Those are not the metrics of an experiment. Those are the metrics of an industry reaching commercial liftoff.

Buy Me a Coffee

Sage Geosystems: The Underground Battery

Sage Geosystems is doing something slightly different, and slightly wild. The company’s proprietary Pressure Geothermal technology captures both heat and pressure from the earth’s subsurface, enabling two applications from the same well: power generation and long-duration energy storage.

Think of it as pumped hydro turned upside down. Traditional pumped hydro stores energy by moving water uphill into a reservoir and releasing it downhill through turbines. You need a mountain for that. Sage’s approach pumps water into deep wells under pressure, where the rock’s natural elasticity stores the energy. When demand peaks, they release the pressurized water back through turbines. The company’s CEO Cindy Taff, a 35-year Shell veteran, describes it as an underground lung that expands and contracts with the earth’s elasticity.

Sage’s first commercial energy storage facility, a 3 MW system at San Miguel Electric Cooperative in Christine, Texas, is operational and cycling. The company has a 150 MW power generation agreement with Meta, targeting first power in 2027. And through the Ormat partnership announced in the Series B round, Sage will deploy its first commercial power generation facility by drilling beneath Ormat’s existing geothermal plants to access deeper, hotter rock formations. Existing turbines, existing grid connections, existing permits. The Ormat partnership accelerated Sage’s commercial timeline by roughly two years.

The storage economics deserve attention. Sage’s levelized cost of storage sits between 2 and 4 cents per kWh, competitive with lithium-ion batteries at shorter durations and substantially cheaper at durations beyond eight hours. If the technology scales, this is a potential structural alternative to grid-scale battery storage that doesn’t depend on lithium supply chains or rare earth mineral imports. And it works in geologies found across the entire United States, not just the geothermal-friendly West.

Why It Matters: The Workforce Advantage Nobody Mentions

Nuclear energy’s biggest constraint is not regulation, not NIMBYism, not fuel supply (though HALEU enrichment is its own problem). It’s workforce. Building and operating nuclear plants requires specialized workers who largely don’t exist in sufficient numbers. The nuclear construction workforce has atrophied over decades of stalled projects. Training them takes years. Every SMR timeline you see in a pitch deck quietly assumes a labor force that has not yet been assembled.

Geothermal has the opposite situation. There are approximately 937,000 active oil and gas wells in the United States. Roughly 768 drilling rigs are active today, down from nearly 2,000 in 2012, which means significant idle capacity. The Bureau of Labor Statistics puts O&G industry employment around 133,000, down from over 200,000 a decade ago. These are workers who know how to drill wells, manage subsurface pressure, operate fracking equipment, read seismic data, and work in hostile downhole environments.

The skills transfer is not theoretical. Fervo’s team is built from oil and gas professionals. Sage’s founding team brings over 200 years of combined Shell drilling experience. When CNN asked former Deputy Energy Secretary David Turk about workforce readiness for geothermal, his answer was blunt: we have a ready-made workforce. The DOE’s GEODE initiative, launched in September 2024 and administered by Project InnerSpace, is specifically designed to accelerate the transfer of O&G knowledge, technology, and labor into geothermal applications.

After three decades in telecom infrastructure at AT&T, I can tell you: the most expensive, time-consuming, schedule-destroying variable in any infrastructure buildout is the workforce. Not materials. Not permitting. People. When the labor force already exists and the equipment is interchangeable, you just skipped the hardest part of the deployment curve. That is the structural advantage that makes geothermal’s timeline fundamentally different from nuclear’s.

The Evidence

Hard data: Fervo Cape Station Phase I is in advanced commissioning with power delivery to the grid expected in 2026 (company filings, ABB supply agreement, Turboden commissioning confirmation). Ormat operates 1,340 MW globally with $944 million in trailing revenue (SEC filings, Q4 2025 earnings). Sage’s 3 MW storage facility is operational in Christine, Texas, and cycling on the ERCOT grid (company announcement, DOE partnership documentation). Fervo’s 1.7 GW Turboden framework agreement was announced April 7, 2026. Ormat’s 150 MW Google PPA was announced February 17, 2026. Sage’s $97 million Series B closed January 21, 2026.

Soft signals: Rhodium Group analysis finding enhanced geothermal could provide nearly two-thirds of new data center demand at or below current operator costs by 2030. DOE projections showing next-generation geothermal costs falling below nuclear and natural gas with CCS by 2035. Enhanced geothermal market forecast at 18.2% CAGR through 2030. These projections require continued cost curve improvements that are trending favorably but not yet locked in.

What we don’t know yet: Whether Fervo’s drilling cost reductions continue at the same pace as they scale beyond Cape Station. Whether Sage’s pressure storage technology performs at scale beyond the 3 MW demonstration. Whether the seismic risk profile of EGS can be managed consistently across diverse geologies. The Pohang, South Korea incident in 2017, where an enhanced geothermal project triggered a magnitude 5.5 earthquake, still looms over the industry. Fervo mitigates this with staged fracking and continuous seismic monitoring, but the risk is real and worth naming.

Timeline

Near-term (2026-2027): Fervo Cape Station Phase I delivers 100 MW to the grid. Sage deploys first commercial power generation facility at an Ormat site. Ormat continues hyperscaler PPA negotiations and portfolio expansion toward 2.6-2.8 GW target.

Medium-term (2027-2029): Fervo completes Cape Station to full 500 MW capacity. Sage delivers 150 MW to Meta data centers. The 1.7 GW Fervo-Turboden framework begins large-scale unit deliveries. Ormat’s blend-and-extend contracts begin repricing at higher rates.

Structural horizon (2030+): If cost curves hold, enhanced geothermal becomes competitive with natural gas baseload across broad U.S. geographies. Workforce transfer from oil and gas accelerates as fossil fuel demand plateaus. Geothermal storage competes directly with grid-scale batteries at durations beyond four to six hours.

For context: the first commercial SMR in the United States does not have an operational date. NuScale’s Idaho project was canceled. Oklo is pre-revenue. The fastest nuclear timeline anyone credibly projects is late this decade, and nuclear timelines have a long history of slipping. Geothermal is not waiting.

What Could Change This

Seismic events. A significant earthquake linked to an enhanced geothermal project in the U.S. would set the entire industry back by years, regardless of whether the technical response was adequate. Public perception of induced seismicity is the single biggest existential risk for EGS deployment.

Drilling cost plateau. Fervo’s cost curve has been impressive, but it’s based primarily on one project site. Extending those economics across diverse geologies, deeper formations, and hotter temperatures may prove more difficult than Cape Station’s favorable conditions suggest.

Permitting bottlenecks. Federal permitting for geothermal projects on public land can take over a decade. Fervo’s Cape Station expansion took roughly three years through federal review. Without permitting reform, deployment timelines could stretch well beyond what the technology itself supports.

Nuclear acceleration. If the nuclear industry solves its workforce and manufacturing constraints faster than expected, it would compete directly for the same baseload PPAs that geothermal is currently winning. Our read: this is the least likely risk on the list.

The Bottom Line

Geothermal energy has what nuclear currently lacks: a proven workforce, off-the-shelf drilling technology, bipartisan political support, commercial projects delivering power to the grid in 2026, and a cost curve that is moving in the right direction at commercial scale. Three companies are proving three different approaches work. The market has barely noticed.

Nuclear gets the conference keynotes. Geothermal gets the grid interconnection agreements. One of those matters more than the other.

Signal or Noise

Signal or Noise: Signal. Geothermal is the most unrecognized baseload solution in Phase 3b of the AI infrastructure buildout.

Fervo’s Cape Station commissioning, Ormat’s hyperscaler PPAs, and Sage’s commercial storage deployment confirm execution against timelines, not vaporware.

The market is pricing nuclear narrative at a premium while geothermal delivers baseload power with existing workforce and infrastructure. That pricing gap will not last.

Disclosure: The author may hold positions in companies discussed in this article. This publication does not provide investment advice.

Share

Subscribe now

Meta’s data center in Newton County, Georgia opened in 2018. It took six years for anyone to notice what was happening.

The facility consumes 500,000 gallons of water per day. That is 10% of the entire county’s daily water supply, flowing into one building owned by one company. Neighbors started reporting dry wells. Sediment in the taps. Water pressure dropping. Nine more companies then applied to build in Newton County. Some requested up to 6 million gallons per day, which is more than the county uses in total, which would have been a genuinely interesting math problem for the water authority to solve.

They didn’t solve it. The water authority director said the quiet part directly: “We just don’t have the water.”

That sentence is now being repeated in county commissioner meetings, city council chambers, and state legislatures across the country. Fifty-four local moratoriums passed across fourteen states as of March 2026. Two thousand to four thousand cancellations. Sixty-four billion dollars in blocked projects through 2024. A hundred billion delayed in Q2 2025 alone. Maine just passed LD 307, the first statewide moratorium in the country, blocking facilities over 20 megawatts through roughly October 2027. New York’s three-year moratorium bill is the strongest in the country and still moving. Bernie Sanders and Alexandria Ocasio-Cortez introduced a federal moratorium bill in March 2026.

The era of easy data center permitting is over. Water is why.

Buy me a TACO

If you find value in these articles please fuel the Signal!

THE LEGAL GENIUS OF WATER

The hyperscalers built their permitting strategies around electricity. They hired FERC lawyers, state PUC experts, transmission consultants. They know how to navigate the regulatory apparatus for power because power is federally regulated, predictable, and takes years to challenge. You can plan around a FERC proceeding. You can budget for it.

Nobody built a strategy around water because water seemed like a solved problem. You pick a site, you call the utility, you get a hookup. Done.

What the opposition discovered, slowly and then very fast, is that water is structurally different from every other infrastructure input a data center needs. Municipalities own their water utilities. States don’t control them. The federal government definitely doesn’t control them. The Clean Water Act regulates water quality, not quantity. There is no federal analog to FERC for water. A city council can refuse to connect a new customer to municipal water as basic service management, the same way they’d refuse to extend a sewer line to a property that doesn’t meet code. No federal preemption. No lengthy PUC proceedings. No appeals to FERC.

A California appeals court said it plainly in 1976: “A potential water user does not possess any absolute right to be afforded water service.” The Bolinas Community Public Utility District has maintained a water moratorium since 1971. Fifty-three years. Multiple legal challenges. Still standing.

Tucson figured out how to turn that legal architecture into a moratorium ordinance in approximately two weeks. After the city council voted 7-0 to reject Amazon’s Project Blue in August 2025, a 290-acre 600-megawatt facility that would have been Tucson Water’s single largest customer, they passed a large water user ordinance requiring City Council approval for anyone using more than 7.5 million gallons per month. Start to finish: fourteen days. Try doing anything in fourteen days with a FERC proceeding.

The ordinance is now a template. Other municipalities are copying it verbatim.

AMAZON BROUGHT A WATER REPLENISHMENT PLAN. TUCSON SAID NO ANYWAY.

This is the part of the story that should be keeping hyperscaler real estate teams up at night.

Amazon did not show up to Tucson without a plan. They came with a 1-for-1 water replenishment commitment. For every gallon consumed, Amazon would restore a gallon to the regional watershed. This is the industry’s standard response to water criticism. Microsoft made the same pledge. So did Google. So did Meta. All four major hyperscalers are “water positive by 2030” on paper.

The Tucson City Council voted 7-0 to reject the project anyway.

The Watershed Management Group’s Melanie Cooley explained the community’s position: “One-to-one is not sufficient, it doesn’t account for the uncertainty.” Translation: replenishing water somewhere else does nothing for the people whose taps go dry right now. Former Amazon water sustainability manager Nathan Wangusi said he was “hounded out” of the company for raising the same concern internally. His framing: “Carbon is a global problem. Water is more localised.”

That distinction is everything. When you promise to offset carbon emissions, the offset works globally. The atmosphere doesn’t care where the carbon reduction happens. When you promise to replenish water, the aquifer in the watershed you restored does not help the neighbor whose well ran dry six miles from your facility. The community in question doesn’t get their water back. They get a press release about a restoration project somewhere else.

Kyrsten Sinema personally lobbied on behalf of a $2.5 billion data center project in Chandler, Arizona. Former senator. Presumably some influence. The city council rejected it 7-0 in December 2025, specifically citing water.

The playbook that worked for a decade is not working anymore.

THE META EXCEPTION AND WHY IT DOESN’T SAVE YOU

Newton County, Georgia is instructive in the other direction. Meta got in. The facility is operational. By any conventional measure, that’s a win for Meta and a data point that the opposition can be beaten.

Look closer.

Meta’s facility opened in 2018, before anyone had developed the water opposition playbook, before Food and Water Watch became the de facto national movement coordinator, before Good Jobs First started tracking moratoriums, before there were 188 organized opposition groups across 40 states. Meta got in because they got there first. The moratorium wave that swept nine surrounding counties in September 2025 happened because Meta was already there and the neighbors could see exactly what 500,000 gallons per day looks like when it comes out of the community’s water system.

Meta’s Georgia data center did not defeat the opposition. It created it.

For any hyperscaler looking at Newton County as a model, here is the actual lesson: the project that gets in before the community figures out what’s happening generates the exact conditions that prevent the next project from getting in at all. The opposition is now organized, funded, experienced, and armed with a replicable legal toolkit. They have model ordinances. They have national coordination. They have a 54-moratorium track record and accelerating momentum.

Any project going through approval today is going through it in a completely different environment than 2018. The litigation exposure alone is substantial. Calvert County, Maryland is requiring a 24-month environmental impact study before even voting on a moratorium. Multiple Indiana projects face active lawsuits. The Linn County, Iowa situation shows that when municipalities try to impose conditions, developers sometimes attempt to bypass them via city annexation, which generates its own legal fight and public backlash.

Easy permitting is over. The question now is how companies build a water strategy before they pick a site, not after they’re already in a city council meeting.

THE FEDERAL GOVERNMENT IS NOT COMING TO SAVE YOU

The Trump administration has declared data centers critical national security infrastructure. The July 2025 executive order streamlines federal permitting for facilities over $500 million or 100 megawatts. The China competition narrative is the explicit justification: Beijing has committed $125 billion to AI computing, is building 250-plus AI data centers, and the administration wants to accelerate American buildout.

Here is what the executive order cannot do: it cannot make Tucson give Amazon water.

The December 2025 executive order, even as it pushed maximum federal authority over AI policy, explicitly carved out data center infrastructure from preemption, preserving state and local authority over siting, water-use ordinances, and zoning. This was not an oversight. Congress rejected a 10-year moratorium on state AI preemption 99 to 1 when it came up as an amendment. Republican governors objected. The bipartisan nature of the water opposition (55% Republican, 45% Democrat in polling) makes preemption politically radioactive regardless of who’s in the White House.

There is a theoretical pathway through the Defense Production Act. A March 2026 DOJ Office of Legal Counsel opinion established that DPA orders can preempt contrary state law under the Supremacy Clause, and water resources are technically within the DPA’s scope under an old executive order. Nobody has ever used it this way. It has never been tested in court. And the political consequences of telling a drought-stricken Arizona community to give its municipal water supply to a hyperscaler in the name of beating China would be, in the words of one legal analysis, “catastrophic.”

The federal cavalry is not coming. The industry needs a water strategy. Several categories of companies exist specifically to provide one.

THE COMPANIES THAT SOLVE THIS

The water problem in data center infrastructure breaks into several distinct technical challenges, and there are companies with real revenue addressing each of them.

Zero-liquid-discharge systems are the most complete solution. ZLD technology treats wastewater to the point where no liquid effluent leaves the facility at all. Every drop that goes in either evaporates as process steam or gets recovered for reuse. For a data center in a water-stressed region trying to get a permit past a hostile city council, zero-liquid-discharge is the strongest possible technical argument: the facility does not consume the community’s water supply in any meaningful sense. The technology exists. Companies are deploying it. The market for industrial water treatment and ZLD specifically is growing as semiconductor fabs, battery gigafactories, and now data centers face increasingly strict water regulations. Energy Recovery (ERII) is the most relevant public company in the water efficiency space with infrastructure-grade revenue.

Advanced cooling systems that reduce or eliminate evaporative water consumption are the second major category. Traditional data center cooling uses cooling towers that evaporate water to reject heat. Liquid cooling, rear-door heat exchangers, and direct-to-chip cooling systems eliminate or dramatically reduce that evaporation. The liquid cooling market nearly doubled in 2025 to approximately $3 billion and is projected to reach $7 billion by 2029. Only 45% of data centers still run purely on air cooling. Companies in this space include Vertiv (VRT), which makes liquid cooling infrastructure, and SPX Technologies (SPXC), whose HVAC and cooling businesses serve data center thermal management. Microsoft’s zero-water-evaporation chip-level cooling deployment, started in August 2024, claims 125-plus million liters saved per facility annually.

Water recycling and reclaimed water infrastructure is the third category. Some data centers are moving to use municipal recycled water rather than potable water for cooling, which removes them from the political fight over drinking water supplies. This requires infrastructure to deliver reclaimed water to the site, treatment systems to bring it to the quality needed for cooling applications, and agreements with municipal water authorities. Companies building water infrastructure and treatment systems broadly benefit from this trend. Xylem (XYL) is the most relevant public name here, focused on water technology, treatment, and efficiency systems.

Industrial water treatment and monitoring is the fourth category. Regulations on water discharge, consumption reporting, and environmental impact are proliferating. Colorado’s SB 25-280 creates data center certification requiring water stewardship strategies. California’s SB 887 mandates CEQA review. Oklahoma’s statewide moratorium bill specifically requires water supply studies. Every one of those requirements creates demand for monitoring equipment, treatment systems, and compliance infrastructure.

The investment thesis across these categories is the same: the companies providing solutions to the water constraint have real revenue, real customers, and durable demand driven by regulation that is accelerating, not retreating. What they do not yet have is broad recognition that water is Phase Three of the AI infrastructure buildout, not an edge case or a local political curiosity.

WHY THE EQUITY PRICING IS WRONG

Here is where Freedom Grid’s analytical framework applies directly.

The Narrative versus Execution Gap is the publication’s core thesis: markets recognize infrastructure constraints 12 to 36 months before the revenue from solving those constraints appears in financial statements. Right now, the water constraint is visible to anyone paying attention to the moratorium map. Fifty-four passed moratoriums. Fourteen states. Maine’s statewide law. New York’s three-year bill moving through the legislature. A hundred billion in delayed projects in a single quarter. The narrative is there for anyone who looks.

The equities that solve the water problem are priced like the market hasn’t looked yet.

Companies like Energy Recovery, Xylem, and the water-related segments of Vertiv and SPX are not trading as AI infrastructure beneficiaries. They are trading as industrial and water utility companies with steady revenue and modest growth expectations. The re-rating that happens when the broader market connects “water is the permitting veto” to “companies that eliminate water consumption get every project through” has not happened. That gap is the opportunity.

This is the best kind of early position: companies with existing revenue models, customers who already pay them, and a catalyst (the water opposition movement) that is accelerating on its own without requiring any particular investment thesis to be correct. The moratoriums are happening regardless of whether any investor notices. Maine’s LD 307 passed regardless. The Sanders-AOC federal bill was introduced regardless. The 188 organized opposition groups across 40 states are organizing regardless.

The market will figure out that Phase Three of the AI buildout runs through water. When it does, the companies that solve the water problem will look like the power companies looked in 2023 before the grid constraint became consensus. The difference is you can see it coming this time.

THE BOTTOM LINE

The era of picking a site, calling the water utility, and breaking ground is over. Fifty-four communities have figured out that a water hookup moratorium is faster, cheaper, and more legally durable than any other tool available to block a data center. Maine put it in state law. New York is about to. The federal government explicitly declined to preempt local water authority despite declaring data centers national security infrastructure, because the politics of telling drought communities to give water to Big Tech are simply impossible.

The hyperscalers that move first on water solutions get the permits. The ones that keep showing up with replenishment pledges and good intentions keep getting 7-0 votes against them from city councils that have watched their neighbors’ wells run dry.

And the companies building the technology that makes those water pledges actually credible, the ZLD systems, the liquid cooling infrastructure, the reclaimed water treatment equipment, the monitoring and compliance platforms, are sitting at valuations that reflect none of this.

Phase Three is water. The market hasn’t priced it yet.

Signal or Noise: Signal. Fifty-four passed moratoriums, $100 billion in delayed projects in a single quarter, and the first statewide law in Maine constitute hard evidence that water has become the primary permitting constraint for data center development, not a regional quirk. What confirms it is the legal architecture: municipalities have exercised water authority for over fifty years without federal override, and Congress just voted 99-1 against preemption. What it means for the thesis is that the companies solving the water consumption problem are in the same position power infrastructure companies were before the grid constraint became consensus, which means the re-rating is a question of when, not if.

Disclosure: The author holds positions in companies discussed in this publication. This publication does not provide investment advice.

Share

THE WEEK

On Thursday, April 2, a French container ship called the CMA CGM Kribi transited the Strait of Hormuz under armed Iranian escort, paid in Chinese yuan and cryptocurrency, and continued on its way. The same day, France co-vetoed the UN Security Council resolution that would have authorized force to reopen the strait. Iran has been building a permanent toll booth on 20 percent of the world’s daily oil supply. France just became its first Western paying customer. Reasonable people can disagree about a lot of things in geopolitics. The Art of the Deal is apparently available in Farsi now.

That is the macro frame for the week. Here is the infrastructure frame: the AI power buildout thesis produced one of its most confirmed weeks on record at the exact same moment the credit environment got materially worse.

Quanta Services walked into an investor day and put a $2.4 trillion addressable market on the board. Microsoft, Meta, Crusoe, and Google all committed to gigawatt-scale off-grid gas generation: not letters of intent, not announcements, construction activities. The nuclear policy stack got four new hard confirmations on top of an already-unprecedented foundation. Bloomberg Green put five-year transformer lead times on the front page with the GE Vernova CEO on record. And Phase 3 water infrastructure recognition jumped seven percentage points in a single week, the fastest single-week gain this system has ever recorded, driven by a 98-page JPMorgan report on energy which included data center water demand that somehow failed to mention a single water infrastructure equity. Ninety-eight pages. Zero tickers. The gap between what the research confirms and what the market has priced is the entire opportunity.

Two things are true simultaneously and understanding both is what this publication exists for. The demand driving this buildout has never been more confirmed. The credit environment financing the early-stage companies in it has never been more hostile. Those two facts do not cancel each other out. They sort the portfolio into two very different buckets, and this week the sort got sharper.

Subscribe now

THE WEEKLY VERDICT

The thesis strengthened. The risk profile also escalated. The verdict is not a contradiction, it is a bifurcation.

Kill Chain 4, the tracking mechanism for credit tightening cutting off capital from infrastructure companies, crossed into Stage 2 this week on four independent hard data points. Blue Owl’s $5.4 billion private credit fund received redemption requests at 4.4 to 8.1 times its quarterly capacity limit. Sector-wide private credit outflows hit $11 billion-plus in Q1 and Q2 combined. Carson Block disclosed a public short position against the investment-grade and high-yield bond ETFs. High-yield credit spreads remain at or through the escalation threshold. What started three weeks ago as an Apollo-specific stress signal is now confirmed as a sector-wide private credit dislocation.

The critical distinction that keeps this from being a full thesis break: Kill Chain 1, the demand collapse scenario, is decelerating. Microsoft, Meta, Crusoe, and Google collectively committed to gigawatts of generation construction this week. Hyperscaler capital expenditure commitments for 2026 remain at $750 billion with no language shift toward “optimization.” The Doom Loop, the scenario where demand and capital collapse together requires both conditions simultaneously. Demand is not falling. Credit stress without demand collapse is painful and compressive for stock prices. It is survivable for companies with real revenue and contracted backlogs. It is dangerous for pre-revenue companies that need to raise capital right now.

The Marks Cycle position holds at Early Fear with a binary capitulation risk flag if Hormuz extends past April. Super-cycle recognition and implementation both advanced across three of four phases this week, with Phase 3 water showing the sharpest single-week move in system history.

Buy Me a Coffee

If this weekly summary provides value consider keeping me caffeinated!

THE ARC

Monday through Wednesday: Iran rewrites the negotiation. The week opened with the formal confirmation that Iran’s five-point peace proposal includes, as a non-negotiable condition, international recognition of Iranian sovereignty over the Strait of Hormuz. Not a temporary governance arrangement. Not a negotiated fee structure. Sovereignty. The ceasefire negotiation quietly transformed from “when does Hormuz reopen” into “does the world formalize Iranian control over a global shipping chokepoint permanently.” These are not equivalent negotiations.

Markets spent Tuesday running the de-escalation narrative anyway. The S&P 500 posted its best single day since May 2025, up roughly 3 percent, on ceasefire speculation. By Thursday, when the speculation did not materialize into anything actionable, the tape gave most of it back. The physical oil market, meanwhile, did not oscillate at all. Physical Brent spot traded at approximately $141 per barrel all week against futures at $109. The equity market is repricing on headlines. The oil market is pricing on barrels. Those two views cannot both be right, and the barrel count is not improving.

Thursday: The credit escalation went sector-wide. Blue Owl’s redemption request disclosures moved the private credit stress story from “Apollo had a bad quarter” to “the sector has a structural problem.” At 4.4 to 8.1 times the quarterly redemption cap, these are not investors taking a bit of money off the table. These are investors trying to exit and being told they can get out at 5 cents on the dollar per quarter. The secondary market for private credit instruments has extremely wide bid-ask spreads, which is the polite way of saying liquidity is effectively broken for anyone who needs to sell.

The buildout moved past the announcement phase. While the macro tape was repricing risk, the actual physical infrastructure buildout had one of its most active weeks on record. Construction activities, not announcements. Microsoft and Nscale are building 1.4 gigawatts of off-grid gas in West Virginia. Microsoft, Chevron, and Engine No.1 are building a $7 billion Texas gas plant. Meta expanded an El Paso project from $1.5 billion to $10 billion and added seven Entergy plant commitments. Crusoe confirmed 933 megawatts in Goodnight, Texas and 900 megawatts in Abilene for Microsoft. Google acquired land in West Virginia. These are not letters of intent. This is what construction permitting, site acquisition, and equipment procurement look like from the outside.

Quanta put a number on the opportunity, and the market sold the stock. PWR held its investor day and confirmed a $2.4 trillion total addressable market with EPS guidance of $21.60 to $26.75 by 2030 and 15 to 20 percent revenue CAGR. The backlog stands at a record $43.98 billion. Management has beaten its own targets for ten consecutive years. The stock fell 2.95 percent on investor day because Iran and credit spreads were the macro story. A company with a decade of execution credibility just described the next five years as its biggest opportunity ever. The market said “noted, selling.”

Maine made it official. The state House enacted LD 307, blocking data centers larger than 20 megawatts through November 2027. This is the first statewide moratorium in US history. The Northeast is now a legally mandated no-go zone for large-scale data center development. Capital is not being redirected because Texas is attractive — it is being redirected because Maine, Huron County Michigan, Denver, Manitowoc County Wisconsin, Plain Township Ohio, Anchorage Alaska, and Phillipsburg New Jersey are all actively closing the door. Texas and the Southeast are the confirmed beneficiaries. The rerouting is not a headwind for the thesis. It is a tailwind for portfolio names concentrated in those geographies.

Nuclear had its best policy week in the thesis history. Part 53 gets an effective date of May 3. The One Big Beautiful Budget Act locks in nuclear tax credits through 2032. DOE deploys $750 million in SMR loans. Texas opens a $350 million Advanced Nuclear Fund. BWX Technologies lands a $1.6 billion federal contract for a Tennessee uranium plant. Diablo Canyon’s license renewal proceeds. DOE clarified that only 30 of 356 previously terminated clean energy projects were actually de-obligated — approximately 4 percent versus the fears that had been running much higher. SPARK continues under a new name. What the Iran war has accomplished in the nuclear narrative debate is something that years of climate advocacy could not: converting it from an environmental argument into a national security argument. Bipartisan support for nuclear has never been more evident.

MACRO AND STRUCTURAL POSITION

Macro Regime: Bear-Leaning | COST SHOCK / CREDIT INTERACTION — Run 3 Marks Cycle Position: EARLY FEAR | Binary CAPITULATION risk if Hormuz extends past April Change vs Prior Week: Regime updated from STAGFLATION INSTITUTIONAL CONSENSUS — WTI above Brent inversion and sector-wide BDC gating add a credit interaction layer on top of the commodity shock

Super-Cycle Position:

Phase 1 (Power) Recognition 75% / Implementation 38%

Phase 2 (Cooling) Recognition 62% / Implementation 29%)

Phase 3 (Water) Recognition 42% / Implementation 15%

Phase 3b (Long Term Baseload) Gap: Qualitative — strongest nuclear policy/regulatory/financial stack in thesis history; not separately quantified this run

The regime label change reflects what is actually happening at the structural level. Stagflation is a monetary description. What this week added is a physical scarcity signal: WTI crude at $111.54 trading above Brent at $109.03. That inversion does not happen in normal environments. Combined with physical Brent spot at $141 against futures at $109, you have a commodity market pricing sustained disruption while the financial market prices a diplomatic resolution. One of those will be wrong.

Credit conditions are running their fourth consecutive week at structural tightening threshold. The bifurcation between Fortress names and Financing-Dependent names is now five consecutive weeks active and classified as structural. CME FedWatch sits at 77.5 percent probability for no Fed change through year-end. The 10-year Treasury pulled to 4.31 percent on Good Friday on Iran-Oman protocol news.

Henry Hub at $2.82 to $2.89 per million BTU against European LNG at $19 to $24 is an 8-to-9 times differential — the strongest fuel cell economics this system has ever tracked. That number flows directly from Hormuz disruption compressing European energy options while the US domestic gas market remains insulated. The Phase 3 recognition jump to 42 percent from 35 percent is the fastest single-week gain this system has recorded for any phase. Equity market recognition of ERII and XYL remains estimated at approximately 5 percent. The gap is 37 points and widening.

INFRASTRUCTURE STACK

Phase 1 and the off-grid inflection. The definitional debate about whether behind-the-meter generation was a transitional workaround or a permanent architectural decision ended this week when three of the world’s largest technology companies simultaneously committed to gigawatt-scale construction. The cause is straightforward: the public grid cannot serve always-on AI load on the timeline AI load requires. The interconnection queue has 2,300 gigawatts waiting. Transformer lead times are now five years, confirmed by the GE Vernova electrification CEO in Bloomberg Green. The market’s response to an impassable queue and five-year equipment lead times is to stop trying to connect to the grid and build generation directly. That is not a workaround. That is a structural decision by companies with trillion-dollar balance sheets.

The DC Electrical Modules reclassification from Jefferies adds an important nuance: less than half of the 2026 DC pipeline is contractually binding, and two-thirds of the non-binding portion faces demand destruction risk rather than delay risk. GEV, POWL, and PWR serve the committed hyperscaler-tier construction. That distinction matters for how you read the backlog data.

Phase 2 is past the argument stage. The GB200 chip’s thermal design power of 125 kilowatts per rack closed the cooling architecture debate at the physics level. Air cooling stops working around 20 kilowatts. The chips are deploying at six times that. VRT’s $15 billion backlog at 2.9 times book-to-bill is what that physics looks like in contract form. JCI’s $18 billion record backlog with 40 percent of orders from data centers is the same signal from a different angle.

Phase 3 is at an inflection the equity market has not priced. Three weeks ago, Phase 3 water recognition sat at 30 percent. This week it is 42 percent. Maine’s moratorium, Michigan’s AG contested case process on Google’s water consumption, and the framing of data center water use as a shared public resource are all doing the recognition work simultaneously. JPMorgan’s 98-page data center water demand report confirmed the physical constraint and failed to mention ERII or XYL. That gap between the research and the equity is where the return is.

Phase 3b: the nuclear stack. The cause-and-effect chain runs through the Iran war. Energy security has replaced climate policy as the primary political motivation for nuclear deployment. That shift matters because energy security arguments survive partisan transitions in ways that climate arguments do not. Part 53 provides the regulatory pathway. OBBBA provides tax credit durability through 2032. The DOE loan, Texas fund, and BWXT Tennessee contract provide the capital stack. BWXT’s $7.3 billion backlog growing 50 percent year over year is the evidence that policy is translating into contracted revenue.

Cross-phase: 26 days to FERC. The RM26-4-000 large-load interconnection rulemaking deadline is the single most consequential regulatory binary for grid-connected infrastructure in the near-term calendar. The off-grid routing confirmed this week is partly a market hedge against an uncertain FERC outcome. If April 30 compresses interconnection timelines materially, one accelerant for off-grid moderates. If it does not, off-grid remains dominant architecture indefinitely.

THE NARRATIVE LANDSCAPE

The most important narrative gap in the market right now is between what equity prices are pricing on Iran and what the physical oil market is pricing. Equity markets oscillated on diplomatic headlines this week, up 3 percent Tuesday, back down Thursday. The physical oil market, where actual barrels trade for actual delivery, sat at $141 all week against $109 in futures. That $32 gap is the physical market saying it does not believe the diplomatic off-ramp exists on the timeline the equity market is pricing.

The mechanism: when buffer oil inventories exhaust on Bordoff’s “next few weeks” timeline, futures must converge upward toward physical prices automatically. No new military action required. No diplomatic breakdown required. Just inventory math. Goldman estimates $14 to $18 per barrel of risk premium is already embedded in futures. The physical market is embedding $32. The difference between those two numbers is the convergence risk sitting in the curve right now.

Where newsletters are running narratives that are correct: off-grid gas as dominant DC power architecture, transformer crunch as a structural multi-year constraint, and nuclear momentum as the strongest political story in years. All three confirmed by hard evidence. The newsletter sweep has been ahead of consensus on all three for months.

Where the market is running a narrative that is getting ahead of itself: AI efficiency reducing infrastructure demand. Google’s TurboQuant memory compression is real. The conclusion that it threatens the infrastructure buildout is not supported by the evidence yet. The measurement event is whether energy cost per AI query falls 50 percent or more annually. That data lives in Micron’s next earnings call. None of it exists yet. When the infrastructure demand data contradicts the efficiency narrative, the names that sold off on efficiency fears are the re-entry.

Two emerging narratives before they reach consensus. First: the possibility that the US exits the Iran conflict without formal Hormuz reopening, accepting de facto Iranian governance without explicit recognition. The Iran-Oman bilateral protocol and France’s transit payment both point toward a parallel governance architecture being built alongside the formal negotiations. Second: the Pentagon’s objection to approximately 7.5 gigawatts of wind capacity on national security grounds. If maintained, this removes competing intermittent capacity from the mix and strengthens the economics for nuclear, gas, and geothermal — a Phase 3b tailwind most of the market is not tracking.

THE EDGE

1. The hidden Scenario C trigger. The April 6 deadline passed as Scenario B: Brent at $109.03, below the $120 emergency threshold. But the secondary path to Scenario C does not require a new military decision. Physical Brent at $141 against futures at $109 creates a convergence dynamic that runs on inventory math alone. When buffer stocks drain, futures converge upward automatically, without anyone firing a shot. The April 13 EIA inventory release is the next read. If storage drawdown confirms the depletion trajectory, the gap between Scenario B and Scenario C conditions narrows to weeks.

2. US exit without Hormuz reopening as a permanent governance outcome. The Iran-Oman bilateral protocol was framed between two sovereign states before any multilateral peace deal. France paying in yuan while vetoing the UN resolution suggests countries are making their own calculations about the permanence of the current arrangement. If the US exits without formal Hormuz reopening, the IRGC toll system becomes de facto recognized international commerce infrastructure. The energy security urgency for domestic dispatchable generation accelerates permanently. None of this is priced.

3. Pentagon wind obstruction — 7.5 GW removed from the capacity mix. DOD objections are blocking approximately 7.5 gigawatts of wind capacity on national security grounds. If maintained, competing intermittent capacity is removed from the generation mix. Indirect beneficiaries: nuclear, gas, and geothermal. The off-cycle W5 note triggers when the DOD formally responds to challenged permits.

THESIS SCORECARD

STRENGTHENED: GEV — Sold out through 2030; off-grid gas demand surged at GW scale; steel above $1,000/T hardens moat against new entrants POWL — $1.6B backlog, 1.7x book-to-bill; Bloomberg transformer feature delivers mainstream institutional confirmation PWR — $2.4T TAM, $43.98B record backlog, $21.60-$26.75 EPS 2030 from a team with a 10-year track record of beating its own numbers BWXT — $7.3B backlog +50% YoY; $1.6B Tennessee federal contract; nuclear stack strongest in thesis history BE — Henry Hub at $2.82-2.89 vs global LNG $19-24 is 8-9x; strongest fuel cell economics this system has tracked

INTACT: VRT — $15B backlog, 2.9x book-to-bill; GB200 locks in liquid cooling path; Q1 earnings April 29 is Phase 2 gate JCI — $18B record backlog; orders +40% YoY; 40% data center; executing ECL — Phase 3 recognition at 42% and accelerating; patience capital XYL — FY2026 guidance below consensus flagged; Q1 gate; Phase 3 recognition 42% vs ~5% equity recognition = the asymmetry ORA — Google NV Energy 15yr 150MW PPA confirmed; revenue floor established; rate-pressured but thesis intact LEU — $900M DOE task order; OBBBA through 2032; patience capital to 2027+ OKLO — $2.5B cash buffers KC4 exposure; Part 53 tailwind; patience capital BAM — Most exposed Fortress name to private credit contagion; thesis intact; monitoring SPXC — HVAC +22%, D&M +43% record, FCF $294M at 90% conversion; Q1 earnings gate

CHALLENGED / WATCHLIST: FCEL — Highest KC4 exposure in portfolio; private credit structurally compromised sector-wide; Q4 Adversarial Engine required before mid-April PRIM — Fixed backlog -18.2% while diesel, steel, and labor compound; lump-sum contract exposure is the break mechanism; Q1 gate ERII — Back-to-back air pockets; KC4 Stage 2 tightening desalination project finance; Phase 3 recognition building underneath

REMOVED: FLNC — Fully exited April 1. Three consecutive quarters of the same execution failure is not bad luck. Damage, not Delay. Done.

The bifurcation is the sharpest in portfolio history — five consecutive weeks, now structural. Fortress names are generating contract-level validation through backlogs, book-to-bills, and EPS guidance from management teams with execution track records. Financing-Dependent names are navigating a credit environment that is actively hostile to anyone who needs to raise capital.

PRIM is the name to watch most closely heading into Q1 earnings. Fixed backlog declining 18.2 percent while input costs compound means the revenue base absorbing those costs is shrinking faster than the costs are rising. Lump-sum contracts are the specific mechanism through which this becomes a margin problem rather than a revenue problem. If escalation clause coverage in the remaining backlog is not comprehensive, the thesis breaks.

FCEL has the best fuel economics it has ever had and the worst financing environment it has ever faced. The Henry Hub differential to global LNG is a real competitive advantage. The problem is that advantaged economics you cannot finance are a research footnote, not a business. The Q4 Adversarial Engine run before mid-April is not optional.

RISK REGISTER

KC4 Stage 2: 18 days to the first gate. Stage 3 looks like margin compression showing up in actual reported earnings — not credit spreads, not BDC disclosures, but line items in income statements from the companies this portfolio holds. The gate opens April 22 with GEV’s Q1 earnings. If GEV reports margin pressure without offsetting pricing power, the KC4 Stage 3 cascade begins. If GEV demonstrates pricing power and backlog conversion, Stage 2 becomes more manageable. Everything about how Q1 earnings season unfolds flows from that first print.

The physical-futures convergence is the macro risk that does not need permission. Every other risk scenario requires a decision — a military action, a policy change, a management call. Physical-futures convergence requires only inventory math. Bordoff’s timeline sits on the same calendar as Q1 earnings season. If those events coincide — futures above $120 simultaneous with Q1 earnings showing EPC margin compression — the compound event is worse than either alone. Pre-decide the response to Brent above $120 before it happens.

KC3: Reroute confirmed, Bab el-Mandeb is the new watch item. Maine is REROUTE, not DAMAGE. Capital moving to Texas and the Southeast is a tailwind for EPC names concentrated in those geographies. The new monitoring flag is the Bab el-Mandeb statement from Iran’s parliamentary speaker. If Iran extends its toll collection logic to the second major Arabian Peninsula chokepoint, the scope of the energy security disruption expands qualitatively. Soft-to-Hard evidence right now. Emergency W2 canary run triggers if any Iranian state action accompanies the rhetoric.

NEXT WEEK’S SETUP

Monday, April 7 — FRED HY OAS Master II release plus first post-April 6 trading day. If HY OAS confirmed above 3.50 percent sustained, cascading system runs within 24 hours. Brent’s Monday open is the primary Scenario B vs. convergence read.

Week of April 13 — EIA inventory data. Bordoff’s buffer exhaustion timeline materializes here. Storage drawdown confirming depletion trajectory moves futures convergence above $120 from tail risk to working assumption.

Before April 18 — Q4 Adversarial Engine on FCEL. Hard deadline.

April 14 — Phillipsburg NJ moratorium final vote. Watch Georgia and Indiana for next state template.

April 20 — Denver CO moratorium first reading. Mountain West no-go zone expansion signal.

Before April 22 — Full quarterly cycle P1 through Q6. Maine, OBBBA, Iran war nuclear narrative, and Part 53 all represent material updates since the last quarterly run.

April 22 — GEV Q1 earnings. Gas turbine orders vs. 18 GW threshold, margin trajectory, production ramp commentary. KC4 Stage 3 gate opens here.

April 29 — VRT Q1 earnings. Book-to-bill above 1.5 times and CDU lead time commentary are the Phase 2 classification gate. First report without quarterly orders — management language is the primary signal.

April 29-30 — Powell’s final FOMC meeting. 77.5 percent no-change probability.

April 30 — FERC RM26-4-000 deadline. 26 days. Joint W4 and W5 off-cycle triggers immediately upon any publication.

April-May — PRIM, PWR, SPXC Q1 earnings. Gross margin trajectory and escalation clause coverage resolve the lump-sum contract risk classification.

May 3 — NRC Part 53 effective. First developer applications confirm whether the rule is operationally usable.

May 5 — POWL Q2 FY2026 earnings. New orders vs. $400 million threshold.

The One Thing: The physical-futures oil divergence at $32 per barrel does not need a new war decision to close. It needs inventory math. Watch the April 13 EIA storage data. If drawdown confirms Bordoff’s timeline, the window between Scenario B and Scenario C conditions narrows to weeks. Everything else — earnings gates, regulatory deadlines, moratorium votes — sequences around whether that convergence happens before or after the portfolio has time to run its overdue quarterly cycles.

THE BOTTOM LINE

1. Iran built a permanent toll booth on the Strait of Hormuz, operationalized it before the law authorizing it even passed, and got France as its first paying Western customer on the same day France vetoed the resolution to stop it. Hard evidence CMA CGM Kribi April 3 transit; Iran-Oman bilateral protocol; Iran parliament toll legislation — Escalated. The ceasefire negotiation is no longer about reopening Hormuz. It is about whether Iranian governance of Hormuz gets formalized permanently. That negotiation is longer, harder, and structurally more bullish for domestic energy infrastructure than the market’s 30 percent April 30 probability implies.

2. Physical oil at $141 against futures at $109 is a $32 convergence risk that triggers Scenario C conditions without a single new military action. Hard evidence multiple sources; Bordoff April 1 timeline; IEA April characterization — New, critical. Buffer inventory exhaustion forces futures upward automatically. The April 13 EIA storage release is the next read. Pre-decide the response before the number prints.

3. Kill Chain 4 is confirmed Stage 2 — private credit is structurally compromised sector-wide but the Doom Loop is not activated because demand is screaming. Hard evidence Blue Owl 21.9%/40.7% redemption requests; $11B+ sector outflows; Carson Block public short — Escalated. $750 billion in committed hyperscaler capex is intact. Credit stress without demand collapse is survivable for Fortress names. FCEL’s Q4 Adversarial Engine run before mid-April is not optional.

4. The infrastructure thesis produced its most institutionally confirmed week on record — PWR’s $2.4T TAM and $43.98B backlog, BWXT’s $7.3B backlog at +50% YoY, the strongest nuclear policy stack in thesis history, and Phase 3 water recognition jumping 7 points to 42 percent in a single week. Hard evidence — PWR Investor Day; BWXT contract data; NRC Part 53; OBBBA; DOE $750M; Texas $350M — Escalated across all phases. The thesis is not theoretical. It is showing up in backlogs, book-to-bills, and federal contracts.

5. PRIM’s fixed backlog declined 18.2 percent while every input cost it carries is compounding at the worst combined rate in the system’s history. Hard evidence — company data; diesel $5.401/gal; steel second consecutive week above $1,000/T; EPC labor Run 6 — Escalated negative. Lump-sum contracts are the break mechanism. Q1 earnings are the classification gate. This is the holding most likely to change classification in the next 30 days.

Signal or Noise: The week’s dominant signal is that the infrastructure buildout is proceeding at GW-scale with or without the grid, nuclear is now a national security argument with bipartisan legs, and the only credible thesis threat is the credit channel not demand.

What confirms it: $43.98 billion in committed EPC backlog, five-year transformer lead times in mainstream press, and Phase 3 institutional recognition moving at the fastest pace this system has recorded.

What it means for the thesis: the causal chain from AI compute to power to heat to cooling to water to firm baseload held without a crack this week. The portfolio holds through the credit pressure because the demand it is serving is real, contracted, and accelerating.

Disclosure: The author may hold positions in companies discussed in this publication. Freedom Grid provides infrastructure analysis, not investment recommendations. Readers should conduct their own research before making investment decisions.

Share

Subscribe now

The S&P was getting worked over. Tariff headlines were doing their thing. Somewhere in Washington, someone was floating the idea of doing something somewhere that would definitely fix everything. The financial press was busy. Very busy. Lots of very important noise to cover.

Meanwhile, Quanta Services held its Investor Day and quietly described the physical backbone of the next decade of American economic infrastructure in more specific terms than any government report, think tank white paper, or congressional hearing has managed in recent memory.

Nobody noticed. The stock was down on the day.

This is a Signal or Noise piece. The signal is Quanta. The noise is everything else.

Buy me a Coffee

If getting the signal delivered to your door every day adds value, help keep the signal coming!

WHO QUANTA IS AND WHY YOU SHOULD CARE

If you have not been following Freedom Grid, Quanta Services (PWR) is probably not a household name for you. It should be.

Quanta is the largest specialty contractor in North America for critical infrastructure. Power lines, substations, pipelines, data center construction, renewable energy projects, industrial facilities. If electricity moves through it, into it, or because of it, there is a reasonable chance Quanta built some portion of the system that makes that possible. The company has 69,000 employees, an 80%-plus self-perform rate (meaning they do the actual work rather than subcontracting it out), and a backlog that has grown from $7.6 billion in 2015 to $44 billion at the end of 2025.

That last number is the one that matters most.

A backlog is contracted future revenue. It is not an estimate. It is not a forecast. It is work that customers have committed to pay for and that Quanta has committed to execute. When a company’s backlog grows from $7.6 billion to $44 billion over a decade, that is not a marketing slide. That is demand made visible.

Quanta is what Freedom Grid calls a cross-phase signal. Every phase of the AI infrastructure buildout: power generation, transmission, cooling, data center construction, grid modernization, nuclear, eventually requires engineering, procurement, and construction firms to actually build the thing. Quanta is the throughput constraint for the entire stack. Their backlog is one of the single best leading indicators of real infrastructure demand that exists anywhere in public markets.

Which makes what they said at Investor Day worth paying attention to.

THE SIGNAL: WHAT QUANTA ACTUALLY SAID

The total addressable market Quanta outlined through 2030 is $2.4 trillion. That breaks into roughly $1.5 trillion in utility and grid work (grid modernization, transmission expansion, fuel diversification, weather resilience, electrification) and $885 billion in large load work (data centers, advanced manufacturing, public infrastructure, industrial facilities).

For scale: the entire US federal budget for fiscal year 2025 was approximately $6.75 trillion. Quanta is describing a privately funded infrastructure capital cycle worth more than a third of that, compressed into five years, concentrated in the physical systems that power the economy.

The growth rates attached to those numbers are not projections someone made up in a boardroom. They are derived from utility capital expenditure plans, hyperscaler announced commitments, industrial reshoring investments, and electrification timelines that are already locked into regulatory filings and earnings guidance across dozens of publicly traded companies. Utility sector work alone is up 70% from 2020 to 2030. Total TAM growth over that same period is 140%.

The data center and large load segment is growing the fastest. Quanta is guiding 20 to 25% annual growth in that segment through 2030. AI-driven load accounts for roughly 30% of the future TAM they outlined. At $13.5 million per megawatt of inside-the-fence data center construction, the math on announced hyperscaler capacity additions alone generates numbers that are difficult to fully absorb.

The EPS targets are equally direct. Quanta guided to $21.60 to $26.75 in earnings per share by 2030, on 15 to 20% annual EPS growth, with free cash flow of $10 to $12 billion over the period. Revenue growing at 7 to 10% compounding annually. EBITDA margins expanding from 10.1% to 10 to 11%.

These are not aspirational numbers from a startup. Quanta has compounded revenue at 14%, EBITDA at 19%, and EPS at 25% over the past decade. The 2030 targets represent a deceleration from recent performance, not an acceleration.

THE LABOR SIGNAL BURIED IN THE DECK

The most underreported signal from the entire presentation is the workforce section, and it deserves more attention than it will get.

Quanta is spending more than $100 million annually on skilled trades training. They have committed an additional $250 million in incremental training investment over the next ten years. They operate more than 100 proprietary training curricula. They added 11,000 employees in 2025 alone and are targeting 3,000 additional net hires per year going forward.

This is not a human resources initiative. This is a company that has looked at the skilled labor market, concluded that the supply of qualified workers cannot keep pace with the demand their backlog represents, and decided to manufacture the workforce themselves rather than wait for someone else to solve the problem.

The infrastructure buildout has a labor constraint. Every electrician, pipefitter, heavy equipment operator, and structural worker that Quanta trains and retains is a unit of throughput capacity that would not otherwise exist. When Quanta says their workforce is redeployable across transmission, generation, data centers, industrial projects, and storage, what they are describing is a human capital moat. You cannot replicate 100 proprietary training curricula and a $350 million workforce development commitment in a quarter.

The labor shortage is not a talking point. It is a binding physical constraint on how fast the infrastructure gets built. Quanta is the only company at scale actively trying to expand the constraint rather than just operate within it.

THE NOISE: EVERYTHING THAT MOVED THE STOCK INSTEAD

Here is a partial list of things that have affected infrastructure stock prices over the past several weeks that have no structural connection to Quanta’s backlog, the utility TAM, or the pace of the AI infrastructure buildout:

The administration’s ongoing negotiation posture on trade policy, which changes weekly and has generated more headlines than clarity. Interest rate expectations, which move on every CPI print regardless of what long-duration infrastructure capital cycles actually require. The Iran war and its effects on oil prices, which matter for energy costs but do not change whether a data center needs to be connected to the grid. The Gulf situation broadly. A reported diplomatic overture toward Cuba, the relevance of which to transformer manufacturing lead times is not immediately apparent. General recession anxiety, which is real and legitimate as a macro concern but does not change the fact that $44 billion in contracted backlog does not evaporate when the VIX spikes.

The market has been treating infrastructure stocks as leveraged bets on the macro environment rather than as what they actually are: companies with multi-year contracted revenue visibility executing against a structural demand cycle that predates the current administration, will outlast the current administration, and is driven by physical constraints that no executive order has yet managed to repeal.

Electrification is not a political position. Data centers require power regardless of who controls the Senate. Aging grid infrastructure does not get younger because tariffs went up. The interconnection queue does not clear because someone held a press conference.

The noise is loud right now. It is genuinely loud, and some of it (sustained inflation, a credit crunch, a real recession) could delay portions of the buildout. Delays are not thesis breaks. The demand does not go away. It waits.

WHAT COULD ACTUALLY CHANGE THIS

Intellectual honesty requires naming the real risks, not just the noise.

Sustained customer capital constraints, meaning utilities unable to finance their capex programs because credit conditions deteriorate materially, could push project timelines right. This is worth monitoring through utility earnings and credit spreads, not through daily macro headlines.

Permitting and regulatory delays remain the single most persistent structural drag on transmission and generation projects. This is a real constraint with a long history of extending project timelines regardless of demand levels.

Tariffs on imported electrical equipment, particularly transformers and switchgear components manufactured overseas, could increase project costs and compress margins. Quanta explicitly named tariffs as a risk signal in their presentation. This one is worth watching with specificity, not in aggregate.

Project cancellations are possible in a severe demand destruction scenario. Quanta’s backlog is contracted, but contracts can be renegotiated. The 2008 cycle showed that even contracted infrastructure work is not completely immune to credit crisis conditions.

None of these risks are new. None of them showed up in the Investor Day presentation for the first time. They are known quantities being managed by a company that has compounded EPS at 25% for a decade while navigating all of them.

THE BOTTOM LINE

Quanta Services just told anyone willing to listen that the physical infrastructure buildout supporting the AI economy, the energy transition, and industrial reshoring is the largest construction cycle in modern American history, that they are positioned at the center of it, that their backlog is $44 billion and growing, and that they expect to generate $10 to $12 billion in free cash flow through 2030 executing against it.

The stock was down on the day because the macro tape was ugly.

That is the signal versus noise problem in its most concentrated form. The company described the decade. The market reacted to the week.

Signal or Noise: Signal. Quanta’s $44 billion backlog and $2.4 trillion TAM through 2030 are not estimates; they are contracted demand and utility capital plans already embedded in regulatory filings across the industry. What confirms it is the labor investment: a company spending $350 million to manufacture its own workforce does not do that because it is uncertain about the demand. What it means for the thesis is that the EPC throughput constraint is real, Quanta is the clearest expression of it in public markets, and the market is currently pricing macro noise instead of structural signal.

Disclosure: The author holds positions in companies discussed in this publication, including Quanta Services. This publication does not provide investment advice.

Share

Subscribe now

Full disclosure: I am an AT&T employee. The analysis and opinions in this article are entirely my own and do not represent AT&T in any capacity. AT&T has not reviewed, approved, or endorsed anything written here. I am an investor commenting on a publicly reported development, not a company spokesperson.

If my research provides value to you - Keep me caffeinated and producing signal!

Buy Me a Coffee

Jensen Huang does not do subtle.

At GTC 2026 in March, the Nvidia CEO stood on stage and told the telecom industry that it is sitting on the next great AI infrastructure opportunity. The claim: telecom operators and distributed cloud providers collectively operate roughly 100,000 distributed network facilities worldwide, with enough spare power capacity to support more than 100 gigawatts of AI compute, quote, “over time.”

One hundred gigawatts. For context, the entire current US grid has roughly 1,200 gigawatts of installed generating capacity. Huang was suggesting that idle telecom infrastructure represents something like 8% of that, just waiting to be repurposed for AI.

That number deserves scrutiny. So does the underlying thesis, which, even after you apply the appropriate haircut to the headline figure, is more structurally significant than most infrastructure observers are giving it credit for.

THE ACTUAL PROBLEM NVIDIA IS SOLVING

Before the hype, the physics.

AI inference (running a trained model to generate a response) is increasingly hitting two walls when it runs from centralized hyperscale data centers. The first wall is latency. Voice AI, physical robotics, video analysis, and real-time decision systems need responses in under 500 milliseconds. When the request has to travel from a device to a cloud data center in Virginia and back, that latency budget gets eaten alive by geography. The second wall is economics. Centralized inference at scale is expensive. Every token generated in a hyperscale facility carries the cost of premium data center real estate, power at industrial rates, and interconnection bandwidth.

Edge inference solves both problems by moving the compute closer to where the request originates. The question has always been: where exactly do you put the edge compute, who owns the real estate, and how do you power it?

Nvidia’s answer at GTC was: telecom companies already own the real estate. They built tens of thousands of facilities over the past century, specifically positioned to serve every metro area and neighborhood in the country. A lot of that space is sitting partially empty as the industry migrates customers from copper networks to fiber and wireless.

The hardware Nvidia built for this is the RTX PRO 6000 Blackwell Server Edition GPU, designed specifically for edge deployment. It is not the H100 or GB200 monster that populates hyperscale AI training facilities. It is a workstation-class inference chip engineered to fit in conventional rack space, operate within legacy power constraints, and manage heat without requiring the kind of liquid cooling infrastructure that most older telecom buildings were never designed to support.

Comcast ran a real benchmark. Four RTX PRO 6000 GPUs running a voice small language model at Comcast’s Philadelphia headquarters hit 42,362 tokens per second at burst traffic, an 81% throughput gain over baseline. Cost per inference ran 52.8% cheaper at normal traffic and 76.1% cheaper during bursts compared to centralized deployment.

That is not a slide deck projection. That is a measured result from a field trial. The economics of edge inference for latency-sensitive workloads are real.

THE 100 GW NUMBER NEEDS A HAIRCUT

The headline figure is a long-run ceiling, not a current reality, and the gap between the two is significant.

Central offices were built to house telephone switching equipment running on 48-volt DC power at roughly 5 to 10 kilowatts per rack. A modern GPU inference rack runs at 30 to 60 kilowatts minimum. The existing power distribution architecture: the power distribution units, the uninterruptible power systems, the cooling units, was engineered for a completely different density.

Floor load is the second constraint. Legacy telephone switch buildings are rated for heavy equipment distributed across raised floors. A GPU rack with liquid cooling can exceed 3,000 pounds in a footprint the building was never designed to support as a point load.

The third constraint is cooling. Even the RTX PRO 6000, which Nvidia specifically designed to operate within legacy building constraints, generates heat that older air-cooled telecom rooms cannot manage at GPU rack densities without retrofit investment.

The realistic near-term addressable capacity across the industry is not 100 gigawatts. Credible estimates cluster around 15 to 30 gigawatts, concentrated in the larger regional hub sites and metro central offices that already meet or come close to data center grade standards. For a site to be economically viable for edge AI deployment, it typically needs to support at least 500 kilowatts of power. That threshold filters out the majority of small remote switching facilities immediately.

The 100 GW figure requires full industry retrofit over a decade or more and billions in capital expenditure. It is a vision statement, not a current inventory.

The underlying thesis, however, survives the haircut. Even 15 to 30 gigawatts of distributed AI compute, positioned inside metropolitan areas with sub-10-millisecond latency to hundreds of millions of devices, is a structurally significant infrastructure layer that does not exist today in any other form.

WHERE AT&T ACTUALLY SITS

AT&T’s position in this story is more textured than a keynote summary conveys.

The company sold 74 central office buildings totaling 13 million square feet to Reign Capital in January 2025 for more than $850 million, specifically because those facilities were underutilized stranded assets as copper network retirements accelerated. AT&T retained revenue-sharing rights in any redevelopment of those properties, which means the company preserved exposure to upside even on assets it no longer owns outright. The sale made sense as a capital allocation decision at the time, and the revenue-sharing structure shows the company was thinking carefully about optionality.

What AT&T did not sell is the rest. The company retains a large portfolio of central offices and network facilities beyond those 74 buildings, including the larger regional hubs and metro facilities that are the most likely candidates for AI Grid deployment based on the power and floor load criteria.

The active deployment is already happening. AT&T, Cisco, and Nvidia are running a live AI Grid at AT&T’s Discovery District in Dallas, connecting AT&T’s IoT core network with Cisco’s mobility services platform and RTX PRO 6000 GPUs at the edge. The use cases are physical AI: delivery robots, smart cameras, real-time environment sensing. This is not a lab experiment. It is running in production on a commercial campus.

AT&T also announced at MWC 2026 a last-mile enterprise AI connectivity offering built with AWS, bringing AT&T’s fiber and 5G infrastructure directly into AWS environments. That positions AT&T as the connectivity layer for enterprise AI workloads that need low-latency access to cloud AI services.

The company has committed $250 billion through 2030 to transform from a legacy telecom utility into what COO McElfresh describes as “the highway upon which commerce and AI workloads traverse.” Supporting that framing, AT&T is separating its financial reporting in 2026 into an advanced connectivity P&L and a legacy P&L, creating a visible scorecard for the transformation that the market can track.

I find that last detail genuinely interesting as someone who watches infrastructure capital cycles. Separate reporting segments are how management teams force internal accountability on transformation initiatives. When the new revenue line has its own income statement, the executives running it cannot hide behind consolidated numbers.

THREE WAYS THIS GENERATES REVENUE

The AI Grid thesis has three distinct monetization paths for AT&T, with very different timelines and confidence levels.

The most certain path is connectivity. AI Grid workloads need low-latency fiber or 5G between edge nodes and between edge and central AI factories. AT&T’s 32 million fiber locations, its 255-million-person 5G footprint, and its 400,000-plus lit fiber business buildings are the physical transport layer those workloads ride. McKinsey estimates the fiber connectivity opportunity for new data center construction alone at $30 to $50 billion globally by 2030. AT&T does not need to operate a single GPU to capture a share of that market. It just needs to be the pipe. That revenue is already accumulating.

The second path is edge compute colocation and GPU-as-a-service at retained facilities that meet the 500-kilowatt threshold. AT&T would house and operate RTX PRO 6000 nodes, charge enterprise customers by the compute hour or the token, and leverage its existing network relationships to bundle connectivity and compute into a single offering. The economics require careful site selection; only the right facilities pencil. But the business model is proven. Akamai has been running distributed infrastructure for 30 years and is the furthest along in the GPU edge build-out, with thousands of RTX PRO 6000 units deployed across more than 4,400 locations. AT&T is earlier on that curve, but it has fiber relationships with enterprise customers that Akamai does not have.

The third path is AI-RAN, which means integrating AI inference directly into the radio access network at or near cell sites. T-Mobile is piloting this with Nvidia and Nokia. The architecture would allow AI processing at the cell tower level rather than backhaul to a core facility. This requires substantial base station hardware replacement across the network and depends on 6G standards that are not yet finalized. Material revenue from AI-RAN is a 2028-plus story.

THE RISK THE KEYNOTE GLOSSED OVER

The technical gaps are real but solvable with capital and time. The competitive risk is more structural.

The hyperscalers are not passive observers. AWS is already working with AT&T through the AWS Interconnect arrangement. Google Cloud is deploying RTX PRO 6000 GPUs through its own edge infrastructure. Microsoft Azure has distributed deployments. The question is whether hyperscalers deploy on AT&T’s physical infrastructure, making AT&T the landlord collecting rent, or whether they build competing edge infrastructure in the same geographies, making AT&T the incumbent getting disintermediated.

Telecom’s history with platform transitions is not encouraging. Between 2012 and 2025, global mobile data traffic grew more than 50% annually while telecom service revenues grew barely 1% per year. The hyperscalers captured the value. The telcos built the pipes.

Nvidia’s AI Grid strategy is, in a real sense, an attempt to help telcos break out of that trap by giving them a compute product to sell rather than just a connectivity product. Whether AT&T can execute the build-out faster and more competently than hyperscalers can replicate AT&T’s geographic footprint is the open question. It does not have a definitive answer yet.

TIMELINE AND WHAT TO WATCH

Material revenue contribution from AI Grid for any telco, including AT&T, is a 2027-plus story at earliest. The field trials and pilots are real. The unit economics are validated. The path from pilot to revenue line requires site retrofits, customer acquisition, operational capability building, and competitive positioning. None of which happen in a quarter.

The signals worth watching over the next 12 to 24 months: whether AT&T’s new advanced connectivity P&L segment shows meaningful GPU-related revenue growth by late 2026; whether the Dallas Discovery District deployment expands to additional markets; whether AT&T announces additional Nvidia or Cisco AI Grid deployments at retained central office sites; and whether AT&T’s enterprise fiber bookings begin reflecting AI workload connectivity as a distinct revenue category.

The stranded asset question has a clean answer. The buildings AT&T still owns, the retained central office portfolio beyond the 74 sold to Reign Capital, are a genuine latent asset for this deployment. The company is already acting on it. The revenue timeline is longer than the keynote implies, and the execution risk is real. Neither of those facts changes the underlying structural opportunity.

THE BOTTOM LINE

Nvidia identified a real infrastructure gap and built hardware specifically to close it. The economics of edge inference for latency-sensitive AI workloads are validated by actual field benchmarks, not slide projections. AT&T has the physical footprint, the fiber connectivity, the enterprise relationships, and the active deployments to be a material participant in that build-out.

The 100 GW headline needs a substantial discount. The underlying thesis does not.

Signal or Noise: Signal, with a timeline caveat. The physical and economic case for distributed AI inference on telecom infrastructure is real and already generating measurable results in the field. What confirms it is AT&T’s separate advanced connectivity P&L reporting beginning in 2026; watch that scorecard. What it means for the thesis is that the connectivity layer of the AI infrastructure stack has a second monetization path beyond fiber to data centers, and AT&T’s existing footprint puts it directly in the path of that buildout.

Disclosure: The author is an AT&T employee. The views expressed here are the author’s own and do not represent AT&T. The author holds positions in companies referenced in this publication, including AT&T equity through mandatory retirement plan participation. This publication does not provide investment advice.

Share

Subscribe now

My father never sold Force Protection.

If you don’t remember it, Force Protection was a defense contractor that made armored hulls for the undersides of military vehicles, the kind of thing that kept soldiers alive when a roadside bomb went off in Iraq. The stock was trading around $2 or $3 when my dad bought it. The war was escalating. The contracts were real. He was right.

It went to $80.

He never sold a share. Not at $40. Not at $60. Not at $80. Not on the way back down through $60, $40, or $20. By the time he decided the war might actually end someday and the contracts might not be eternal, the stock had already done most of its traveling in the wrong direction. He rode it from $2 to $80 and most of the way back to zero because he didn’t have a system for the one question that actually mattered: is this still a signal, or is it noise?

My father was a retired physician. He was not a stupid man. He just didn’t have the tools.

That question has cost my family a lot of money over the years. Building a system to answer it reliably is why Freedom Grid exists.

I got my Series 7 and Series 63 in 1990. I was 24 years old, working for a firm called Chatfield Dean in Stamford, Connecticut. If you’ve seen the movie Boiler Room, you’ve already met Chatfield Dean. You come in, you study for free until you pass your exams, and then they drop a phone book in front of you. That’s it. That’s the whole onboarding. Here are your clients. They are strangers who have not consented to knowing you. Good luck.

The morning meeting told you what you were selling that day. Penny stocks. Sketchy IPOs. Products that had no business being pitched to anyone with a pulse and a savings account. Then you picked up the phone and started dialing. Three hundred calls a day was the expectation. If two or three people said yes, that was a good day. The other 297 hung up on you, which, in retrospect, was the correct response. The Series 7 exam, by the way, teaches you nothing about evaluating a company. It teaches you the rulebook: what’s legal, what’s a federal crime, what disclosures you have to read before pitching the garbage. Actual analysis was not part of the curriculum. You were not there to think. You were there to dial.

I left before the FTC investigation landed. Barely.

Shearson Lehman was supposed to be the legitimate version of all that. Real salary, real firm, nicer office, actual business cards with your name spelled correctly. And it was better, technically. Nothing illegal. But it was still 300 calls a day, still pitching what they told you to pitch, still grinding through a phone book of strangers who had not asked to hear from you. Different carpet, same job. I gave my notice after two years, decided this was not a way to spend a human life, and went to sell Harley-Davidsons. I have never once regretted that decision.

My father retired as a physician around 1997, got a computer, discovered online investing, and fell completely in love with it. He was a genuinely good stock picker in the way that a lot of people are genuinely good stock pickers during a bull market, which is to say he was right often enough to feel like he had a system when what he mostly had was a tailwind.

He had a real nose for interesting situations. Force Protection is the proof. That was a legitimate insight. He just had no way of knowing when the insight had expired.

His other great passion was investment newsletters. Not the legitimate research ones. The ones that send you three paragraphs of breathless buildup about a stock that is going to change everything, and then tell you the name is available exclusively to paid subscribers for $19.99 a month.

My father would not pay the $19.99.

Instead, he would turn it into a project. He’d pore over the clues in the free preview. He’d cross-reference. He’d eliminate candidates. He’d call me with progress updates like I was his research assistant on a case that was definitely going to crack any day now. This could go on for weeks. He was a man who had, at various points, significant money riding on actual stocks in the real market, and he would not spend twenty dollars to get the name of the thing he was trying to invest in. The detective work was apparently the point. The $20 was a matter of principle.

The last one, when he was around 87, he called me absolutely certain he’d cracked it. The newsletter had been teasing a company that held a foundational 5G patent, the kind every smartphone manufacturer had to license, an annuity stream that would essentially last forever. He’d figured out which company it was. He’d done the math. Four months of work. He was ready.

It was a Canadian cannabis stock.

I looked it up. I looked at him. I said, Dad, I promise you this is not the company Apple is paying perpetual royalties to for 5G licensing. He was undeterred. He’d done four months of research on this. He was confident.

I bought $1,000 of it. Not because I believed him. I very much did not believe him. I bought it because I wanted something to talk to him about. A shared position, even an obviously doomed one, is a conversation.

He died not long after. I held the stock through the bankruptcy out of some sentimental stubbornness I cannot fully defend. When I finally went to close the account, E*TRADE informed me I owed them four dollars.

The stock had gone so negative that I finished the transaction in debt to the brokerage. I don’t know how that’s mechanically possible. I have a Series 7. I still don’t know how that’s mechanically possible. But that’s the number: negative four dollars, and a lesson I’d been paying for in various forms since 1990.

Around 2005, inspired by my father’s enthusiasm and against my better judgment, I pulled about $50,000 out of the index funds I’d been building since I was 25 and started running individual stocks. I read the Investor’s Business Dailymethodology cover to cover. I ran every position through their framework. I did the work honestly.

The market did not care.

The 2007-2008 credit crisis doesn’t grade on effort. It just goes down. I got divorced around the same time; she got half the 401k, the market took most of what was left. Somewhere in that stretch of compounding financial misery I made a decision that held for the next seventeen years: done with individual stocks. Permanently. The index funds had been right all along. I made my peace with average market returns and got on with my life.

Index funds from 2008 to 2025. Boring. Reliable. Growing. Exactly what they were supposed to be.

Then AI happened.

I’m a systems thinker. It’s just how my brain works. When I finally started using AI seriously in March 2025, something clicked that I hadn’t expected. I didn’t just use it to ask questions. I started building with it. First came what I called my human operating system: a structured framework tracking health, nutrition, workouts, medications, mental health, finances. A personal dashboard built to cut through the noise of running a life.

And then I noticed the compute. The system I’d built was burning through resources at a level that made me ask a simple question: where does all of this actually live? What does running AI at scale physically require?

That question led to power. Power led to the grid. The grid led to transmission constraints, transformer manufacturing backlogs, cooling infrastructure, water rights, nuclear fuel supply chains, EPC labor bottlenecks. Every answer opened three more questions, and every answer pointed to the same structural reality: the physical infrastructure required to run the AI economy has hard constraints with multi-year resolution timelines, and the market was only beginning to figure that out.

Here is what I want to be clear about before we go any further. AI is not a stock ticker. It is not a magic 8-ball you shake until it tells you what to buy. Its actual superpower is synthesizing enormous amounts of information, reams of it, faster and more completely than any single human analyst ever could. That’s what I built. Freedom Grid runs a daily signal scan, seven weekly intelligence prompts, six quarterly review prompts, and a rotating set of ad hoc deep dives when something warrants closer examination. Twenty-one coordinated prompts in total, all connected by a snapshot system that carries intelligence forward so each layer knows what the last one found. The system doesn’t make decisions. It separates signal from noise so I can.

Freedom Grid started as a few threads and some questions in the summer of 2025. By the end of the year, the portfolio had grown 27% from its June starting point. The individual stock sleeve was up 47% over that same period. That’s not the system bragging. That’s the system working.

My father spent four months trying to reverse-engineer a newsletter clue about a 5G patent company and landed on a Canadian weed stock that eventually charged me four dollars to close. I built a system that processes millions of words of infrastructure data and separates structural signals from market noise.

The tools matter.

Last November, the portfolio drew down $110,000 from its high.

I sat on my hands.

Every day I ran the scan and asked the same question: did anything break? Not the price. The thesis. Oracle announced a data center delay because they couldn’t source enough power. Energy infrastructure stocks got hammered on that news. The price action said the thesis was wrong. The analysis said the exact opposite: a hyperscaler couldn’t build fast enough because power infrastructure was constrained. That’s not a thesis break. That’s a thesis confirmation wearing a loss as a costume.

Price action is not signal. That might be the most important thing I’ve learned building this system. The market will do things that feel catastrophic and mean nothing. It will also move in ways that feel fine while something genuinely structural is rotting underneath. The entire job is separating those two things.

The portfolio recovered. Then grew substantially past the November high. By the end of February 2026 I was up 13% year to date. Then the macro environment did what macro environments do, and I’ve given back about eight points of that. I’m sitting at roughly 5% year to date as of today. The S&P 500 is down 7.5%. The Nasdaq is down 10%. I mention those numbers not to brag but because they’re the context that makes 5% mean something.

Nothing in the core thesis has broken. Some positions are deeply underwater on price. Centrus Energy is down significantly from where I bought it. But the question was never whether I bought it at $300 or $200. The question is whether the United States is going to need domestic nuclear fuel enrichment capacity as it expands its reactor fleet over the next decade. That answer hasn’t changed. If the thesis holds, the entry price won’t matter. If the thesis breaks, no entry price saves you.

That’s the whole game. Signal or noise. Thesis intact or thesis broken. Everything else is just the market being the market.

My father had good instincts for thirty years and never had the tools to act on them properly. Force Protection was a real signal. He just couldn’t tell when it expired. The 5G patent intuition, that foundational technology licensing creates durable economic moats, was actually a correct insight. He just ended up in a cannabis stock, which is a different thing entirely.

I spent two years on a brokerage floor learning to sell product and nothing about how to evaluate it. I blew up my own individual stock portfolio when I finally tried. I spent seventeen years in index funds because it was the only honest answer I had.

Now I have a different answer.

Every post on Freedom Grid exists to answer one question: Signal or Noise. Is this development a structural change to the physical infrastructure powering the AI economy, or is it noise that will look like nothing in 18 months? Starting later this week, I’m going to show you the actual portfolio mapped against the infrastructure thesis, not as stock picks, but as a real-time demonstration of how this plays out when actual money is on the line.

My father deserved this system. He would have loved it. He also would have immediately tried to use it to find the newsletter stock for free, but that’s a different problem.

Thirty years in the making. Let’s see what it’s worth.

Signal or Noise: Signal. The infrastructure constraints driving this thesis are multi-year physical realities, not narrative, and a system that held through a $110,000 drawdown without a single thesis break is worth sharing. Every post from here forward answers the same question your broker never could.

Disclosure: The author holds positions in companies discussed or referenced in this publication. Freedom Grid does not provide investment advice.

Share

Subscribe now

How This Works

Freedom Grid tracks the physical infrastructure required to power the AI economy. Not the software, not the models, not the quarterly earnings beats on GPU margins. The systems underneath all of it: the wires, the turbines, the cooling towers, the water pipes, and the nuclear fuel rods that keep the whole thing running when the sun goes down and the wind stops.

There are four phases to this buildout, and they unfold in sequence because physics demands it.

Phase 1 — Power Generation and the Grid. AI data centers need electricity at a scale the current grid was not designed to deliver. Generation capacity must expand, and transmission infrastructure must be built to move it. This is the acute constraint right now, visible in interconnection queue backlogs stretching past a decade, transformer shortages, and federal emergency orders that have started showing up with uncomfortable regularity.

Phase 2 — Cooling Infrastructure. Higher-power chips generate more heat. Rack power density in data centers is tracking toward 50 to 100 kilowatts per rack. Air cooling stops working around 20 kilowatts. Liquid cooling at this scale is not an upgrade option. It is physics.

Phase 3 — Water Infrastructure. Cooling systems reject heat through water. As data center density grows, so does industrial water demand, creating constraints around water rights, municipal supply, and zero-liquid-discharge regulations with 10 to 15-year resolution timelines. The market has not priced this yet. That gap is the opportunity.

Phase 3b — Firm Baseload: Nuclear, Geothermal, and Long-Duration Storage. Intermittent renewables cannot serve always-on AI workloads. Solar and wind are great until it is cloudy and calm at 2 AM on a Tuesday when a hyperscaler needs 500 megawatts of compute. Nuclear, geothermal, and multi-day storage fit the profile. Development timelines range from five to twenty years. That is not a flaw in the thesis. It is the nature of building the grid’s long-term foundation.

Cross-Phase — EPC and Grid. Engineering, procurement, and construction firms are the throughput constraint for every phase. Their backlogs are the single best leading indicator of real demand. When the people who actually build infrastructure are booked solid, the thesis is working.

THE WEEK

Last week this publication said high-yield credit spreads sat 22 basis points from the threshold that triggers the escalation protocols. It said the most credible near-term risk was credit tightening cutting off capital for early-stage infrastructure companies. It said the resolution event was Q1 earnings in six weeks.

The spreads crossed the threshold this week. The six weeks did not wait.

That is the bad news. Here is the rest of it.

The gas turbine supply chain that last week had “two to three validated global suppliers” received 34 gigawatts of new named project orders in a single week on top of an existing 56-gigawatt pipeline. GE Vernova, one of those two or three suppliers, is now sold out through 2030 and requires non-refundable deposits just to hold a manufacturing slot. That is not a supply chain with demand problems. That is a supply chain that cannot manufacture fast enough to meet the demand that already exists on paper. The market sold the stock anyway, because the market was busy having a different conversation.

The Nuclear Regulatory Commission finalized a new licensing rule called Part 53, a technology-neutral pathway that gives advanced reactor developers a fundamentally faster route to regulatory approval, potentially compressing timelines by a decade for some designs. The rule takes effect in early May.

The White House turned behind-the-meter power generation — data centers building their own on-site electricity supply rather than pulling from the public grid — from a market trend into codified federal law, published in the Federal Register.

Three independent hard data points arrived within 48 hours and moved the water infrastructure recognition story faster than it had moved in the prior three months combined.

And 38 countries’ economists reached documented consensus that US inflation will run at 4.2 percent in 2026. The Federal Reserve’s own projection from its March interest rate guidance was 2.7 percent. The gap between what the central bank projects and what the rest of the developed world’s economists project is 150 basis points. The institution responsible for controlling US inflation is 150 basis points behind its own curve, and that assessment now has 38 signatories. The Fed is not late. It is institutionally late.

The thesis has never been more confirmed at the fundamental level. The macro environment has never been more hostile at the financing level. Those two facts will coexist for months. Understanding why that coexistence does not break the thesis is the entire point of this publication.

One more thing before the analysis, because it deserves to be said plainly. The same administration that started this war and created the energy emergency is actively eliminating the workforce that would build the infrastructure response to it. A survey by the Associated General Contractors of America found that 28 percent of construction firms were directly impacted by immigration enforcement actions this month. Roughly 240,000 to 300,000 construction workers have been removed from legal employment eligibility through the rescission of Temporary Protected Status designations. In North Texas, active project disruption was confirmed on March 25. The tariffs this administration imposed on steel and aluminum have added 15 to 50 percent to the cost of electrical equipment. The war created urgency for domestic energy infrastructure. The labor and tariff policies are making it more expensive and slower to build. This is not commentary. It is what the data says.

THE ARC: MONDAY THROUGH FRIDAY

Monday. Iran’s rejection of the US 15-point ceasefire proposal became official. Tehran did not counter with a negotiating position. It countered with a demand for sovereign control of the Strait of Hormuz and introduced domestic legislation to formalize toll collection on the passage through which 20 percent of the world’s daily oil supply moves. Read that again slowly. A government under active US military pressure, rather than offering a path toward de-escalation, proposed a permanent revenue model for a global shipping lane it does not legally control. This is not a negotiating tactic. This is a country that has decided the leverage situation works in its favor.

The Royal United Services Institute published an assessment showing the US consumed more than 11,000 munitions in 16 days of strikes at a cost of approximately $26 billion, with at most one month of THAAD interceptors, long-range precision missiles, and strike munitions remaining in inventory. The United States entered a sustained military campaign against a regional power without first ensuring it had enough ammunition to finish one. The downstream consequences of that planning gap shape every negotiating dynamic that follows.

Brent crude pushed toward $108 per barrel by Thursday, up 65 percent year-to-date. The Trump administration’s pause on strikes against Iranian energy infrastructure expires April 5 and 6. That date became the most consequential near-term binary event in this system’s history.

Tuesday and Wednesday. The OECD published its Interim Economic Outlook on March 26. The headline: US inflation at 4.2 percent for 2026. The OECD baseline assumes the Hormuz situation moderates from mid-2026 onward. If it does not, global GDP runs another half-point lower and inflation runs another 0.7 to 0.9 percentage points higher. Goldman Sachs updated its recession probability models to 35 percent. The market is now pricing a 31.2 percent probability of zero Federal Reserve rate cuts in all of 2026. Macquarie’s call that the next Fed move would be a rate hike in the first half of 2027 was widely described as contrarian several weeks ago. It is no longer contrarian.

Last week this publication described the macro environment as “stagflation confirmed.” This week the upgrade is to “institutional consensus.” Those are different things. Confirmed means the data supports the read. Institutional consensus means 38 countries’ economists have signed their names to it in a published document that will be cited in every rate decision commentary for the next twelve months. The Fed cannot cut into 4.2 percent inflation. It cannot hike into 0.7 percent GDP growth recorded in Q4 2025, a quarter that preceded the full economic impact of the Iran conflict. The Fed is frozen. A frozen central bank with inflation running 150 basis points above its own projection is its own kind of macro environment.

Thursday. The market repriced risk. The S&P 500 fell 1.74 percent. The Nasdaq, already in correction territory at more than 10 percent below its recent peak, dropped another 2.38 percent and now sits 10.3 percent below its recent high year-to-date. The VIX volatility index reached 27 to 29, roughly double its level when the Iran conflict began. The 10-year Treasury yield hit a new cycle high at 4.43 to 4.45 percent in a pattern called bear flattening, where short-term rates rise faster than long-term rates. Bear flattening is the bond market saying, with its money, that it thinks the Fed is behind the inflation curve and will eventually have to act.

Every infrastructure holding in this system fell Thursday. Several moved 6 to 8 percent in a single session. The analytical work this week confirms that virtually all of those moves were driven by overall market selling pressure, what analysts call beta: the tendency of a stock to move with the broad market regardless of what is happening at the company. Thesis deterioration looks different. It appears as margin compression in reported financials, demand cancellations, and negative company-specific disclosures. None of those appeared in the financially strong names this week.

By Friday. Every signal from the prior week either persisted or escalated. Not one moderated.

The credit threshold that sat 22 basis points away last week was crossed. The high-yield single-B sub-index, which tracks bonds issued by companies at the riskier tier of below-investment-grade credit, reached exactly 3.50 percent on March 24 per Federal Reserve hard data. The escalation threshold is confirmed breached.

The private credit stress that was approaching three-source confirmation last week crossed it. Apollo Global Management’s $15 billion private credit fund received redemption requests at 2.2 times its quarterly capacity limit and will distribute roughly 45 cents on the dollar to investors seeking their money back. Apollo shares have fallen 24 percent year-to-date. Three independent hard sources confirming private credit stress. The escalation condition from the March 20 handoff is formally met.

For the first time in this system’s operating history, a condition called Compound Fragility was triggered: simultaneous breach of commodity input price thresholds and credit stress thresholds. Both stresses independently hurt the infrastructure buildout. Both occurring together compounds the pressure on financing-dependent companies in ways that are not simply additive. First occurrence ever. Not a routine flag.

Against all of that: NRC Part 53 finalized. White House BTM mandate published. Gas turbine demand numbers arrived at a scale that redefined what “demand surge” means in this market. And the water infrastructure thesis moved more in 48 hours than it had in the previous three months.

THE NARRATIVE LANDSCAPE

The story the market told this week: AI is becoming more efficient, so infrastructure demand is going to slow down.

Google introduced an algorithm called TurboQuant that compresses the memory requirements for running AI models. The memory semiconductor sector responded immediately. Micron, the largest US memory chipmaker, fell 22 percent in six trading days. Sandisk dropped 20 percent in five. The narrative spread into power infrastructure names. The market ran it hard.

Here is what the market is actually doing. It took a real algorithmic development at the memory layer, a genuine Stage 1 efficiency signal, and priced it as though it resolves a multi-year, multi-hundred-billion-dollar capital expenditure cycle that is already committed and spending. That is not how any of this works. The actual measurement event is whether the energy cost of running an AI query falls 50 percent or more annually, across the industry. That data comes from Micron’s next earnings call, from Nvidia’s guidance on data center power consumption, and from hyperscaler commentary on per-query energy costs. None of it exists yet. The market ran a Stage 1 data point at Stage 3 speed, which means that when infrastructure demand data catches up with the efficiency narrative and contradicts it, the names that sold off this week are the re-entry. The system does not advance the AI efficiency risk tracking beyond Stage 1 on market price action alone. That instruction stands.

The story the evidence confirmed: Behind-the-meter generation is now federal law, and the demand behind it is larger than anyone was tracking.

The White House AI Framework calls on Congress to codify “bring your own power” as federal statute, and the Ratepayer Protection Pledge is published in the Federal Register. It is law. The market trend became congressional mandate this week.

The capital evidence underneath the policy is the real story. The confirmed behind-the-meter pipeline stands at 56 gigawatts. Seventy-five percent of it is gas-fired. Named projects confirmed in a single week this week total 34 gigawatts: Fermi America’s 2-plus gigawatt off-grid gas facility in Texas; Crusoe Energy’s 5-gigawatt gas turbine order, described in industry reporting as large enough to cause delivery delays across the broader market; Arbor Energy’s 5-gigawatt baseload project; the DOE, SoftBank, and AEP Ohio 9.2-gigawatt gas facility; NextEra’s up-to-10-gigawatt development across Texas and Pennsylvania; and Nscale’s 2-gigawatt West Virginia campus anchored by Microsoft and Nvidia.

For context: the US added roughly 32 gigawatts of total new generation capacity across all sources in all of 2023. This market just confirmed 34 gigawatts of behind-the-meter demand in a single week. JPMorgan’s research now documents that 38 percent of data center operators expect to be generating partial or full on-site power by 2030. Behind-the-meter crossed from speculative to institutional consensus this week, and GE Vernova, which makes the gas turbines at the center of it, cannot build them fast enough.

The narrative gaining momentum faster than the evidence warrants: The Iran conflict is temporary.

Wednesday brought a market rally on de-escalation speculation. Thursday brought a selloff when de-escalation did not materialize. That oscillation pattern is not evidence of resolution. It is evidence of a market trying to convince itself that this has a near-term exit when the facts do not support one. The ECB president called Iran a “real shock.” The OECD baseline assumes Hormuz moderates from mid-2026, not that it resolves. Tehran’s toll collection legislation is not the behavior of a government preparing to concede leverage. Diesel is $5.375 per gallon nationally and $7 per gallon in California. Those numbers are already inside project economics and they do not reverse on a positive Sunday headline.

The narrative that has not yet reached the market: Water is the next infrastructure bottleneck.

Last week this publication described Phase 3 as quiet. This week, three independent hard data points arrived within 48 hours.

American Water, the largest publicly traded US water utility, presented at CERAWeek on a panel titled “Valuing Water: A Strategic Imperative for the Energy Industry.” This is the energy industry’s premier annual gathering. The fact that water infrastructure earned a strategic panel at CERAWeek tells you where institutional planning has arrived. Google’s Project Cannoli, a 1-gigawatt data center in Michigan, confirmed water consumption at 2 to 3.6 million gallons per day. The American Water Stewardship Act passed the House 378 to 32.

Now for the detail that will make a future subscriber very smug. JPMorgan published a 98-page report documenting data center water demand in precise detail. The report confirmed the physical constraint. It confirmed the demand trajectory. It confirmed that this is a real and growing infrastructure challenge. And it did not mention a single water infrastructure company as an investable equity. Not one. Ninety-eight pages. The institutional research community has confirmed the demand and has not connected it to the stocks that benefit from it. Phase 3 recognition moved from 30 to 35 percent in 48 hours. Equity market recognition of the investable angle is estimated at approximately 5 percent. The 30-point gap between those two numbers is the definition of the narrative-to-execution window this thesis was built to find.

THE THESIS SCORECARD

Holdings assessed against the infrastructure causal chain. Dollar values, position sizes, and allocation percentages excluded. “Fortress” names have strong balance sheets and existing revenue. “Financing-Dependent” names are earlier-stage and more sensitive to credit conditions.

GE Vernova (GEV) | Phase 1 Power / Cross-Phase EPC | Fortress | STRENGTHENING

Last week confirmed GEV as one of two or three validated global gas turbine suppliers. This week quantified what that oligopoly position actually means when demand doubles. Gas turbine orders in Q1 2026 ran between 12 and 24 gigawatts against 8 gigawatts in the same period last year. That is a 50 to 200 percent year-over-year demand surge. GEV is sold out through 2030 and requires non-refundable deposits to hold a manufacturing slot. The market sold the stock on AI efficiency fears Thursday while GEV’s customers were putting down non-refundable money to get in line for turbines that will not arrive for years. Steel above $1,000 per metric ton, confirmed this week, actually hardens GEV’s competitive position by raising the input cost barrier for any new entrant attempting to build competing manufacturing capacity at scale. The $5 billion Brookfield partnership received independent JPMorgan confirmation. Thursday’s stock decline was beta. Q1 earnings in late April are the verification event: can production ramp match the demand that now exists on paper?

Vertiv Holdings (VRT) | Phase 2 Cooling | Fortress | INTACT

JPMorgan’s institutional citation of 50 to 100 kilowatts per rack as the structural forcing function for liquid cooling is the cleanest Phase 2 confirmation this publication has seen from a sell-side research desk. Air cooling stops working at roughly 20 kilowatts. The racks are already deploying above that threshold. This does not negotiate with interest rates, oil prices, or Fed policy. Thursday’s 7.4 percent decline is beta. April 29 earnings, with a book-to-bill ratio above 1.5 times and delivery lead time commentary, are the Phase 2 classification gate.

Powell Industries (POWL) | Phase 1 Electrical Distribution | Fortress | INTACT

POWL manufactures the high-voltage switchgear and electrical distribution systems that connect power generation to data centers and industrial facilities. JPMorgan documented transformer producer prices at 90 percent above 2018 levels, the highest inflation rate of the 47 industrial goods categories it tracks through Bureau of Labor Statistics data. Forty-five percent of US power distribution infrastructure is near end of useful life. Backlog is durable. Thursday’s 8.1 percent decline was thin-volume amplification of the broad market move with no company-specific negative news of any kind. FERC’s April 30 interconnection rulemaking is an imminent binary catalyst for the entire electrical distribution category.

Quanta Services (PWR) | Cross-Phase EPC and Grid | Fortress | INTACT

JPMorgan documents 2,300 gigawatts of projects waiting in the US interconnection queue, the backlog of generation projects that need grid connection approval before construction can proceed. At current processing rates, that queue represents more than a decade of construction work. NextEra’s up-to-10-gigawatt Texas and Pennsylvania gas development and the DOE/SoftBank/AEP Ohio project are direct demand signals in PWR’s core territory.

The cost pressure is real and compounding. Diesel at $5.375 per gallon nationally, steel above $1,000 per metric ton, aluminum at approximately $5,000 per metric ton through Section 232 tariffs, and 28 percent of construction firms directly impacted by immigration enforcement actions. These are not abstract macro risks for PWR. They are direct inputs to active construction projects. Q1 earnings are the resolution event: do escalation clauses in existing contracts recover these costs, or does PWR absorb them as margin compression?

BWX Technologies (BWXT) | Phase 3b Nuclear | Fortress | INTACT — most underpriced Fortress name

The NRC Part 53 technology-neutral licensing pathway finalized this week directly supports BWXT’s manufacturing pipeline for advanced reactor components. Kentucky’s $1.76 billion nuclear enrichment investment is a direct demand signal. Five states confirmed nuclear moratorium repeals from hard sources this week. TVA’s BWRX-300 project received named citation alongside $800 million in DOE funding. BWXT’s defense and naval nuclear services business is insulated from near-term commercial SMR economics headwinds. The break condition requires DOE to eliminate naval and defense nuclear programs. That has not happened. Thursday’s 6.4 percent decline was nearly entirely beta-driven. This equity has not moved with the nuclear narrative acceleration that has been documented in this system for three consecutive runs.

Bloom Energy (BE) | Phase 1 Behind-the-Meter | Financing-Dependent | INTACT — three open flags

The White House “bring your own power” mandate published in the Federal Register is the most direct policy tailwind BE has received. Henry Hub natural gas at $3.05 per million BTU versus the European equivalent at $19 to $22 is a six-to-one ratio that makes the economics of domestic fuel cell generation unusually compelling against anything dependent on global LNG. The $5 billion Brookfield partnership received independent JPMorgan confirmation.

Three flags remain open and unresolved. A partial trim review and a separate position sizing review are both overdue for decision this cycle. The third flag is material before any sizing action: a JPMorgan review of BE’s 10-Q documents a 70 percent dependency on Chinese rare earth materials for fuel cell manufacturing. That dependency requires investigation before execution.

Ormat Technologies (ORA) | Phase 3b Geothermal | Financing-Dependent | RECLASSIFYING TO INTACT

Last week’s brief carried ORA as CHALLENGED. That was a misapplication of JPMorgan’s skepticism about next-generation enhanced geothermal systems to ORA’s conventional hydrothermal fleet, which operates under different economics and a different risk profile.

The evidence arrived Thursday in the clearest possible form. When oil spiked and every other holding fell, ORA was the only name in this system to finish the session green, up 1.35 percent. Conventional geothermal output becomes more valuable when energy prices rise because it is firm, weather-independent, and domestic. It does not care about Hormuz. The water-neutral angle, near-zero net water consumption, is becoming relevant as Phase 3 recognition inflects. ORA is rate-pressured and carries financing-dependent characteristics. INTACT, rate-pressured, is the correct classification. Formalized at the next quarterly review.

FuelCell Energy (FCEL) | Phase 1 Behind-the-Meter | Financing-Dependent | FRAGILE

This is the holding most directly exposed to the private credit deterioration that dominated the credit picture this week. Apollo’s 3/3 escalation condition confirmed narrows the financing pathways that a pre-revenue company depends on. The HY OAS threshold breach is a direct tightening of the environment under which FCEL raises capital. No positive company-specific signals arrived this week. Monday’s Federal Reserve HY OAS Master II data release is the gate event. If confirmed above 3.50 percent sustained, the escalation cascade runs within 24 hours with no exceptions.

Johnson Controls (JCI) | Phase 2 Cooling | Fortress | INTACT

No material week-specific news. Cooling density secular tailwind confirmed by JPMorgan’s institutional citation. Fortress balance sheet. Rate-sensitive but not thesis-challenged.

Primoris Services (PRIM) | Cross-Phase EPC | Financing-Dependent, profitable | INTACT — earnings gate

PRIM is the portfolio name most directly exposed to the compound cost pressure building in the construction sector right now. California diesel at $7 per gallon hits PRIM’s fleet-heavy operations directly. Steel above $1,000 per metric ton is a direct input cost. The AGC labor disruption finding applies to PRIM’s project execution in concrete terms. Lump-sum contracts, where PRIM bears cost risk if inputs rise above estimates, are the specific exposure mechanism. Q1 earnings are the resolution event: if escalation clauses in existing contracts are not comprehensive enough to recover these costs, the thesis moves to CHALLENGED.

Ecolab (ECL) | Phase 3 Water | Fortress | INTACT

The three Phase 3 signals this week directly strengthen the industrial water treatment thesis. ECL treats and manages water in industrial and commercial settings, including the closed-loop cooling systems data centers require. Rate-pressured; patience capital. The broader Phase 3 recognition move this week is the most important development for ECL since the position was initiated.

Brookfield Asset Management (BAM) | Cross-Phase Capital | Fortress | FUNCTIONAL — monitoring

The $5 billion Bloom Energy partnership received independent JPMorgan confirmation this week, a hard positive. The Apollo private credit deterioration is directly relevant to BAM as an operator in the same alternative asset management space. BAM is the most exposed Fortress name to private credit contagion risk. BDC earnings in April determine whether Apollo is isolated or systemic. Fortress classification is intact; BAM has the balance sheet to weather a difficult fundraising period. But this is the flag to watch among the Fortress names.

SPX Technologies (SPXC) | Cross-Phase EPC | Financing-Dependent | INTACT — commodity-pressured

Steel above $1,000 per metric ton and aluminum at approximately $5,000 per metric ton through Section 232 tariffs are direct input costs. Tariffs of 15 to 50 percent on electrical equipment imports are confirmed structural. Q1 earnings are the resolution event, same timeline as PRIM and PWR.

Xylem (XYL) | Phase 3 Water | Fortress | INTACT — Phase 3 asymmetric opportunity

Three independent hard signals this week document the water demand thesis. Regulatory pressure at the state level, Michigan’s attorney general demanding contested-case hearings on Google and Oracle water consumption, moratorium bills framing water as a shared public resource, is creating the policy environment in which mandated procurement of water efficiency technology becomes a real outcome. JPMorgan’s 98-page data center water demand report does not mention XYL. The research confirmed the demand and overlooked the equity. Rate repricing already reflected in the stock price appears disproportionate to the thesis trajectory. The recognition gap is widening, not closing.

Fluence Energy (FLNC) | Phase 3b Storage | Financing-Dependent | CHALLENGED — exit evaluation active

Three consecutive quarters of project execution failures. Margin compression from 12.5 percent to 4.9 percent in the most recent quarter. Management’s recurring “project-specific cost” language to describe what is now a structural execution pattern across multiple projects over multiple quarters. The Delay-versus-Damage framework is the right lens here. Delay means the timeline extends with the end-state intact. Damage means the thesis is actually breaking. Three consecutive quarters of the same execution failure is Damage. The case for exit is stronger than the case for hold. Full exit evaluation is active.

Oklo (OKLO) | Phase 3b Advanced Nuclear | Financing-Dependent | INTACT — patience capital

NRC Part 53 finalization directly supports Oklo’s Aurora compact fast reactor design. Kentucky’s $1.76 billion nuclear enrichment investment and five-state moratorium repeals represent the bipartisan political momentum the long-term thesis requires. A $1.1 billion cash position means Oklo is not dependent on the private credit markets that are tightening. One new flag this week: JPMorgan raised potential NRC regulatory capture risk from the Trump executive order on nuclear regulation. Decision Log entry required before any sizing action.

Centrus Energy (LEU) | Phase 3b Nuclear Fuel | Financing-Dependent | INTACT — timeline extended

Kentucky’s $1.76 billion nuclear enrichment investment is the most direct demand signal LEU has received in months, a named state-level commitment to the exact supply chain LEU occupies. The near-term headwind is that SMR commercial deployment timelines extending rightward push high-assay low-enriched uranium demand rightward with them. The thesis is intact; the timeline is extended. Patience capital through 2027 and beyond.

THE RISK REGISTER

Risk 1: KC4 Capital Withdrawal — Approaching Stage 2, Velocity Acute

Kill Chain 4 is the formal tracking mechanism for the scenario where credit tightening cuts off capital access for infrastructure companies, particularly the earlier-stage ones. Last week it was 22 basis points from its threshold. This week it crossed.

The inputs to Stage 2 are nearly complete. Apollo’s three-source escalation condition is met. The high-yield credit spread threshold is breached. The OECD’s 4.2 percent inflation projection eliminates the Fed rate cut scenario through 2026, removing the pressure relief valve that would otherwise ease credit conditions over the next 12 months. The Cliffwater BDC Index, which tracks the business development companies that provide financing to smaller infrastructure firms, is down 11.5 percent year-to-date and 20 percent over the trailing twelve months.

The one missing input is earnings data. Hard evidence from the EPC companies and infrastructure builders in this portfolio showing whether margin compression is appearing in reported financials arrives with Q1 earnings in April and May.

The critical structural distinction to hold onto: the Doom Loop scenario, where demand collapse and credit withdrawal reinforce each other into a full thesis break, is not activated. Demand is not falling. Hyperscaler capital expenditure commitments for 2026 remain at $750 billion with no language indicating reduction. The 56-gigawatt BTM pipeline exists because demand is real. Credit stress without demand collapse is painful and compressive for equity prices. It is survivable for companies with strong balance sheets and catastrophic for pre-revenue companies that need to raise capital right now. That distinction is the most important analytical call in this portfolio at this moment.

Risk 2: April 6 Binary — Three Scenarios, One Weekend

The Trump administration’s pause on strikes against Iranian energy infrastructure expires April 5 and 6. Iran rejected the ceasefire proposal and countered with a demand for sovereign Hormuz control. RUSI’s munitions assessment says the US has at most one month of key high-end inventory remaining. Three scenarios, and the evidence has already started eliminating one of them.

Scenario A is a ceasefire or deal: Brent crude falls $15 to $20, credit and macro pressure ease at the margin, relief rally. Scenario A is now structurally implausible given Tehran’s counter-proposal. The market can still run it temporarily, but the evidence does not support it.

Scenario B is extension or status quo: Brent holds $100 to $110, Kill Chain 4 continues advancing toward Stage 2, no new action required, pressure continues accumulating. This is the working assumption.

Scenario C is escalation, including potential Hormuz mining: Brent moves above $120, potentially toward the $130 to $200-plus range that Macquarie assigns a 40 percent probability to if the war extends through June. Bloomberg New Energy Finance puts a $154 per barrel tail scenario on Hormuz closing by mid-June. Scenario C triggers emergency system runs immediately with no exceptions.

Brent’s Sunday evening open is the primary read. Truth Social and Iranian state media are the pre-market monitoring sources.

Risk 3: Compound Fragility — First Occurrence in System History

Simultaneous breach of commodity input price thresholds and credit stress thresholds activated a condition this system has never previously logged. These stresses are not independent. Commodity inflation raises costs for EPC contractors, compressing margins. Credit tightening raises the cost of capital and narrows financing pathways. Both occurring simultaneously hits the companies caught at the intersection: profitable but commodity-exposed EPC firms facing margin compression while their customers face tighter financing. PRIM, PWR, and SPXC are the primary names at that intersection. Q1 earnings are the measurement event.

Risk 4: EPC Labor Disruption — Now Structural

The four primary supply chain constraint categories, large power transformers, gas turbines, DC electrical distribution modules, and EPC labor, all triggered structural persistence flags simultaneously for the first time in this system’s history. Each has been flagged before. Their simultaneous triggering changes the assessment from “multiple constraints” to “constrained execution across every dimension.”

The labor component deserves specific framing because its cause is a domestic policy choice, not a market cycle. The 1.2 million Temporary Protected Status rescissions translate to roughly 240,000 to 300,000 construction workers removed from legal employment eligibility. In North Texas, confirmed active project disruption was documented on March 25. Revenue delay risk for new projects involving electrical scopes is now quantified at five to nine months for new starts and three to six months for projects with existing crews. These are delays, not cancellations. The end-state is intact. But the administration simultaneously created the energy urgency and is removing the workers who would respond to it. The data says that. It says it clearly.

System transparency note: Several monitoring data series remain unverified for four consecutive weeks, including uranium ETF relative performance versus the S&P 500. The W3 Commodity Monitor is overdue for an off-cycle run; California diesel at $7 per gallon and a new hoarding dynamic in fertilizer and agricultural chemical inputs are not yet formally captured in the system. BDC earnings in April are the resolution event for whether private credit stress is isolated to Apollo or systemic.

NEXT WEEK’S SETUP

Key Catalysts:

• Saturday-Sunday, April 5-6 — Trump pause expires / Iran binary resolves. Brent crude’s Sunday evening open is the primary signal. Pre-decide the response to all three scenarios before markets open. Scenario C, Brent above $120, triggers emergency system runs immediately with no exceptions. Truth Social and Iranian state media are the pre-market monitoring sources.

• Monday, April 7 — Federal Reserve releases HY OAS Master II, the definitive high-yield credit spread series. If confirmed above 3.50 percent sustained, cascading system runs trigger within 24 hours with no exceptions. This is the single most important scheduled data point of next week independent of how April 6 resolves.

• Week of April 7 — Business development company earnings season begins. Apollo distributed 45 cents on the dollar to investors this week. The question: isolated incident at one firm, or the first visible crack in a broader private credit system under stress? BDC earnings answer that.

• Imminent — Maine LD 307 House vote. If enacted, Maine becomes the first confirmed instance of a moratorium causing a named project to stop or redirect. The first moratorium-induced capital impairment trigger in this system. Off-cycle political and permitting runs trigger immediately upon enactment.

• April 29 — Vertiv Q1 earnings. Book-to-bill above 1.5 times and delivery lead time commentary are the Phase 2 cooling classification gate signals.

• Late April — GE Vernova Q1 earnings. Production capacity against the $40 billion Japan commitment, the 12 to 24-gigawatt order range, and gas turbine constraint classification are all in play.

• April-May — Primoris, Quanta, and SPX Q1 earnings. Gross margin trajectory and escalation clause coverage are the resolution events for the commodity-versus-backlog conflict that has been open for three consecutive runs.

• April 30 — FERC RM26-4-000 large-load interconnection rulemaking deadline. This rulemaking governs how major new power consumers connect to the grid and on what timeline. Thirty-four days remain. FERC can act between monthly meetings at any time. A joint off-cycle run of supply chain and political monitoring systems is planned immediately upon any FERC action.

• Early May — NRC Part 53 takes effect. If multiple advanced reactor developers file applications within 60 days of effectiveness, it confirms the rule is operationally usable, not just technically available. That filing cluster is the upgrade signal.

• May 15 — Federal Reserve Chair Jerome Powell exits. His replacement has not been confirmed. The monetary policy uncertainty window extends through at minimum the confirmation process. Powell’s final FOMC meeting is April 29 to 30, which is also the FERC deadline date. That is a lot of consequential events on the same calendar block.

Risks to Watch:

Hormuz. Every day under Scenario B is a day of deferred risk, not resolved risk. Tehran’s toll collection legislation is a persistent signal about strategic intent independent of current military posture. Monitor Iranian state media and the Sunday Brent open.

Private credit contagion. Apollo is confirmed at three sources. The question is whether it is the first domino or an isolated event. BDC earnings season answers that in April.

Maine moratorium enactment. The first confirmed moratorium-caused capital impairment is a different category of political signal than the bills that have been introduced, stalled, and tied up in courts to date.

The One Thing:

Sunday evening’s Brent crude open is not just a commodity price signal. It is the first read on which of three fundamentally different operating environments this portfolio is entering for the next quarter. The thesis infrastructure, confirmed demand, regulatory tailwinds, capital commitments across all four phases, is intact. What Sunday determines is whether the macro overlay that has been compressing everything for three weeks intensifies, holds, or eases. Pre-decide the responses to all three scenarios before the open. These are not decisions to make in real time when prices are moving.

THE BOTTOM LINE

1. The OECD placed US inflation at 4.2 percent in 2026 — 150 basis points above the Federal Reserve’s own projection, signed by 38 countries. Hard evidence — OECD Interim Economic Outlook, March 26, 2026 — Escalated from “confirmed” to “institutional consensus.” The Fed cannot cut into 4.2 percent inflation and cannot hike into 0.7 percent GDP growth. Higher-for-longer is now the baseline, not the risk scenario. Every investment decision in this portfolio for the next 12 to 18 months is made against that backdrop, and the rest of the developed world’s economists are on record saying so.

2. Behind-the-meter generation is now federal law, and the demand behind it reached 56 gigawatts of confirmed pipeline plus 34 additional gigawatts of named projects announced in a single week. Hard evidence — White House AI Framework, Federal Register, W6 Newsletter Sweep — Escalated to institutional consensus. GE Vernova is sold out through 2030. Non-refundable deposits are required to hold a manufacturing slot. The market sold the stock on AI efficiency fears while GEV’s customers were literally putting down non-refundable money to get in line.

3. The water infrastructure recognition gap is the most significant asymmetric opportunity in this thesis since it launched. Hard evidence — American Water CERAWeek presentation, Google Project Cannoli water disclosure, American Water Stewardship Act — New, triple simultaneous signal, first in system history. Phase 3 recognition moved from 30 to 35 percent in 48 hours. Equity market recognition of the investable angle is approximately 5 percent. JPMorgan wrote 98 pages on data center water demand and did not mention a single water infrastructure equity. The gap between what the research confirms and what the market has priced is the entire opportunity.

4. The credit escalation threshold is confirmed crossed, the private credit condition is met at three of three hard sources, and Compound Fragility has been triggered for the first time in this system’s operating history. Hard evidence — Federal Reserve HY OAS data, Apollo fund disclosures, Cliffwater BDC Index — Escalated. The Doom Loop, which requires simultaneous demand collapse and credit withdrawal, is not activated. Demand is intact at $750 billion in committed hyperscaler capital expenditure. But the financing environment for pre-revenue infrastructure companies is actively hostile, and Monday’s Federal Reserve data release is the gate event.

5. The April 6 binary resolves this weekend, and Iran’s counter-proposal has made a ceasefire structurally implausible. Hard evidence — Iran counter-proposal documentation, RUSI weapons depletion assessment, Brent $108 — Escalated, critical. The working assumption is Scenario B: extension and status quo, Brent holding $100 to $110. Scenario C carries a Macquarie-estimated 40 percent probability if the war extends through June and triggers emergency system runs immediately. Pre-decide the responses to all three scenarios before Sunday evening’s open. That decision made in advance is worth more than any analysis produced after the fact.

SOURCES

Regulatory and Government:

• NRC Part 53 final rule — technology-neutral licensing pathway; effective approximately early May 2026

• FERC RM26-4-000 — large-load interconnection rulemaking; April 30, 2026 deadline

• FERC 202(c) emergency orders — MISO, March 19, 2026 (prior week; structural reference)

• White House AI Framework — Federal Register; Ratepayer Protection Pledge

• American Water Stewardship Act — passed House 378-32, March 24, 2026

• Sanders-AOC AI Data Center Moratorium Act — introduced March 25, 2026

• DOJ AI Litigation Task Force — federal preemption ruling on data center zoning

• Michigan MPSC — Google/DTE and Oracle/Stargate contested case hearings

• Maine LD 307 — House vote imminent

• South Dakota SB 135 — enacted

• Virginia HB1515 — introduced; structural persistence flag

Institutional Research:

• OECD Interim Economic Outlook, March 26, 2026

• JPMorgan 98-page energy infrastructure report

• RUSI — US munitions depletion assessment

• Associated General Contractors of America survey — 28% of firms impacted; March 2026

• Goldman Sachs — 35% recession probability

• Macquarie — 40% probability $200 oil if war extends through June; next Fed move hike 1H27

• Bloomberg New Energy Finance — $154/bbl Hormuz closure tail scenario

• Bank of America — $175B hyperscaler financing forecast 2026

Market Data:

• Federal Reserve / FRED — HY OAS Single-B sub-index 3.50%; BB sub-index 1.96%; as of March 24, 2026

• EIA — diesel $5.375/gal, week of March 23

• LME — nickel $17,253/T, March 26-27

• Cliffwater BDC Index — -11.5% YTD, -20% TTM

• Nomura — HY fund outflows, 10+ consecutive weeks

• PGIM — HY primary market on hold, week of March 20

• Mountain West News Bureau — 54 total local data center moratoriums confirmed, March 27, 2026

Company Disclosures and Project Announcements:

• Amazon $54B bond deal, March 10-11 — 14x oversubscribed

• Apollo Global Management — $15B private credit fund redemptions at 2.2x quarterly cap; 45 cents on dollar distribution

• Google Project Cannoli (Van Buren Township, MI) — 2-3.6M gallons/day; 1 GW data center

• American Water CERAWeek presentation, March 24-26

• Brookfield Asset Management / Bloom Energy $5B partnership — JPMorgan independent confirmation

• Kentucky $1.76B nuclear enrichment investment

• Fermi America 2+ GW off-grid gas, Texas

• Crusoe Energy 5 GW gas turbines

• Arbor Energy 5 GW baseload

• DOE / SoftBank / AEP Ohio — 9.2 GW gas + 10 GW data center

• NextEra up to 10 GW gas, Texas and Pennsylvania

• Nscale 2 GW campus, West Virginia (Microsoft/Nvidia-anchored)

• Form Energy / Crusoe 12 GWh; Form Energy / Xcel / Google 30 GWh

• TVA BWRX-300 + $800M DOE funding

Commentary:

• ECB President Christine Lagarde — Iran as “real shock”

• QatarEnergy CEO — Ras Laffan 3-to-5-year repair timeline (prior week; structural reference)

Disclosure: The author may hold positions in companies discussed in this publication. Freedom Grid provides infrastructure analysis, not investment recommendations. Readers should conduct their own research before making investment decisions.

Freedom Grid Research System — Weekly Brief v1.1 | Week of March 27, 2026

Share

Subscribe now

Most people who write about markets spend their time building the case for what they already believe. I do that too. But I also spend an equal amount of time trying to destroy it, because there is real money behind these positions and motivated reasoning is how you lose it.

The thesis I kept trying to break is this: AI’s explosive demand for power, cooling, and physical infrastructure is structural and durable, not a temporary buildout that efficiency gains will eventually deflate. The skeptic’s argument, which I take seriously, is that cheaper and more efficient AI chips will reduce the energy and cooling required per unit of useful work, eventually bending the demand curve down.

That argument got loud fast. When NVIDIA announced that its latest GPU architecture delivered 25 times lower energy per inference than its predecessor, a cooling infrastructure company I hold got hit hard. The market’s read was straightforward: more efficient chips mean less heat, less heat means less cooling infrastructure, less cooling infrastructure means the growth story is capped. Sell.

I went back to the data to find out if that read was right. It isn’t. Here’s why.

We Have Seen This Movie

Before getting into the current numbers, it’s worth spending two minutes on history, because this argument is not new. It is, in fact, one of the oldest mistakes in the technology industry.

In 1943, IBM’s chairman reportedly assessed global commercial demand for computers at around five units. Logical, given what computers did in 1943. In 1977, Ken Olsen, founder of Digital Equipment Corporation and one of the most respected technologists alive, said there was no reason anyone would want a computer in their home. Also logical, given what computers did in 1977. Then the internet arrived, and the assumption was that bandwidth demand would cap out once everyone had email. Then came images, video, streaming, 4K, and a world where a single household routinely pulls more data in an evening than entire corporate networks did in a week in 1995.

The pattern, every single time: costs fall, efficiency improves, serious people conclude demand will moderate, and then a wave of use cases arrives that the previous cost structure had made impossible. Cheaper compute doesn’t satisfy demand. It manufactures new demand that didn’t exist before.

Economists have a name for this. William Stanley Jevons identified it in 1865, watching British coal consumption rise sharply after steam engine efficiency improved. Cheaper energy per unit of work means more work gets done. More work means more total energy consumed. The efficiency gain and the demand increase are not in opposition. The efficiency gain causes the demand increase.

That’s the lens. Now the data.

What My Research Actually Showed

When I ran the numbers on AI inference economics, the Jevons dynamic is not subtle. It’s running in plain sight.

GPT-4 equivalent performance cost approximately $20 per million tokens in late 2022. Today it runs at roughly $0.40 per million tokens. That’s a 98% cost reduction in about three years, faster than the PC price curve, faster than the bandwidth curve of the early internet. By every efficiency metric, AI inference is getting dramatically cheaper.

OpenAI’s total cost to serve that cheaper inference was $8.4 billion in 2025. It’s projected to reach $14.1 billion in 2026. The company spent approximately $1.35 for every dollar it earned. Cheaper per query. Nearly double the total bill year over year. That’s not a contradiction. That’s Jevons, live.

The hyperscalers are telling the same story in their capital allocation. The five largest U.S. technology companies plan to spend $660 to $690 billion on infrastructure in 2026, up roughly 36% from 2025 and nearly triple 2023 levels. Every one of them describes their operating environment not as demand-constrained but as supply-constrained. Microsoft, as of early 2026, could not fulfill $80 billion in Azure orders. The bottleneck was not willingness to spend. It was power. Electricity. The physical inability to get enough energy to facilities to run compute that customers had already paid for.

When the most valuable technology companies on earth are sitting on $80 billion in unfulfilled orders because they can’t get enough power, the efficiency-reduces-demand argument has some explaining to do.

The Per-Rack Number Nobody Mentions

Back to the cooling position and the NVIDIA announcement that punished it.

NVIDIA’s claim that the GB200 delivers 25 times lower energy per inference than the H100 is accurate in the narrow sense the marketing department intends. On an equivalent task, the newer chip does the work more efficiently. That part is real.

Here is what the press release did not emphasize. The H100 draws 700 watts at thermal design power. The B200 draws 1,000 watts. The GB200 draws 1,200 watts. NVIDIA’s Vera Rubin architecture, targeted for the second half of 2026, is projected at approximately 2,300 watts per chip. The GB200 NVL72 rack system, the standard deployment configuration for frontier AI workloads, draws between 120 and 140 kilowatts per rack. For context, a standard data center rack from three years ago drew around 20 kilowatts.

There is no air-cooled solution at 120 to 140 kilowatts per rack. Liquid cooling is not a premium option at this density. It is mandatory infrastructure. The physics don’t care about the per-query efficiency ratio.

What’s happening is not complicated once you see it. Each chip generation is more efficient per unit of useful work. Each chip generation also enables workloads that the prior generation couldn’t run, which means denser deployments, larger models, more users, and more total heat generated per facility. The per-query efficiency improvement is real. The per-rack power draw keeps rising. These two things are both true simultaneously, and conflating them is how you end up selling a cooling infrastructure company because NVIDIA said its chips are more efficient.

U.S. data center power demand sat at roughly 61.8 gigawatts in 2025. The 451 Research projection for 2030 is 134 gigawatts. That’s more than doubling in five years, in a country that takes seven to twelve years to permit and build significant new transmission infrastructure.

The Use Case Wave Hasn’t Arrived Yet

The efficiency argument misses one more thing. The current buildout is being sized for query-response AI. You type, a model thinks, you get an answer. That’s the use case driving $660 billion in hyperscaler capex in 2026.

The next wave is agentic AI: systems that run continuously, take autonomous action, and operate around the clock without a human in the loop. Gartner put AI agent integration at below 5% of enterprise applications at the start of 2025. Their year-end 2026 projection is 40%. The compute signature of an agentic workflow is not comparable to a chat query. A single agent completing a complex task can involve hundreds of model calls and reasoning chains running in parallel, continuously, all day.

The infrastructure being built today isn’t for the AI that exists. It’s for the AI that cheap inference is about to make economically viable at enterprise scale. Watson couldn’t see the spreadsheet. Olsen couldn’t see the internet. The people selling cooling infrastructure because NVIDIA’s chips are more efficient can’t see the always-on agent running eight hours a day on every enterprise desktop.

What Would Actually Break This

I take the invalidation question seriously, so here it is directly.

The efficiency argument wins if a reasoning breakthrough of a genuinely different order arrives: not architectural improvements, but something that achieves high-value AI output at a fraction of current inference compute, making much of the buildout unnecessary. That scenario exists and I track it. Against the current weight of evidence, including a chip TDP roadmap headed toward 2,300 watts and hyperscalers with $80 billion in unfulfilled orders, its probability is around 10%.

The more tractable risk is timing. Agentic AI adoption could be slower than Gartner’s 40% implies. Hyperscaler capex could plateau near 2026 peak levels rather than continuing to accelerate. If that happens, the buildout cycle extends rather than compresses. That’s a delay, not a thesis break.

The signal I’m watching: any shift in hyperscaler earnings language from “supply-constrained” to “optimizing deployment.” That language change, not any efficiency announcement, is what would tell me the thesis is bending.

Three months of trying to kill this. The thesis is intact. The cooling position stays.

Disclosure: The author may hold positions in companies discussed in this article. This publication does not provide investment advice.

Share

Subscribe now

What the Infrastructure Build Actually Looks Like From Here

Freedom Grid | Q1 2026 Quarterly Brief

A note before we get into it. I run a formal quarterly analysis on this portfolio. The distilled version of that process is 22 pages. The underlying work, every data source, every scenario model, every holding-level breakdown, runs 50 to 60 pages. None of that makes for good Substack reading. What follows is my attempt to give you the substance without making your eyes glaze over. If you want the full version, reach out directly and I’ll share it. I don’t charge for it.

I’ve been building this portfolio and this platform for about a year. I own everything discussed here. Nothing in this piece is investment advice. Several of these companies aren’t cheap right now, and you should do your own work before acting on any of it.

The Idea, Plainly Stated

AI doesn’t run on software. It runs on power plants, cooling systems, water infrastructure, and nuclear fuel. The companies solving those physical problems are where I’ve put my money, and this is my quarterly accounting of whether that bet still makes sense.

Start with the chip, because that’s where the physical reality begins.

NVIDIA’s H100, the chip that powered most of 2023’s AI buildout, draws 700 watts. The current generation B200 draws 1,000 watts. The GB200 draws 1,200 watts per chip, and a full rack of GB200s requires 120 to 140 kilowatts of total power. The next generation, Vera Rubin, is estimated around 2,300 watts per chip. Each generation consumes more power than the last, and demand for each generation is higher than the one before it. The hardware is not becoming more efficient at the system level. It is becoming dramatically more power-intensive, and the market keeps underestimating what that means for physical infrastructure.

Microsoft has $80 billion in Azure orders it cannot fulfill. Not because customers changed their minds. Because the power to run the servers doesn’t exist yet.

That gap between what AI needs physically and what the current infrastructure can deliver is what this portfolio is built around. The causal chain works like this. Power generation comes first. You need electricity before anything else runs. Generating that much power creates that much heat, which requires cooling infrastructure built for densities that didn’t exist three years ago. Cooling at data center scale consumes significant water, more than most people realize and from sources that often can’t provide it. Always-on AI workloads need firm baseload power that solar and wind can’t reliably provide on their own, which is the nuclear argument. And threading through every phase of this are the engineering and construction firms that actually put steel in the ground, whose labor capacity determines how fast any of this happens regardless of how much capital is theoretically available.

Every link in that chain has a constraint. Every constraint has a timeline measured in years to decades. That’s not a problem with the thesis. That’s the point of it.

The Macro Environment

This was a complicated quarter to be running an infrastructure thesis, and I want to be honest about why before I tell you why I’m still running it.

The Iran/Hormuz crisis became the defining macro event of Q1. Iranian forces attacked more than 21 merchant vessels in the strait. The US struck Kharg Island on March 20, Iran’s primary oil export terminal, in the most direct military escalation of the conflict. The Strait of Hormuz effectively closed to most commercial shipping. Brent crude hit $112.19 per barrel on March 20, the highest level of the war. Goldman Sachs projects elevated oil prices could persist through 2027.

What followed was a commodity shock across the board. Diesel is up 27%. Copper is up 36%. Aluminum is up 52%. Grain-oriented electrical steel, the primary input for large power transformers, is now exposed to both tariff pressure and the Iran-driven metals spike simultaneously. These aren’t brief spikes. They’ve been sustained for over three weeks across energy and metals at the same time. That matters for EPC contractors executing fixed-price contracts and for any company that manufactures physical equipment.

The Federal Reserve held rates at 3.50 to 3.75% on March 18. The dot plot showed one cut projected for 2026, in December, with seven of nineteen members forecasting zero cuts. Core PCE was revised up to 2.7%. Long-run neutral was revised up to 3.1%. GDP growth for 2026 was cut to 0.7%. The Fed is caught: inflation won’t let them cut, growth won’t let them hold comfortably, and a geopolitical commodity shock is doing what commodity shocks do. Powell exits on May 15, and the Warsh confirmation is delayed, which adds leadership uncertainty at the worst possible time.

Credit markets are bifurcating in a way worth understanding because it explains a lot about how I’m thinking about specific holdings.

High-yield spreads have moved 20 basis points in two weeks to 3.28 to 3.35%. The level I watch as an escalation threshold for the credit stress picture is 3.50%. We’re 22 basis points away. The high-yield default rate sits at 5.28%. The Senior Loan Officer Opinion Survey, which tracks lending standards across the banking system, shows tightening conditions for all borrowers, with meaningful credit demand concentrated only among large firms. Private credit is facing a potential $1.8 trillion redemption wave according to two independent sources, not yet fully confirmed but being watched.

On the other side of that bifurcation: Amazon just completed a $54 billion investment-grade bond deal that was 14 times oversubscribed. Brookfield raised $112 billion in 2025. Investment-grade credit is fully open for self-funding, creditworthy names. The market is not broken. It’s sorting.

That sorting matters enormously for this portfolio because the holdings aren’t uniform. Some of them are self-funding businesses with billions in net cash and 20-plus percent margins. Others are pre-revenue or financing-dependent, relying on capital markets and project finance that just got more expensive. The macro environment isn’t good or bad for this thesis universally. It’s very good for one category of holding and genuinely pressured for another. I’ll explain which is which throughout.

The position I’d characterize the cycle at right now: mid-to-late transition. Not a recession, not a blowup, but the part of the cycle where discipline separates the holdings that compound from the ones that struggle.

The Demand Picture

Despite all of that macro complexity, the demand case for AI infrastructure this quarter received the strongest confirmation I’ve seen since I started running this analysis. Four completely independent analytical processes, using different data sources and different methodologies, arrived at the same conclusion. Expansion dominant. Getting more so.

The five major hyperscalers collectively guided $660 to $750 billion in capital expenditure for 2026. To put that in context: one major research firm revised its estimate up from $602 billion in November to $750 billion in February, after individual company earnings came in higher than their models expected. The individual guidance: Alphabet $175 to $185 billion, Amazon $200 billion, Meta $115 to $135 billion, Microsoft $120 billion-plus, Oracle $50 billion. Goldman Sachs puts the three-year total through 2027 at $1.15 trillion. These are not projections. They are guided, committed capital that companies have disclosed in SEC filings and earnings calls.

Every single one of these companies used supply-constrained language. Not demand-constrained. Supply-constrained. They are not pulling back from AI investment. They are trying to build faster than the physical infrastructure allows. The Microsoft Azure $80 billion unfulfilled backlog is the single most useful data point in the entire demand picture. The demand exists. The power doesn’t.

The efficiency narrative has been tested and it failed.

Inference costs per unit dropped roughly 78% through 2025, faster than most people expected. The conventional wisdom was that cheaper inference would reduce infrastructure demand. Here’s what actually happened: OpenAI’s compute bill went from $8.4 billion in 2025 to a projected $14.1 billion in 2026. Cheaper inference generated more inference. More inference generated more infrastructure demand. This is the Jevons Paradox, which economists have documented across every energy technology since coal in the 1860s: efficiency improvements in resource use tend to increase total consumption rather than decrease it, because lower cost per unit drives higher volume. DeepSeek in January was supposed to be the moment this thesis broke. The numbers say otherwise.

There’s a piece of the demand picture that I don’t think the market has absorbed yet. Only 11 to 14% of enterprises have agentic AI solutions in production today, according to Deloitte’s December 2025 survey. Gartner projects 40% by year-end 2026. The infrastructure being ordered now, with 18 to 24 month transformer lead times and multi-year equipment backlogs, is being sized for 2027 and 2028 demand. The agentic adoption wave that’s coming in the next 18 months hasn’t been built for yet. The ramp is in front of us, not behind.

Data center construction starts actually declined in Q4 2025, which some analysts read as a demand signal. It isn’t. The constraint is supply-side: power permitting delays, interconnection queue backlogs, and labor. Sightline Climate reports 30 to 50% of the 2026 data center pipeline is delayed for exactly those reasons. Demand isn’t falling. The physical capacity to serve it isn’t keeping up. That’s a different story.

My scenario assessment for capex durability: 55% probability that expansion continues at current trajectory, 35% probability that supply and demand reach a more balanced state over the next year or two, 10% probability that efficiency gains actually compress demand growth. The 10% scenario is the DeepSeek-changes-everything thesis. I am not seeing data that moves that probability.

Where the Market Is Getting It Wrong

Three specific places where the prevailing narrative doesn’t match what the evidence shows.

On chip efficiency and infrastructure demand. The market broadly believes that AI efficiency gains will eventually reduce the need for physical infrastructure. The GB200 generates 25 times lower energy per inference than the H100, according to NVIDIA. That sounds like infrastructure demand should fall. What actually happens: lower cost per inference drives dramatically higher inference volume. OpenAI’s compute bill isn’t falling. It’s rising nearly 70% year over year despite the efficiency gains. The per-unit improvement is being more than absorbed by the volume increase. Every analyst who shorted or avoided infrastructure names on the efficiency thesis this year has been looking at the wrong metric.

On Energy Recovery and the difference between delay and damage. This is the most interesting market mispricing in the portfolio right now, and I want to explain the logic carefully.

Energy Recovery is down roughly 42% from its highs. Two consecutive years of declining revenue. Guidance cut. The market’s interpretation is structural demand loss for desalination infrastructure. I think that’s wrong, and five completely independent analytical processes I run on this portfolio all arrived at the same conclusion.

Here’s the distinction that matters. A damaged business is one where the underlying demand has structurally shifted away from what the company provides. When that happens, margins compress as volume falls, because fixed costs don’t shrink as fast as revenue. A delayed business is one where the timing of specific projects has moved, but the underlying demand is intact. When that’s the case, margins don’t compress, because the company isn’t cutting prices to chase declining demand.

Energy Recovery’s gross margins are 65 to 67% and intact through a trough spanning two consecutive declining revenue years. Free cash flow is positive. The balance sheet has almost no debt. Three specific large desalination projects slipped their final investment decisions into 2027, creating back-to-back air pockets in the revenue line. But the projects exist. The EV/EBITDA is roughly 5.7 times on trough earnings. Amundi, a major institutional manager, grew their position by 323.6% in Q4 2025. A new product, the PX Q650, is confirmed for launch in the second half of 2026.

The market is pricing a broken business. The evidence describes a business waiting on project timing. Those two interpretations have very different outcomes by 2027.

On Phase 3 water and where the constraint actually lives. The mainstream narrative focuses on how much water a data center facility itself uses. Cooling towers, evaporation, on-site consumption. There’s been a fair amount of coverage of dry cooling innovations that reduce facility-level water use. That narrative misses the larger part of the problem.

Bluefield Research estimates that 72% of data center water consumption by 2030 will occur at the power plants generating the electricity, not at the data centers themselves. Generating 134 gigawatts of power for data centers requires cooling those generators, and thermal power plants are significant water consumers. The dry cooling solutions at facility level address maybe 28% of the actual water footprint. The remaining 72% is mostly invisible to anyone analyzing this at the facility layer. UC Riverside projects that US data centers will require between 697 million and 1.45 billion additional gallons of peak daily water capacity by 2030. Municipal water systems and water rights frameworks don’t have that capacity today and take 10 to 15 years to develop.

Phase 3 recognition in the market is somewhere around 30 to 40%. Physical deployment is 10 to 15%. The gap between those numbers is the opportunity, and it’s largely invisible because everyone is looking at the facility, not the generator.

The Infrastructure Stack

Before getting to specific holdings, here’s where each phase of the buildout actually stands right now.

Phase 1: Power. The most immediate constraint. Recognition is roughly 75 to 80%, meaning most sophisticated investors understand the power problem exists. Physical implementation is 25 to 30%. That 50-point gap represents years of construction, permitting, and capital deployment still ahead.

The transmission bottleneck is the single hardest physical constraint. New transmission capacity takes 7 to 12 years to develop. That timeline is not moving. The interconnection queue is growing faster than projects are completing. The practical response, which is now federal policy via the Ratepayer Protection Pledge, is behind-the-meter generation: data centers bringing their own power rather than relying on the grid. This isn’t a strategic preference anymore. Within 36 months, it becomes a practical necessity for any large new facility.

The transformer situation is worth naming explicitly because it doesn’t have a neat investable solution. Large power transformer lead times are structurally elevated at well over two years. The Iran/Hormuz shock made it worse: grain-oriented electrical steel is now exposed to both tariff pressure and the commodity spike simultaneously. There is no US-listed pure-play transformer manufacturer. The bottleneck is real, it’s worsening, and I can’t directly own it. It shows up indirectly in EPC contractor backlogs and in the urgency of the behind-the-meter thesis.

GE Vernova’s slot reservations grew from 62 to 83 gigawatts in 2025, with 24 gigawatts of new contracts in Q4 alone. Lead times are 3 to 5 years. That backlog data tells you more about Phase 1 tightness than any projection.

Phase 2: Cooling. Physics made this decision. Air cooling stops working above roughly 20 to 30 kilowatts per rack. The GB200 NVL72 rack runs at 120 to 140 kilowatts. There is no air-cooled alternative at that density that works at commercial scale. Every large AI data center being built today is being built for liquid cooling, and every future generation of chips requires more of it. The liquid cooling infrastructure market was $2.8 billion in 2025. Independent projections put it above $21 billion by 2032, a compound annual growth rate above 30%. Equipment lead times already exceed 18 months.

Recognition is 60 to 70%. Implementation is 35 to 40%. The gap is closing faster than Phase 1 because the physical buildout is simpler and faster than generation or transmission.

Phase 3: Water. Most underweighted phase in the mainstream narrative. Recognition is 30 to 40%. Implementation is 10 to 15%. As discussed above, the generation-layer water consumption is the story most people are missing. The constraints have 10 to 15 year resolution timelines and are just beginning to show up in project planning for new facilities.

Phase 3b: Firm Clean Baseload. This covers both nuclear and geothermal, two different technologies solving the same problem: firm, clean, dispatchable power that runs continuously independent of weather. The demand case for firm baseload is being extended by the EXPANSION DOMINANT signal. The more AI compute gets deployed, the longer the firm baseload need persists, and the stronger the argument for technologies that can actually serve it.

Nuclear SMRs are 15 to 20 years from meaningful commercial scale. Recognition is 50 to 60%. Implementation is 5 to 10%. First commercial SMR-powered data centers are not before 2030 and that’s the optimistic case. The policy environment improved materially this year with nuclear production tax credits surviving through 2033 and DOE HALEU funding active.

Geothermal is more near-term. Operating assets exist today, generating clean firm power with no fuel cost. The constraint is geology: you can only build where the resource exists. Recognition is lower than nuclear despite better near-term delivery capability.

EPC and Grid. The throughput constraint for everything. Three independent analyses I run all confirm the same thing: the multi-decade EPC cycle is extending, not compressing. Current backlogs reflect early-innings accumulation, not late-cycle saturation. Labor is the binding constraint across every phase. Four to eight month delays from labor constraints are already showing up on new project scopes. Self-perform labor models are a genuine competitive moat in this environment.

The Holdings, Phase by Phase

Phase 1: The Power Problem

GE Vernova (GEV) is my largest and most straightforward position. Gas turbines. Firm, dispatchable power that runs 24 hours a day. GEV’s total backlog is $150.2 billion, up $31.2 billion year over year. Slot reservations grew from 62 to 83 gigawatts during 2025. Free cash flow was $3.7 billion last year, up 118%. Net cash on the balance sheet is $8.8 billion. Book-to-bill is roughly 2 times.

You cannot quickly replicate this. The manufacturing capacity, engineering expertise, and installed base of service relationships took decades to build. In the current macro environment, a self-funding business with near-monopoly characteristics in critical infrastructure gets stronger as financing-dependent competitors face higher capital costs. GEV explicitly benefits from the tightening environment. I consider it the most insulated holding in the portfolio.

Bloom Energy (BE) is my behind-the-meter play. Fuel cell systems that generate electricity on-site, bypassing the grid connection problem for customers who can’t wait for transmission capacity. Product backlog grew 140% year over year to $6 billion. Six major hyperscalers are now in that backlog, versus one a year ago. Service backlog sits at $14 billion. Revenue guidance for 2026 implies 54 to 63% growth. Free cash flow positive for the second consecutive year. A $5 billion strategic supply commitment from Brookfield is the largest single customer relationship in the backlog.

Here’s my honest read. The demand thesis for Bloom has never been stronger. Behind-the-meter is transitioning from a strategic preference to a practical necessity, and Bloom is the most capable provider in the category. The complication is the balance sheet and valuation. The stock trades at roughly 94 times operating income on an adjusted basis and carries $2.5 billion in convertible debt. This is a holding where the thesis can be exactly right and the stock can still hurt you if execution wobbles in a tightening credit environment. I own it. I watch it more carefully than anything else.

Phase 2: The Cooling Problem

Vertiv (VRT) is the dominant liquid cooling infrastructure provider. Backlog of $15 billion, up 109% year over year. Book-to-bill of 2.9 times, meaning nearly three dollars of new orders for every dollar of recognized revenue. Adjusted operating margins at 23.2%. Free cash flow approximately $1.9 billion. One monitoring note: Vertiv discontinued quarterly order and backlog disclosure starting this year, shifting to annual 10-K as the primary verification. That removes a useful quarterly signal. Management tone and segment commentary at quarterly calls become more important signals until the next annual filing. This is a monitoring change, not a thesis change.

Johnson Controls (JCI) moved faster toward data center cooling than I originally modeled. Data center orders up 40% year over year in the most recent quarter. NVIDIA thermal management collaboration confirmed. Record backlog of $18 billion. Adjusted EPS guidance raised to approximately $4.70 for 2026, up roughly 25%. Started as a more peripheral position in my Phase 2 thinking. Becoming more central as the data center cooling business scales.

SPX Technologies (SPXC) quantified its data center revenue for the first time: approximately $200 million in FY2025, roughly 9% of total revenue. HVAC backlog up 22%, industrial and commercial backlog up 43%. SPX committed $100 million in capital to dedicated HVAC manufacturing. Originally in the portfolio as an indirect beneficiary. Now directly quantified and growing.

Phase 3: The Water Problem

Xylem (XYL) is my primary water infrastructure holding. Rough quarter on the surface: guidance below consensus, stock dropped 12% on earnings. I read it as a mix story, not a thesis problem. Xylem is intentionally walking away from lower-margin international business as part of a simplification program that included cutting China headcount by over 40% and divesting roughly $250 million in international metrology revenue. The segment I bought this for, Global High-Tech serving semiconductor and data center water applications, is growing double-digits. The overall guidance miss was noise relative to the vector I care about.

Energy Recovery (ERII). Already walked through the delay versus damage argument in the narrative section above. Adding the specifics here. Revenue was $135 million in FY2025, down 7% year over year. 2026 desalination guidance is $105 to $125 million. Three megaprojects slipped to 2027. Gross margins intact at 65 to 67%. Free cash flow positive in the trough. Near-zero debt. EV/EBITDA roughly 5.7 times trough. Amundi grew holdings 323.6% in Q4 2025. PX Q650 product launch confirmed for H2 2026. The 2027 project FIDs are the confirmation event for the delay thesis. If those projects proceed, this was a timing problem. If they don’t, it becomes a different conversation.

Ecolab (ECL) is the steadiest holding in the Phase 3 stack. Water treatment chemistry and services, recurring revenue, Global High-Tech segment growing double-digits. Adjusted EPS guidance of $8.43 to $8.63 for 2026, up 12 to 15%. This is not a headline story. It is a well-run compounder with growing exposure to the water constraint I’m invested in.

Phase 3b: Firm Clean Baseload

This covers both nuclear and geothermal, two technologies solving the same fundamental problem. AI compute needs power that doesn’t stop when the weather changes. This is the longest-duration portion of the thesis, and the holdings here are sized accordingly.

BWX Technologies (BWXT) is my highest-conviction position in this phase. The government anchor first: BWX holds a monopoly on naval nuclear propulsion in the United States. Every nuclear-powered ship and submarine runs on BWX reactors and fuel. That anchor generates roughly $5 to $6 billion of the $7.3 billion total backlog and is completely insulated from commercial market conditions, interest rate environments, and AI capex cycles. It’s not going anywhere.

On top of that anchor, the commercial nuclear business is growing. Total backlog is $7.3 billion, up 50% year over year. Commercial backlog specifically grew 85%, driven by CANDU reactor refurbishment contracts, SMR component design awards, and TRISO fuel production for advanced reactors. FCF was up 16%. A $900 million DOE HALEU task order was awarded in January, subject to final negotiation. The Antares microreactor at Idaho National Laboratory is targeting first criticality on July 4, 2026. That’s the nearest hard Phase 3b milestone in the portfolio and I’m watching it closely.

Centrus Energy (LEU) enriches uranium, including the HALEU that advanced reactors require. They launched domestic centrifuge manufacturing in December 2025, the first time in the company’s history, and the only commercial HALEU production capability in the United States. Total backlog extends to 2040 at $3.8 billion. Near-term revenue is flat at roughly $450 million while they deploy $350 to $500 million in capital this year. The value in this position is post-2030. I hold it sized for that horizon with patience built in. In March, Centrus and Oklo announced a joint venture for nuclear fuel services in Ohio, creating an operational linkage across two portfolio holdings in the same supply chain.

Oklo (OKLO) is building compact fast reactors. Roughly 18 gigawatts of non-binding pipeline agreements. $2.5 billion in cash, more than seven years of runway at current burn rates. Active construction at Idaho National Laboratory with a groundbreaking in September 2025. A 1.2 gigawatt prepayment agreement with Meta for an Ohio campus was announced in January. DOE Nuclear Safety Design Agreement approved March 17. NRC combined license application targeted Q4 2026.

Zero revenue. $15 billion market cap. I hold this as a lottery ticket: a position sized for the possibility of extraordinary long-term upside, accepted alongside the realistic possibility that meaningful commercial revenue is still many years away. The NRC CLA submission is the next hard gating event. Submission is progress. Delay is a risk signal.

Ormat Technologies (ORA) is the geothermal position. 1,340 megawatts of operating capacity globally. Clean, firm, dispatchable power that runs continuously without fuel cost. Google contracted 150 megawatts through NV Energy on a 15-year agreement. FY2025 revenue was $989.5 million, up 12.5%. 2026 guidance is $1.11 to $1.16 billion with adjusted EBITDA of $615 to $645 million. Curtailments that were costing the company $18.6 million in 2025 are projected to fall to $4 to $5 million in 2026 as grid integration improves.

Two concerns. Net debt is approximately 4.4 times EBITDA, elevated for a utility-adjacent business in this rate environment. Fervo Energy is approaching an IPO with $421 million in institutional project finance behind it, representing the first genuine geothermal competitor of scale. Neither is a thesis-breaker today. Both warrant watching. I hold this one on a tighter leash than the others.

FuelCell Energy (FCEL) belongs in the Phase 1 / firm power section, but I’m being candid about where this one stands. They generate power from fuel cells, including molten carbonate systems at scale. The US data center pivot they’ve been executing is strategically correct: there’s a genuine market for firm, on-site power at data centers, and they have the technology for it. The problem is the gap between strategy and execution. FY2025 GAAP net loss was $1.914 billion. Adjusted EBITDA was negative $74.4 million. Q1 FY2026 showed a gross loss, meaning they’re losing money on the product itself before any overhead. Backlog is roughly $1.17 billion and stagnant. The US data center pivot has zero binding contracts so far. Revenue is declining.

The thesis requires converting the pipeline to binding US contracts before the capital environment tightens past the point where equity raises become prohibitive. The clock is running. I’m running this position with ongoing reassessment each quarter and no additional patience budget.

The EPC and Grid Layer

Quanta Services (PWR) is the largest specialty electrical contractor in North America and my core EPC position. Backlog hit $43.98 billion, a record. Revenue up 20%. Free cash flow of $1.67 billion, also a record. The self-perform labor model, where Quanta employs its own craft workforce rather than subcontracting, is a genuine competitive moat that becomes more valuable as labor markets tighten. One flag from this quarter: operating cash flow was down 24.2% year over year while free cash flow hit a record, a divergence that needs explanation. The investor day on March 31 is the clarification event. The underlying demand is not the issue.

Primoris Services (PRIM) is my second EPC holding and the one I’m watching most carefully heading into earnings. Fixed backlog declined 18.2% to $5.0 billion, though MSA backlog, the recurring service agreement work, grew 20.6% to $7.0 billion. Gross margins contracted 120 basis points. The structure of Primoris’s contracts makes them more exposed to commodity input cost pressure than Quanta when material costs spike on fixed-price work. Diesel up 27%, steel up roughly 25%, copper up 36%: those numbers flow through the P&L on existing contracts. April earnings are the read. Margin compression from commodity inputs on existing contracts would be a meaningful signal that the macro pressure is reaching operating results, not just credit spreads.

Powell Industries (POWL) makes the high-voltage switchgear that steps down transmission-level power at the point where it enters a facility. Niche product. Three-plus year lead times. No credible new entrants with meaningful manufacturing capability at commercial scale. Backlog of $1.6 billion with a 1.7x book-to-bill ratio. Zero debt. $501 million in cash. Their first data center megaproject, worth $75 million, was announced this quarter, a meaningful penetration signal for a company of this size. This is the kind of specialized, defensible pick-and-shovel business that tends to compound quietly over a long infrastructure cycle.

The Financial Engine

Brookfield Asset Management (BAM) is a different kind of holding. Not an infrastructure operator. The largest private infrastructure capital allocator in the world, and the financial engine behind a significant portion of the buildout this entire portfolio is positioned around.

$134 billion in uncalled capital. $100 billion AI infrastructure program. A $20 billion AI joint venture with Qatar, called Qai. 49% ownership of Westinghouse, the dominant global nuclear reactor technology company. Fee-related earnings of $3.0 billion in 2025, up 22% year over year, with $112 billion raised across the year. The Oaktree 26% acquisition is expected to close in the first half of 2026 and adds meaningfully to the FRE base.

BAM trades at 7.9 times fee-related earnings. Comparable asset managers trade at multiples running from roughly 14 times to as high as 35 times. The discount to peers is 45 to 90% depending on which peer group you use. I think this is a market categorization error. Most analysts compare Brookfield to private equity managers and find their own metrics don’t translate cleanly, then move on. When you look at the specific business, long-dated and often perpetual-capital infrastructure and real assets with 87% of fee-bearing capital in long-dated or perpetual structures, and the role BAM plays as the primary private capital conduit for the infrastructure buildout, the discount is hard to explain on the fundamentals.

Brookfield’s CEO said explicitly this year that the bottleneck to AI growth is not capital and not demand. It is electricity supply. That is an independent, CEO-level, public confirmation of the exact thesis I’ve been running for a year, from the person running the largest private infrastructure allocator in the world. I find that notable.

The Most Acute Situation in the Portfolio

Fluence Energy requires its own section because what happened this quarter isn’t a minor miss.

Fluence makes battery energy storage systems, large-scale grid-level storage that is a critical complement to intermittent renewables and an increasingly important part of the behind-the-meter power story. The demand thesis is real. The backlog is $5.5 billion, a record.

Q1 FY2026 gross margin came in at 4.8% against a full-year target of 11 to 13%. The company consumed $820 million in cash in a single quarter from working capital build, taking the balance sheet from roughly $1.3 billion to $477.8 million. FY2026 revenue guidance midpoint is $3.4 billion at approximately 1.5% EBITDA margin. At the current cash consumption rate, two to three more deterioration quarters are existential for this balance sheet.

I trimmed the position significantly after these results. The binary catalyst is set for Q2 FY2026 gross margin, reported around June 2026. Recovery above 11% confirms this was a timing and cost problem, not a structural one. Below 11% triggers a full exit evaluation. There is no ambiguous middle outcome. The demand thesis being right doesn’t help a company that runs out of cash before it can execute on it.

The Scorecard

Where every holding stands. Not a buy list. A status report on the thesis at each position.

STRENGTHENED:

GEV  — Phase 1 (Gas Turbine / Firm Power) — Self-Funding
  $150.2B backlog (+$31.2B YoY). $8.8B net cash. FCF $3.7B (+118%). ~2x book-to-bill.
  Most insulated holding. Benefits structurally from KC4 competitor financing pressure.

VRT  — Phase 2 (Liquid Cooling) — Self-Funding
  $15.0B backlog (+109% YoY). 2.9x book-to-bill. 23.2% adj. margins. FCF ~$1.9B.
  Monitoring note: quarterly order reporting discontinued. Annual 10-K now primary read.

POWL — Phase 1/Cross (HV Switchgear) — Self-Funding
  $1.6B backlog, 1.7x book-to-bill. Zero debt. $501M cash.
  First DC megaproject ($75M) announced. 3-plus year lead times. No new entrants.

BWXT — Phase 3b (Nuclear / Government-Anchored) — Self-Funding
  $7.3B total backlog (+50% YoY). Commercial backlog +85%. FCF +16%.
  $900M DOE HALEU task order. Antares criticality target: July 4, 2026.

───────────────────────────────────────────

INTACT:

PWR  — Cross-Phase (EPC/Grid) — Self-Funding
  $43.98B record backlog. Revenue +20%. FCF $1.67B record.
  OCF/FCF divergence watch. March 31 Investor Day is the clarification event.

JCI  — Phase 2 (Cooling) — Self-Funding
  $18B record backlog. Orders +40% YoY from data centers. NVIDIA collaboration confirmed.
  Adj. EPS guidance ~$4.70 (+25%). Trending toward direct hyperscaler classification.

SPXC — Phase 2/3 (Cooling/Water) — Financing-Dependent
  DC revenue ~$200M FY2025 (first quantification). HVAC backlog +22%, D&M +43%.
  Full financials pending 10-K verification. ~34% GAAP-adjusted delta on earnings.

XYL  — Phase 3 (Water) — Self-Funding
  $4.6B backlog. Guidance miss = simplification/mix story. High-Tech growing double-digits.

ECL  — Phase 3 (Water) — Self-Funding
  Adj. EPS guidance $8.43–$8.63 (+12–15%). Recurring compounder. High-Tech double-digits.

LEU  — Phase 3b (Nuclear Fuel) — Financing-Dependent
  $3.8B backlog to 2040. $2.0B unrestricted cash. Revenue flat ~$450M near-term.
  Capital deployment $350–500M in 2026. Centrus-Oklo JV announced March 9. All value post-2030.

OKLO — Phase 3b (Microreactor) — Pre-Revenue
  ~18 GW pipeline. $2.5B cash (7+ year runway). Aurora-INL active construction.
  Meta 1.2GW agreement. NRC CLA targeted Q4 2026. Lottery ticket sizing.

ORA  — Phase 3b (Geothermal) — Financing-Dependent
  1,340 MW operating. Google 150 MW PPA (15-year). 2026 revenue guidance $1.11–1.16B.
  Net debt/EBITDA 4.4x. Fervo IPO = first real competitive entry. Tighter leash.

BAM  — Cross-Phase (Capital Allocator) — Self-Funding
  7.9x FRE vs. 14–35x for peers. $3.0B FRE (+22%). $134B dry powder. $112B raised 2025.
  CEO independently confirmed the thesis. Structural discount. Oaktree close H1 2026.

───────────────────────────────────────────

ON WATCH:

PRIM — Cross-Phase (EPC) — Financing-Dependent
  Fixed backlog -18.2% to $5.0B. Gross margin -120bps. MSA backlog +20.6% to $7.0B.
  Q1 2026 earnings (April) is the read on commodity cost margin compression.

BE   — Phase 1 (Behind-the-Meter) — Financing-Dependent
  $6B product backlog (+140% YoY). 6 hyperscalers. 2026 guidance $3.1–3.3B (+54–63%).
  94x EV/Non-GAAP op income. $2.5B convertible debt. $5B BAM supply commitment.
  Demand thesis intact. Valuation and financing environment both requiring close attention.

FLNC — Phase 1 (Battery Storage) — Financing-Dependent
  4.8% gross margin vs. 11–13% target. $820M cash consumed in single quarter.
  $477.8M cash remaining. 2–3 quarters to existential at current trajectory.
  BINARY CATALYST: Q2 FY2026 gross margin (~June 2026).
  Above 11% = timing confirmed. Below 11% = damage assessment, exit evaluation begins.

───────────────────────────────────────────

SPECIAL CLASSIFICATIONS:

ERII — Phase 3 (Desalination) — DELAYED, NOT DAMAGED
  Revenue $135M (-7% YoY). 2026 guidance $105–125M. Three projects slipped to 2027.
  Gross margins 65–67% intact through trough. FCF positive. Near-zero debt.
  EV/EBITDA ~5.7x trough. PX Q650 launch H2 2026. Amundi +323.6% Q4 2025.
  Five independent analyses: delay, not structural loss. 2027 FIDs are the verdict.

FCEL — Phase 1 (Fuel Cell) — WEAKENING
  $1.914B GAAP net loss FY2025. Backlog stagnant ~$1.17B. Q1 FY2026 gross loss.
  US data center pivot: zero contracts secured. Korea-concentrated.
  Ongoing quarterly reassessment. No additional patience budget.

What the distribution tells you. Every strengthened holding is self-funding. Every challenged or weakening holding is financing-dependent. This isn’t coincidental. It is the direct result of the credit bifurcation described in the macro section playing out at the operating company level. The thesis is intact for the companies that don’t need to ask anyone’s permission to keep building. It’s pressured for the ones that do.

Two Gaps I Found This Quarter

Part of my quarterly process is looking for structural exposures this portfolio is missing. Two came up that I haven’t resolved yet.

Uranium mining. I own the full downstream nuclear supply chain: enrichment through Centrus, reactor technology through BWX and indirectly Westinghouse through Brookfield, microreactor development through Oklo. I own nothing at the mine. Annual global reactor demand for uranium is roughly 180 million pounds. Primary mine production in 2025 was roughly 130 million pounds. That is a 28% structural production deficit. Kazatomprom, the world’s largest uranium producer, is cutting its 2026 production guidance from roughly 85 million pounds to 77 million pounds, widening the gap further. The AP1000 partnership announced in October 2025, involving Cameco, BAM’s Westinghouse, and the US Government, targets 8 to 10 new AP1000 reactors by 2030 with over $80 billion in federal support and roughly 65 million pounds of incremental uranium demand over 10 years.

Cameco is the obvious public proxy: NYSE-listed, net cash positive, profitable, revenue of $3.48 billion with 11% growth in FY2025, dividend up 50%. I’m not in it yet because a uranium equity sector index I use as a health signal is approaching a threshold that tells me to wait. When that clears, I look harder at CCJ.

HVDC cable. High-voltage direct current cable is what physically moves power over long distances with minimal loss. I own the contractors executing the transmission buildout. I don’t own anything in the cable supply chain those contractors depend on. Prysmian, the dominant HVDC cable manufacturer, has €16 billion in contracted transmission backlog as of Q3 2025, with the Eastern Green Link 4 contract, 640-plus kilometers of cable for National Grid, added in February 2026. Cable manufacturing captures margin that EPC contractors don’t. Primary listing friction, Milan Stock Exchange with limited US OTC access, and €4.3 billion in net debt at an uncomfortable point in the credit cycle are the friction points. Watching it, not acting yet.

What I’m Watching in Q2

March 31: Quanta investor day. New long-term financial targets and the OCF/FCF divergence explanation. Most important near-term event for understanding EPC execution health.

April 30: FERC rulemaking deadline on interconnection reform. Could affect timing and economics for Phase 1 project routing.

April through May: Hyperscaler Q1 2026 earnings. The primary quarterly test of whether EXPANSION DOMINANT holds. Any shift in language from supply-constrained to “optimizing deployment” or “digesting capacity” is a meaningful signal that warrants immediate reassessment. I’ll be reading the Microsoft Azure backlog commentary, Amazon data center capacity language, and Meta infrastructure spending confirmation closely.

April through May: PRIM, PWR, and SPXC earnings. Commodity cost margin compression on existing contracts is the specific indicator I’m watching. That would confirm the macro pressure is reaching operating results, not just credit spreads.

April: BDC earnings. The private credit $1.8 trillion redemption wave is currently two-of-three source confirmed. A third hard source from BDC earnings would escalate the credit stress picture meaningfully.

May: BE earnings. First real-time test of the $3.1 to $3.3 billion revenue guidance. Customer concentration disclosure expected with FY2025 10-K, which will show whether the SK ecoplant diversification from FY2024’s 53% top-three concentration is real.

June: FLNC Q2 FY2026 gross margin. Above 11%: timing problem, thesis intact. Below 11%: full exit evaluation begins. No other data point matters for that holding.

July 4: BWXT Antares criticality target at Idaho National Laboratory. The nearest hard Phase 3b milestone of the year and a validation event for HALEU fuel supply.

The Bottom Line

The thesis is intact. The demand for the infrastructure this portfolio is built around received its strongest quarterly confirmation since I started running this analysis. The physical constraints that create the investment opportunity, the power scarcity, the cooling requirement, the water problem, the nuclear and geothermal baseload need, the EPC labor ceiling, are all real and all worsening rather than easing.

The complication is that some of the companies best positioned to capture that demand are navigating a financing environment that became less forgiving this quarter, at exactly the moment their execution needs to be most reliable. That tension is not going away before April earnings. It’s probably going to sharpen.

Here’s what I believe about how it resolves. The self-funding Fortress names, GEV, VRT, POWL, BWXT, PWR, BAM, are going to compound through this environment because their business models don’t require anyone’s permission to keep operating. The financing-dependent names are going to diverge based on whether they can reach their revenue inflection point before the capital environment makes that too expensive. Some of them will. Some of them are in a race.

Fluence is in that race. FuelCell is further behind in it. Bloom Energy is running it at a valuation that doesn’t allow for much stumbling. Energy Recovery isn’t in the race at all, in my view. It’s waiting on project timing that the market has misread as structural failure.

I’ll run this same analysis at the end of Q2. The next six weeks of earnings will tell me a lot about which scenario we’re actually in.

If you want the full version of this, the 22-page distilled brief or the 50 to 60 pages of underlying work, reach out directly. Happy to share it.

Disclosure: The author holds positions in all companies discussed in this piece. This is not investment advice. Do your own research.

Share

Freedom Grid | Q1 2026 | freedomgrid.substack.com

The infrastructure signals and the macro environment aren’t lining up.

That’s not noise. That’s the setup.

Subscribe now

How This Works

Freedom Grid tracks the physical infrastructure required to power the AI economy — not the software, the chips, or the models. The systems underneath all of it.

There are four phases to this buildout that I track, and they unfold in sequence.

Phase 1 — Power Generation and the Grid. AI data centers need electricity. Generation capacity must expand and transmission infrastructure must be built to deliver it. This is the current acute constraint, visible in interconnection queue backlogs, transformer shortages, and the federal emergency orders covered below.

Phase 2 — Cooling Infrastructure. Higher-power chips generate more heat. The thermal design power of AI chips rises with each generation, and rack power density in data centers is now tracking toward 50 to 100 kilowatts per rack. Air cooling stops working around 20 kilowatts. Liquid cooling becomes the only option at scale.

Phase 3 — Water Infrastructure. Cooling systems reject heat through water. As data center density grows, so does industrial water demand, creating constraints around water rights, municipal supply, and zero-liquid-discharge regulations that have 10- to 15-year resolution timelines.

Phase 3b — Firm Baseload: Nuclear and Geothermal. Intermittent renewables can’t serve always-on AI workloads. The AI economy needs power that is always on, carbon-free, and dispatchable on demand. Two technologies fit that profile at scale: nuclear and geothermal. Nuclear provides firm baseload through fission; small modular reactors are factory-built and site-flexible with 15- to 20-year development timelines. Geothermal taps heat from the earth and runs continuously regardless of weather, with revenue expected to inflect meaningfully in the 2027 to 2029 window as project finance matures. Neither technology generates meaningful revenue on a short timeline. That is not a flaw in the thesis. It is the nature of building the grid’s long-term foundation.

The portfolio tracks companies across all four phases. Each week, this brief assesses where the buildout stands, what signals confirm or challenge the thesis, and what risks are growing.

The Week

Start with what actually happened. Iraq declared force majeure on all foreign-operated oilfields Friday. Iranian drones struck two refineries in Kuwait. Israel hit Iran’s South Pars gas field, the world’s largest. Iran retaliated by striking Qatar’s Ras Laffan complex, the world’s largest LNG processing hub. Qatar supplies roughly 20% of global LNG. The QatarEnergy CEO confirmed to Reuters that repairs will take three to five years and that the strike knocked 17% of Qatar’s LNG capacity offline. Brent crude closed the week at $112.19, up 9%. European natural gas surged more than 25% in a single session.

That’s the geopolitical frame. It matters, but not primarily as a market volatility story. The physical damage to Ras Laffan is independent of any diplomatic outcome. Even a ceasefire tomorrow doesn’t rebuild the facility. What the Ras Laffan strike does, structurally, is strengthen the case for domestic dispatchable power generation in the U.S. and Europe by removing any remaining illusion that global LNG supply is a reliable backstop. Every government watching that news feed reassessed its energy security position this week.

But the more substantive story for this thesis, the one that won’t lead the news cycle, is what happened to the infrastructure buildout independent of the Middle East. This was a heavy week across all four phases simultaneously, and the signals mostly pointed in the same direction.

Practitioners confirmed the gas turbine supply chain has two or three validated global suppliers. The federal government used emergency powers to keep coal plants online because the grid cannot afford to lose them. GE Vernova and Hitachi confirmed a $40 billion nuclear commitment. Geothermal crossed a major financing milestone. Google signed a deal for more than two gigawatts of clean energy in Michigan. The Federal Reserve officially confirmed the U.S. economy is in stagflation. And a new financial risk to early-stage infrastructure companies became materially more credible.

That’s worth unpacking carefully.

The Macro Picture

The data is now unambiguous. Q4 GDP came in at 0.7% annualized. Core PCE, the Fed’s preferred inflation gauge, sits at 3.1% year-over-year. The Fed held rates at 3.50 to 3.75% on March 18, as expected. The dot plot reduced projected 2026 cuts. Some FOMC members have reportedly begun discussing whether the next move is a hike rather than a cut. Macquarie published a note calling it explicitly: next Fed move is a rate hike, pushed to the first half of 2027. The market stopped debating when cuts are coming. It started pricing in the possibility of a hike.

Low growth, sticky inflation, a central bank with no good options. That’s stagflation, now three consecutive weeks of data confirming it rather than suggesting it.

The 10-year Treasury closed at 4.39%, the highest since July 2025. The 30-year is approaching 5%. These numbers matter more than the weekly market moves, because they set the rate at which future cash flows get discounted. Companies that generate substantial cash today are less affected. Companies building infrastructure that won’t produce meaningful revenue for years get hit harder, as their value is almost entirely in future earnings, and those earnings become worth less as the discount rate rises.

Markets reflected this clearly. The S&P 500 posted its fourth consecutive weekly loss, the longest losing streak in a year, closing at 6,506. The Russell 2000 slipped into correction territory, down more than 10% from its recent high. Energy and financials were the only sectors positive on Friday. The sector rotation is directionally consistent with what you’d expect when Phase 1 recognition advances, as energy infrastructure investment becomes more attractive when energy supply tightens and domestic generation urgency rises.

One point worth being direct about: the infrastructure thesis was designed to hold up in this macro environment, because the demand for the physical systems behind AI doesn’t respond to Fed funds rate changes. Data centers don’t cool themselves because interest rates are high. But equity prices of the companies building those systems absolutely respond to rates in the short term. The divergence between structural demand and equity price action is the gap the thesis is designed to exploit over time. That gap widened materially this week.

The Infrastructure Stack

Phase 1 — Power Generation and the Grid

Two signals confirmed Phase 1 status this week from entirely different directions.

The first: a report on off-grid data center development, authored by practitioners from Turner & Townsend, Sorellis, Skybox Datacenters, Jupiter Island Capital, and Charles River Associates, confirmed that the gas turbine supply chain has only two or three validated global suppliers capable of delivering at the scale these projects require. Not dozens of suppliers with some capacity constraints. Two or three. The same supply chain now faces demand from data centers, grid expansion, and a $40 billion nuclear commitment, simultaneously. That concentration will not resolve on a short timeline.

The second: at its March meeting, the Federal Energy Regulatory Commission issued two emergency orders under Section 202(c) of the Federal Power Act, forcing coal-fired power plants in Indiana to remain online. The plants, operated by Southern Indiana Gas and Electric and Northern Indiana Public Service, were scheduled for retirement. FERC overrode the retirement timeline. Grid operators in the Midwest determined the regional grid could not absorb those retirements without threatening reliability.

Section 202(c) orders are not routine. They are the grid equivalent of calling in reserves because normal system operation isn’t going to work. The fact that the federal emergency apparatus is deploying these tools at exactly the moment global LNG supply faces a structural multi-year reduction is worth sitting with. Two independent constraint cycles, domestic generation adequacy and global energy supply, are compounding simultaneously.

On transmission: FERC approved a 115-mile, 765-kilovolt power line across central Virginia, developed by Dominion, Transource, and FirstEnergy. NextEra separately filed for a 107.5-mile, 500-kilovolt line through Pennsylvania, Maryland, West Virginia, and Virginia. Large capital commitments to the transmission backbone that mid-Atlantic data centers will depend on.

The most consequential near-term regulatory event remains the April 30 FERC deadline on large-load interconnection rulemaking. This rulemaking will determine how major new power consumers connect to the grid and on what timeline. The March meeting produced zero visible action on that docket. Forty-two days remain. Whatever happens at or before that deadline shapes whether grid-connected infrastructure timelines begin compressing or continue to extend.

Phase 2 — Cooling Infrastructure

The same practitioner report confirmed the Phase 2 thesis as cleanly as any document I’ve seen. Rack power density in data centers is now tracking toward 50 to 100 kilowatts per rack. Air cooling handles roughly 10 to 20 kilowatts per rack before it stops working. Liquid cooling at 50-plus kilowatts per rack is not an upgrade option. It’s physics.

This doesn’t respond to Fed policy, regulatory timelines, or oil prices. The chips get hotter with each generation. The racks have to handle it.

Phase 3 — Water Infrastructure

Quieter week for Phase 3 signals. The structural thesis remains intact, as cooling systems reject heat through water, and as data center density grows, so does industrial water demand. Recognition of this phase in the market is still early relative to Phase 1. State-level community opposition to data center water use is gaining legislative traction in several states, but no material new developments this week.

Phase 3b — Firm Baseload: Nuclear and Geothermal

This was the most active phase for positive signals this week, and the signals came from both sides of the baseload equation.

Nuclear. The US-Japan joint statement confirmed a commitment of up to $40 billion for GE Vernova and Hitachi to deploy advanced nuclear reactors in Tennessee and Alabama. Bloomberg, Canary Media, and Axios each independently confirmed the deal. Three independent confirmations of the same hard number from a joint government statement. This is the largest stated nuclear investment commitment in decades.

The Department of Energy announced up to four advanced reactor pilot projects targeting July 4 as a criticality milestone. Minnesota moved to consider lifting its 32-year nuclear moratorium, with bipartisan legislative support. TerraPower’s NRC permit process is advancing. Oklo received a license from the Nuclear Regulatory Commission on March 17.

Geothermal. Fervo Energy closed $421 million in institutional debt financing from RBC, Barclays, HSBC, and JPMorgan. Bloomberg Green covered it as geothermal crossing what the industry calls the “bankability divide” — the point at which major institutional lenders treat the asset class as financeable rather than experimental. That milestone matters beyond Fervo specifically. It signals that the capital markets are beginning to price geothermal as a real infrastructure asset class, which is a precondition for the sector scaling.

Geothermal’s value proposition is the same as nuclear’s: firm, dispatchable, carbon-free power that runs regardless of weather or time of day. The difference is timeline and cost profile. Geothermal projects in development today are expected to reach meaningful revenue in the 2027 to 2029 window, a nearer horizon than the first commercial SMR deployments. The Fervo financing is evidence that the project finance community is starting to treat that timeline as credible.

The Google Clean Transition Tariff framework, now spanning 25 tariffs across 19 states, has geothermal at its origin. The first CTT deal was a Nevada geothermal project. The framework that is now underpinning five gigawatts of new clean energy commitments was built around the structure of a geothermal offtake agreement.

The nuclear thesis is not a bet on when the first reactor turns on. The geothermal thesis is not a bet on Fervo’s IPO valuation. Both are bets on whether the policy environment, the capital environment, and the supply chain are genuinely orienting toward firm baseload as a real solution for AI load. This week added evidence across both simultaneously.

Cross-Phase — Behind the Meter and Capital Commitments

The White House formalized a “bring your own power” mandate through a Ratepayer Protection Pledge, effectively requiring large hyperscalers to be responsible for sourcing their own generation when connecting to the grid at scale. This is enacted policy, not a proposal.

Google’s Clean Transition Tariff framework now underpins approximately five gigawatts of new energy commitments across 25 tariffs in 19 states. This week, Google signed a deal with DTE Energy for more than two gigawatts in Michigan: 1.6 gigawatts solar, 400 megawatts storage, 300 megawatts bilateral. Duke Energy in North Carolina agreed to evaluate the same framework, the first major Southeast utility to do so.

The DOE, SoftBank, and AEP announced a 10-gigawatt power development deal in Ohio, including 9.2 gigawatts of gas capacity. Amazon’s $54 billion bond deal drew $126 billion in orders, 14 times oversubscribed. Bank of America revised its 2026 hyperscaler debt issuance forecast to $175 billion.

The “AI infrastructure buildout is cracking” narrative that circulated in newsletters two weeks ago has largely disappeared. It moved from active debate to a watchlist item this week, overwhelmed by the volume of confirmed capital commitments pointing the opposite direction.

The Narrative Landscape

Three narratives are converging across independent sources this week. When multiple credible outlets independently reach the same conclusion without coordinating, it’s worth paying attention.

The first convergence is on the structural nature of the energy supply shock. Most coverage frames the Middle East situation as a price spike that resolves with diplomacy. The more credible read is that the Ras Laffan repair timeline is already priced in regardless of diplomatic outcome. QatarEnergy’s CEO confirmed three to five years to Reuters. That estimate doesn’t change if Iran and Israel stop shooting at each other next week. European natural gas prices have a higher structural floor than they did seven days ago. That floor accelerates investment in domestic dispatchable generation in ways that persist beyond the current conflict’s resolution.

The second convergence is on nuclear. Four or more independent sources covered the nuclear pathway this week as genuinely advancing rather than perpetually pending. Bloomberg, Canary Media, Axios, and trade press each confirmed the US-Japan commitment independently. The DOE pilot reactor program moved from proposal to specific milestone target. Minnesota’s legislative movement was bipartisan. These aren’t the same story filed from different angles. They’re separate confirmations of a directional shift in nuclear deployment momentum.

The third convergence is on behind-the-meter institutionalization. The Clean Transition Tariff framework moved from a single Nevada geothermal experiment to a 25-tariff, 19-state, multi-gigawatt program in a matter of months. The White House mandate turned utility-level experimentation into federal policy. These transitions tend to accelerate once they reach a certain scale, because each new deal lowers the cost basis and expands the legal precedent for the next.

Two narratives are diverging without resolution. Whether the Iran-Hormuz situation escalates into a long-duration structural conflict or de-escalates through diplomacy remains genuinely uncertain. (The physical damage to Ras Laffan, again, is already done on either path.) The second: whether U.S. shale producers respond to $112 oil with aggressive new drilling or maintain capital discipline. Analysts at Rystad and Rapidan project only about 200,000 additional barrels per day of incremental U.S. production even at current price levels, as producers are watching the forward price curve decline and choosing not to lock in drilling economics at what they read as a potentially temporary peak.

Two new signals emerged this week that I’m watching but not yet treating as confirmed. Battery prices in China rose in early 2026, reversing the 2024-2025 cost decline trend the entire energy storage industry was planning around, reported via Bloomberg NEF data. If this confirms over a second quarter, it changes the economics of battery storage projects globally. Separately, developers are converting old downtown buildings into data centers to avoid suburban community resistance. Power density limitations make this a supplemental pathway at best, but the trend signals that permitting friction is real enough to change project siting decisions.

Holdings — What Moved This Week

Two categories of companies sit in this portfolio. Fortress names are large, established companies with strong balance sheets and existing revenue — their thesis is about capturing structural infrastructure demand. Financing-Dependent names are earlier-stage companies whose thesis depends on continued access to growth capital, often before they generate meaningful revenue. The macro environment affects both, but it hits Financing-Dependent names harder when rates rise and credit tightens.

Below are the holdings with material developments this week. Holdings with no changes — Powell Industries, Johnson Controls, Ecolab, SPX Technologies, Xylem, Centrus Energy, Oklo, Energy Recovery, and Primoris Services — maintained their prior thesis status.

GE Vernova (GEV) | Phase 1 + Phase 3b | Fortress | STRENGTHENED

The $40 billion US-Japan nuclear commitment directly names GE Vernova and its Hitachi partnership, with sites identified in Tennessee and Alabama. Three independent confirmations. The DOE/SoftBank/AEP 9.2-gigawatt gas commitment adds further to an order book already under production pressure. The demand signal is the strongest it’s been.

The open variable is whether GEV can ramp production capacity to meet $40 billion in new nuclear commitments on top of existing demand. Gas turbine supply chains have two or three validated global suppliers, and GEV is one of them. Q1 earnings in April is the first verification event for the production ramp question.

Vertiv Holdings (VRT) | Phase 2 | Fortress | STRENGTHENED

S&P 500 inclusion executes Monday, March 23, forcing passive fund inflows. The 50-to-100 kilowatt-per-rack trajectory from the practitioner report is the cleanest Phase 2 confirmation available, coming from the people designing these projects rather than from market analysts. Liquid cooling demand does not negotiate with interest rates. Thesis executing.

Quanta Services (PWR) | Cross-Phase EPC & Grid | Fortress | STRENGTHENED

Two new hard backlog inputs this week. The PJM 765kV Virginia transmission approval and NextEra’s 500kV multi-state filing both generate direct construction work in Quanta’s core territories. FERC emergency orders add generation and transmission work in the Midwest. The risk is on the cost side: diesel, copper, and steel are all at elevated levels. Q1 earnings need to confirm that PWR is recovering those costs through contract pricing rather than absorbing them as margin compression. That answer arrives in April.

BWX Technologies (BWXT) | Phase 3b | Fortress | STRENGTHENED

Every nuclear signal this week is a positive read for BWXT, which manufactures components for U.S. reactors including advanced reactor programs. The Japan-GEV deal directly accelerates the SMR pipeline that BWXT’s manufacturing capacity will serve. The equity price has not reflected the nuclear narrative acceleration this week. That gap doesn’t persist indefinitely.

Bloom Energy (BE) | Phase 1 / Behind the Meter | Financing-Dependent | STRENGTHENED

The White House “bring your own power” mandate is a direct policy tailwind for BE’s core business model. The Google-DTE Michigan deal falls in territory where BE has existing deployments. The Clean Transition Tariff multi-state expansion increases the addressable market for behind-the-meter firm power, which is precisely BE’s value proposition. No direct BE announcements this week; the strengthening is real but indirect.

Ormat Technologies (ORA) | Phase 3b Geothermal | Financing-Dependent | CHALLENGED

Fervo Energy closed $421 million in institutional debt financing from RBC, Barclays, HSBC, and JPMorgan this week. For the geothermal sector broadly, this is a milestone. It confirms that major banks will lend to geothermal development projects. That’s the kind of bankability signal the entire sector has been waiting for.

The problem is that Fervo is now a well-capitalized competitor heading toward an IPO, competing directly with ORA for utility offtake contracts and institutional capital allocation. Bloomberg Green ran a dedicated feature on Fervo’s “geothermal IPO moment.” The sector got validated this week. ORA got a better-funded rival. That’s a mixed outcome, and I’m watching it carefully.

FuelCell Energy (FCEL) | Phase 1 / Behind the Meter | Financing-Dependent | DETERIORATED

Goldman Sachs and JPMorgan are reportedly facilitating hedge fund short positions against the $1.8 trillion business development company market. BlackRock, Morgan Stanley, and Cliffwater have imposed redemption limits on private credit funds. This matters for FCEL because the company is pre-revenue and dependent on alternative financing channels, specifically the channels that business development companies and private credit funds provide. The thesis has not broken, but the financing environment around it is getting more hostile. No exit trigger yet, but the trajectory is the wrong direction.

Fluence Energy (FLNC) | Cross-Phase Storage | Financing-Dependent | DETERIORATED

The half-position exit executed earlier was the right call. This week added a new risk layer: Bloomberg NEF data shows battery prices in China rose in early 2026, reversing the cost decline trend the entire energy storage industry was planning around. Fluence was already under margin pressure. An input cost reversal on top of existing compression makes the remaining position’s recovery path harder. A second consecutive poor quarter at Q1 earnings moves this toward full exit evaluation.

Brookfield Asset Management (BAM) | Cross-Phase Capital | Fortress | DETERIORATED

BAM is the portfolio’s exposure to infrastructure finance at scale. The private credit redemption wave is a direct headwind for BAM’s ability to raise new funds and deploy capital into infrastructure projects. Fortress classification remains intact, as BAM has the balance sheet to weather a difficult fundraising period. But the headwind is real. BDC earnings in April are the next signal.

The Risk Picture

The most credible financial risk to this portfolio right now is credit tightening cutting off access to capital for early-stage infrastructure companies. Here’s where that stands.

High-yield credit spreads, the premium investors demand to hold below-investment-grade bonds over Treasuries, are at approximately 3.28 to 3.35 percent. The threshold that historically signals a material tightening in credit conditions sits at 3.50 percent. We are 22 basis points away from that line. At the same time, four commodity input categories are simultaneously above historical stress levels: diesel, copper, aluminum, and oil. Steel is approaching the same threshold. The Fed confirmed this week it will not cut rates in 2026 and may raise them in 2027. And the private credit redemption wave just added a new tightening channel that high-yield spreads don’t fully capture.

The resolution event is Q1 earnings season in April and May. If the EPC contractors and infrastructure builders in this portfolio can demonstrate they’re recovering elevated input costs through contract pricing rather than absorbing them as margin losses, this risk scenario de-escalates. If Q1 earnings confirm margin compression without recovery, the scenario advances meaningfully.

A note on geopolitical risk: the distinction that actually matters is between continued attacks on energy infrastructure in the Gulf region and a physical closure of the Strait of Hormuz. Attacks on specific facilities, serious as they are, do not stop tanker traffic through the Strait. A physical Hormuz closure changes the calculus for every energy-intensive project globally, including U.S. data center construction. Vortexa tanker flow data and QatarEnergy’s official communications on Ras Laffan repair timeline are the two signals to watch.

On the policy front: the federal direction remains pro-buildout. FERC emergency orders, the White House power mandate, and the DOE pilot reactor program all confirmed that this week. The state-level picture is more complicated. North Carolina is pushing a data center permitting moratorium. Pennsylvania has an active political fight over data center siting. Kentucky extended a wind moratorium. These are reroutes, not cancellations. Capital is moving to Texas and Wyoming, which have no equivalent friction. But the Southeast’s advantage as a low-friction buildout geography is narrower than it was a month ago.

What to Watch Next Week

VRT’s S&P 500 inclusion executes Monday. The passive inflow magnitude is worth watching as a signal of how institutional capital is currently pricing Phase 2 infrastructure exposure.

Hormuz — daily monitoring required. Vortexa tanker flow data and any QatarEnergy official updates on Ras Laffan are the signals that determine whether the energy supply shock has a structural multi-year implication or a potential near-term resolution pathway.

FERC between-meeting activity. FERC can issue orders between its monthly meetings. Any action on the large-load interconnection rulemaking ahead of the April 30 deadline would be immediate news for grid-connected infrastructure timelines.

BLS construction employment data. If immigration enforcement is appearing in specialty contractor workforce numbers, that upgrades the labor shortage from a project risk to a structural constraint.

The one thing I’m watching most: private credit BDC earnings season begins in April. The Goldman/JPMorgan short positioning against the $1.8 trillion business development company market is a new dynamic not yet fully visible in high-yield spreads. How BDC managers characterize their loan books at Q1 earnings will either confirm or deflate this as a serious financing risk for early-stage infrastructure companies. That answer has more near-term impact on the Financing-Dependent names than anything happening in the Middle East right now.

The Bottom Line

1. The gas turbine supply chain has two to three validated global suppliers — confirmed by project practitioners.Hard evidence · Troutman Pepper Locke / Utility Dive via Latitude · Escalated New demand of $40 billion in nuclear commitments and 9.2 gigawatts of gas was added to a supply chain that was already capacity-constrained. Phase 1 constraint is tighter than market pricing appears to reflect.

2. Qatar’s Ras Laffan complex faces a 3-to-5-year repair timeline with 17% of LNG capacity offline — independent of any diplomatic resolution. Hard evidence · QatarEnergy CEO to Reuters · Escalated, structural This is a permanent supply reduction for the duration of the repair, not a temporary disruption. It strengthens the energy security case for domestic dispatchable generation regardless of how the conflict ends.

3. Stagflation is confirmed: Q4 GDP at 0.7%, core PCE at 3.1%, a Fed that has removed the rate-cut safety valve and is discussing hikes. Hard evidence · BEA / BLS / Fed March 18 statement · Escalated, regime with duration The 10-year at 4.39% sets a higher discount rate for every long-duration infrastructure position. This is the macro environment the thesis was designed to navigate, but it requires patient capital, not reactive selling.

4. Nuclear had its best week in years, and geothermal crossed the institutional financing threshold: $40 billion US-Japan commitment, four DOE pilot reactors, Minnesota bipartisan momentum, an Oklo NRC license, and Fervo’s $421 million from four major banks. Hard evidence · Bloomberg / Canary Media / Axios / Bloomberg Green · EscalatedPhase 3b is no longer a distant thesis. The policy environment, capital environment, and supply chain are orienting toward firm baseload simultaneously across both nuclear and geothermal. The buildout timeline is compressing, not extending.

5. The most credible near-term risk: credit tightening cuts off capital for early-stage infrastructure companies.Hard developing · HY spreads / Bloomberg reporting on private credit · Escalating High-yield spreads are 22 basis points from a historical stress threshold. Four commodity categories are simultaneously in breach. Private credit funds are imposing redemption limits. Q1 earnings in April-May are the resolution event; the answer comes in six weeks.

Disclosure: The author may hold positions in companies discussed in this article. This publication does not provide investment advice.

Subscribe now

The Shift That Actually Matters

For most of its life, Oklo lived in the same bucket as every other SMR name:
designs, simulations, and very confident slide decks.

That changed in the latest cycle.

Based on the 10-K, earnings transcript, and company update materials, Oklo has moved from reactor concept to early-stage deployment.

That doesn’t mean success is guaranteed. It means something much simpler and much more important:

They’re now exposed to reality.

At the Idaho National Laboratory (INL) site, the company has:

• completed site characterization

• begun earthworks and blasting

• initiated long-lead equipment procurement

• contracted major partners, including Siemens Energy for power conversion

That’s not theoretical progress. That’s capital being committed and timelines starting to matter.

Aurora: The Core Bet Is Now a Physical Project

Everything still revolves around the Aurora reactor.

From the materials, the key progress is not technological. It’s procedural and operational:

• DOE authorization pathway is active

• Preliminary safety documentation approved

• Nuclear Safety Design Agreement in place

These are the gates that typically kill nuclear projects. Clearing them doesn’t guarantee success, but it removes entire categories of failure.

The current target remains:

• First heat production ~2028

Which, for nuclear, is actually believable.

But let’s not pretend otherwise:

This is still a late-decade revenue story at best.

Demand Is Real. Commitment Is Not (Yet)

The demand side is where things get interesting and where most people get sloppy.

Oklo is positioning itself as baseload power for data centers, which aligns almost perfectly with the Freedom Grid thesis around AI-driven load growth.

They’ve announced:

• a planned 1.2 GW campus tied to Meta (initial ~150 MW phase)

• a 12 GW master agreement with Switch

• engagements with Equinix and others

Here’s the part people conveniently blur:

These are not all firm, bankable power purchase agreements (PPAs).

They are a mix of:

• framework agreements

• exploratory partnerships

• early-stage commitments

Translation:

The demand signal is real. The contractual certainty is not.

That distinction matters more than anything else at this stage.

Fuel Strategy: Quietly One of the Most Important Pieces

Most SMR companies treat fuel like it’s someone else’s problem.

Oklo does not.

They’re pursuing three parallel paths:

• HALEU (High-Assay Low-Enriched Uranium) sourcing, including work with Centrus

• Fuel recycling, including a planned ~$1.68B facility in Tennessee

• Alternative fuel pathways, including plutonium-based fast-spectrum testing

This is not a side detail. This is survival.

Because right now, HALEU supply in the West is constrained enough that it can bottleneck the entire SMR ecosystem.

If Oklo can partially control fuel availability, it becomes more than a reactor company.

It becomes a fuel + generation platform, which is structurally more defensible.

If they can’t, they’re just another customer in a tight supply chain.

The Overlooked Piece: Isotopes

Buried in the filings is a smaller but near-term business:

Isotope production.

With the Atomic Alchemy acquisition and NRC licensing for their Idaho facility, this becomes:

• an early revenue stream

• a technical capability builder

• a way to monetize nuclear-adjacent operations before power generation

This won’t justify the valuation.

But it does matter for timeline credibility.

Financial Reality

The company reported:

• ~$139M operating loss (expected for this stage)

• roughly $2.5B total capital available after recent raises

That’s actually strong relative to peers.

But nuclear math is brutal.

Even with that capital:

• first-of-a-kind reactor costs will be high

• project financing will be required

• dilution or debt is almost inevitable over time

This is not a self-funded path to scale.

The Business Model Is Not What Most People Think

Oklo is not trying to sell reactors.

They want to:

• build

• own

• operate

• sell power under long-term contracts

This makes them closer to:

• a merchant power developer

• with embedded nuclear technology

That model can work.

But it also means:

• execution risk is higher

• capital requirements are larger

• timelines matter more

What Actually Improved

Across the materials, four things clearly progressed:

1. Physical execution started (site work, procurement, vendor contracts)

2. Regulatory pathway advanced (multiple approvals cleared)

3. Demand interest validated (hyperscaler engagement is real)

4. Fuel strategy matured (not solved, but actively being built)

This is meaningful.

They are no longer a “maybe someday” company.

What Did Not Change

And this is where most people lose discipline.

• No meaningful power revenue until ~2028–2030

• First-of-a-kind project risk remains high

• Fuel supply is still a system-level constraint

• The valuation still assumes success across multiple uncertain layers

So while the company improved, the risk profile did not fundamentally compress.

How This Fits in the Freedom Grid Framework

This is clearly a Phase 3B asset:

• long-duration

• infrastructure-heavy

• dependent on system-wide shifts (AI load, grid constraints, policy support)

The stock is effectively a bet on three stacked assumptions:

• hyperscalers adopt nuclear for baseload

• SMRs successfully deploy at scale

• Oklo executes ahead of competitors

If any one of those breaks, the narrative weakens quickly.

Where This Likely Trades Over Time

This is not a near-term earnings story.

It will trade on:

• milestone credibility (permits, construction progress, contracts)

• narrative momentum (AI + power demand)

• capital access

Expect:

• volatility tied to news flow

• re-rating on major project milestones

• drawdowns when timelines slip or contracts fail to firm up

In other words:

This behaves more like a long-duration infrastructure option than a traditional operating company.

The Two Questions That Matter Going Forward

Everything else is noise compared to these:

1. Are the power agreements becoming real PPAs?
Not frameworks. Not MOUs. Actual contracted revenue.

2. What is the true cost per MW of Aurora?

Because if the economics don’t compete with:

• natural gas

• geothermal

• or even hybrid solar + storage

then none of this scales the way the market expects.

Bottom Line

Oklo made real progress.

It crossed the line from:

idea → execution

That’s not trivial.

But it is still:

• capital intensive

• time delayed

• and dependent on multiple external systems aligning

So nothing about this story has become simple.

It just became real enough to start failing or succeeding in public.

Which is where the useful information finally begins.

Share

Subscribe now

My wife sent me a news story this morning.

We live about thirty minutes from Social Circle, Georgia, so this thing has been bouncing around local news for a while. The federal government wants to turn a warehouse there into a massive ICE detention facility. Residents are furious. Nobody wants something like that dropped into their backyard.

But what caught my attention wasn’t the politics.

It was how the town shut the whole thing down.

They didn’t file lawsuits. They didn’t launch a protest movement.

They turned off the water.

Literally.

The city manager walked out to the building and put a physical lock on the municipal water meter serving the warehouse. Water off. End of story.

And the reason they did it is almost funny in its simplicity.

The math didn’t work.

The Infrastructure Math That Killed the Project

Social Circle has about 5,000 residents.

Their wastewater treatment system currently handles around 660,000 gallons per day.

The proposed ICE facility would generate roughly 1,000,000 gallons of sewage per day.

Let that sink in for a second.

One building would produce more sewage than the entire town already does.

So the city told the federal government something that sounds ridiculous until you understand how infrastructure actually works.

You can’t build that here.

The pipes can’t handle it.

The treatment plant can’t handle it.

And until you solve that problem, the water stays off.

You can raise billions of dollars. You can have federal backing. You can have lawyers and consultants and a stack of permits.

But if the town sewage plant is the size of a Taco Bell bathroom, the project isn’t happening.

What This Has to Do With AI

Now here’s where the story gets interesting.

Most investors thinking about artificial intelligence follow a pretty simple chain.

AI is exploding.
AI needs GPUs.
Buy chip companies.

That’s first-level thinking.

Second-level thinking goes a little deeper.

Those GPUs consume an absurd amount of electricity.

So the constraint must be power generation and the electrical grid.

That’s already becoming obvious. Utilities are scrambling to add generation capacity. Gas turbines have multi-year backlogs. Grid interconnection queues are jammed with projects waiting to connect.

But the system doesn’t stop there.

Electricity turns into heat.

Heat requires cooling.

Cooling requires water.

And that’s where the infrastructure math starts to look a lot like Social Circle.

The Hidden Water Problem Behind AI

Modern AI data centers are essentially giant heat engines.

A single hyperscale facility can require 100 to 500 megawatts of electricity, roughly the power demand of a small city.

All of that energy eventually becomes heat.

And the most common way to remove that heat is evaporative cooling, which uses water to carry the heat away.

The numbers are staggering.

A 100-megawatt data center can require 500,000 to 2.5 million gallons of water per day, depending on climate and cooling design.

That’s roughly the daily water usage of 10,000 to 20,000 people.

Now consider that many of the newest AI campuses are three to five times that size.

Meta has talked about building 10-gigawatt AI clusters over time. For context, ten gigawatts is roughly the electricity demand of New York City.

Imagine trying to drop something with the power demand of New York City into the desert somewhere and asking the local water system to handle the cooling.

You start to see the problem.

By some estimates, U.S. data centers could consume 700 million to 1.4 billion gallons of water per day by 2030.

That’s comparable to the entire daily water supply of New York City.

And just like in Social Circle, that water eventually becomes wastewater.

Which means it has to go through pipes, pumps, and treatment plants that were never designed for that kind of load.

The Real Veto Power

This is where infrastructure becomes political in a very practical way.

Local governments control the systems that make these projects possible.

Water supply.

Wastewater treatment.

Permitting.

Zoning.

Land use approvals.

When those systems hit their limits, the project stops.

It doesn’t matter how much money is involved or how strategic the project is.

The infrastructure veto wins.

And we’re already seeing early versions of this around data centers.

Projects in Arizona have run into water restrictions.

Communities in Virginia are pushing for water-use disclosures as aquifers come under pressure.

Parts of Texas are beginning to scrutinize large cooling systems in drought-prone regions.

In Ireland, the grid operator has effectively limited new data centers around Dublin because the power system simply can’t support them.

Across the United States, nearly 100 data center projects are facing organized local opposition, and roughly $60 billion worth of developments have already been delayed or blocked.

Water is one of the biggest concerns.

Why Investors Aren’t Talking About This Yet

None of this is on CNBC.

Jim Cramer isn’t yelling about wastewater treatment plants.

Bloomberg hasn’t turned it into a daily headline.

And that’s exactly why it’s interesting.

The market is finally starting to recognize the energy constraints behind AI infrastructure. Power generation and grid expansion are now part of the conversation.

But water sits one layer further down the stack.

It’s boring.

It’s local.

It’s controlled by municipalities and utilities that most investors have never heard of.

Which means it’s early.

Where This Shows Up in the Market

Freedom Grid isn’t a stock-picking newsletter, and this series isn’t meant to be investment advice. The goal is to understand how the infrastructure behind the AI economy actually works.

But if water becomes a real constraint in the AI buildout, certain parts of the infrastructure ecosystem become more important.

Some of the companies exposed to this theme are not flashy technology firms at all. They’re the kind of steady industrial and water-infrastructure businesses that rarely get mentioned in AI conversations.

Ecolab, for example, is one of the largest global companies focused on water treatment, industrial water management, and efficiency systems. They work with large industrial facilities, utilities, and data centers to reduce water usage, recycle water streams, and manage cooling systems. It’s a mature, revenue-generating business that sits quietly underneath many large infrastructure systems.

Another example is Energy Recovery (ERII). ERII specializes in pressure-exchange technology used primarily in desalination plants, where seawater is converted into freshwater. Their systems dramatically reduce the energy required for desalination, which is one reason they are widely used in large desalination facilities around the world.

That technology has obvious implications if water scarcity becomes a constraint in data center regions. Desalination can expand freshwater supply in coastal regions, and pressure-exchange systems can also be used in certain industrial and wastewater processes where high-pressure fluid systems are involved.

Both companies are real businesses with real revenue streams today. They’re not speculative startups chasing the latest AI headline.

And interestingly, many of these water-related infrastructure companies haven’t moved much at all during the recent AI investment frenzy.

Meanwhile everything above them in the stack, from chips to construction, has been running hard.

When the top of a system is booming and the physical infrastructure underneath it isn’t moving yet, that usually means one thing.

The market hasn’t connected the dots.

The Freedom Grid View

The Freedom Grid framework starts with a simple idea.

The digital economy runs on physical infrastructure.

Every layer of that infrastructure has constraints.

Power generation.

Transmission.

Cooling.

Water supply.

Wastewater treatment.

Construction capacity.

Most people stop thinking after the electricity layer.

But the deeper you follow the chain, the more interesting the bottlenecks become.

Sometimes those bottlenecks show up in places nobody expects.

Like a small town in Georgia.

Where a city manager walked up to a warehouse, locked a water meter, and effectively stopped a federal project with a wrench and a padlock.

The Bottom Line

The AI boom may move at the speed of software.

But the infrastructure that powers it moves at the speed of pipes, pumps, permits, and construction crews.

And when those systems hit their limits, the solution is rarely complicated.

Sometimes someone just turns off the water.

Share

THE WEEK

The infrastructure thesis strengthened this week, even as the near-term macro backdrop got uglier. That is the whole week in one sentence. Markets spent the week reacting to oil volatility, sticky inflation, and a more hostile rate environment, while the physical system kept sending the same message it has been sending for months: electricity demand is outrunning infrastructure capacity, and the buildout required to fix that is still nowhere close to ready.

That tension matters more than the tape. Stocks sold off, yields rose, and the usual crowd acted like higher rates somehow repeal physics. They do not. They change who gets funded, on what terms, and how fast. They do not change the need for more generation, more cooling, more water handling, more grid throughput, and eventually more firm baseload. The infrastructure problem did not weaken this week. It became more selective.

The real shift was not just in markets. It was in narrative. The week started with the familiar framing of an artificial intelligence power boom and ended with something sharper: an artificial intelligence energy crisis. That is not semantics. A boom story excites investors. A crisis story pulls in regulators, utilities, governors, federal agencies, and eventually subsidy mechanisms. Once the story moves from opportunity to bottleneck, capital allocation and public policy start changing around it.

That is why the week was more constructive for the long thesis than the market action implied. The macro environment got less forgiving. Credit got pickier. Energy geopolitics got more dangerous. But those pressures are not random headwinds to the Freedom Grid thesis. They are accelerants for the exact parts of the system that matter most: domestic generation, contracted infrastructure, grid upgrades, nuclear fuel chain development, and anything else that reduces dependence on fragile global energy flows.

The weekly conclusion is blunt. The thesis did not get cleaner. It got more real.

THE WEEKLY VERDICT

The verdict is that the thesis is intact and strengthening, but the risk profile shifted materially toward bifurcation. In plain English, the winners this cycle are increasingly likely to be the companies and asset classes with contracts, balance-sheet strength, regulated returns, and government alignment. The losers are the stories that still require generous capital markets, forgiving investors, or faith-based timelines.

Three data points carried that verdict. First, core Personal Consumption Expenditures inflation, the Federal Reserve’s preferred inflation gauge, held at 3.1% year over year. That keeps the disinflation stall alive. Second, the 10-year Treasury yield pushed up toward 4.24% while the curve bear-steepened, which is a fancy way of saying long-end financing conditions got worse, not better. Third, Brent crude violently moved from roughly $93 to $118 before settling near $100 on Iran escalation and Strait of Hormuz fears. That combination is a damn nuisance for macro tourists, but it is exactly the kind of setup that makes energy security spending politically durable.

This is where the week stopped being a generic market story and became an infrastructure story. If inflation stays sticky, rate cuts get pushed out. If oil stays volatile, policymakers start caring less about elegant transition narratives and more about reliability, domestic capacity, and anything that can be built inside the fence line. If capital gets selective, speculative energy tech gets starved while contracted infrastructure gets funded first. Same environment, very different consequences depending on where you sit in the stack.

The other reason the verdict is constructive is the recognition-versus-implementation gap. Recognition is racing. Implementation is crawling. Power generation is now widely understood as constrained, with recognition near 82%, while implementation is only around 23%. Cooling is being recognized faster, but actual deployment remains early. Water is still badly under-recognized relative to its eventual gating role. Nuclear has regained narrative momentum while the actual fuel-cycle and fabrication stack remain almost comically underbuilt. That widening gap is not a sign the thesis is late. It is evidence the market is still narrating a future the physical world has not yet delivered.

That does not mean all risk goes away. It means the shape of the opportunity is getting more specific.

THE ARC

The week’s arc was a progression from macro discomfort to structural confirmation.

At the start of the week, the market was already uneasy. Inflation had stopped improving fast enough to give the Federal Reserve breathing room, and rates were leaning the wrong way for long-duration projects. On its own, that would have been enough to keep speculative capital on edge. But then energy geopolitics added fuel to the fire. The Iran situation escalated, markets started gaming out some probability of Strait of Hormuz disruption, and oil responded the way oil always does when the world is reminded that concentrated chokepoints exist for a reason. First panic, then policy chatter, then partial retracement, but not a full return to calm.

That mattered because energy shocks hit the infrastructure story in two directions at once. In the short run, they raise inflation pressure and tighten financial conditions. In the longer run, they make energy security non-optional. That second effect is the more important one. The world has seen this movie before. The 1970s oil shocks helped push nuclear expansion. The gas crises of the 2000s accelerated liquefied natural gas buildout. This week’s version looks like the opening phase of an electricity-security cycle shaped by artificial intelligence, domestic industrial policy, and the simple fact that always-on compute does not care whether the grid planner had a nice spreadsheet.

By midweek, the capital message also became clearer. Money is still available, but it is discriminating. Credit spreads are not yet blowing out in full crisis mode, but they are widening. High-yield option-adjusted spreads around 3.08% are not catastrophic, but the direction matters because infrastructure does not need a formal credit event to feel pain. It only needs the cost of capital to rise enough that shaky projects fall below hurdle rates. That shifts the advantage toward regulated utilities, backed projects, contract-heavy industrial names, and nuclear or grid infrastructure with political support. It shifts the disadvantage toward unproven hydrogen dreams, reactor vaporware, and anything that still needs the market to hand it cheap money and applause at the same time.

By the end of the week, the narrative layer caught up. The newsletter ecosystem moved from “AI energy boom” toward “AI energy crisis.” That is a critical transition. A boom narrative attracts capital and headlines. A crisis narrative attracts institutions. It forces the question that is now becoming unavoidable across states and utility territories: who pays for the grid needed to support hyperscaler load growth? That political fight is the next major bottleneck. Not whether more power is needed. That argument is basically over. The fight is who eats the cost.

And that is where the week ended: with stronger proof that the demand side is real, stronger proof that the physical response is lagging, and stronger proof that the next phase of the battle is political and financial rather than conceptual.

MACRO & STRUCTURAL POSITION

Macro Regime: Mid-fear transition

Marks Cycle Position: Between normal and early fear

Change vs Prior Week: Shifted toward tighter selectivity, not full panic

Super-Cycle Position:

Phase 1 Gap: 59 percentage points, with recognition at 82% and implementation at 23%

Phase 2 Gap: 39 percentage points, with recognition at 58% and implementation at 19%

Phase 3 Gap: 24 percentage points, with recognition at 32% and implementation at 8%

Phase 3b Gap: 43 percentage points, with recognition at 48% and implementation at 5%

The macro read is ugly enough to matter and not ugly enough to break the thesis. That is the sweetly irritating middle ground markets hate. Rates are high enough to stress project economics. Inflation is sticky enough to delay easy monetary relief. Oil is volatile enough to reintroduce energy insecurity into policy thinking. But credit is still functional, which means the system is not frozen. It is sorting.

That sorting is the point. This is not a regime where anything related to “energy transition” or “AI power” gets funded. This is a regime where the world starts differentiating between projects that can survive real financing conditions and projects that were only ever alive because money had no standards. It is a healthier environment for serious infrastructure and a worse one for PowerPoint fraud dressed up as industrial strategy.

The structural gaps across all four phases remain the load-bearing fact. Recognition is ahead of implementation everywhere, especially in power and the nuclear supply chain. That means the thesis runway is extending, not closing. Markets can understand the problem much faster than steel, turbines, cooling loops, pipelines, enrichment facilities, transformers, and transmission corridors can be built. Humans are always shocked to learn that reality has lead times.

INFRASTRUCTURE STACK

Phase 1 — Power Generation

Power remains the core bottleneck. Utilities are reporting multi-gigawatt connection requests, and hyperscaler load forecasts continue to surprise to the upside. The system is now well past the stage where this can be dismissed as a few overexcited technology companies leasing too much capacity. The problem is systemic. The market increasingly understands that. The buildout does not.

That mismatch is why Phase 1 remains the center of gravity. Gas turbines still carry multi-year lead times. Transmission remains stuck in multi-year engineering and permitting bottlenecks. Nuclear is gaining narrative acceptance, but commercial-scale relief is still far out on the calendar. The result is simple: scarcity persists.

Phase 2 — Cooling Infrastructure

Cooling kept advancing from secondary issue toward recognized constraint. Artificial intelligence compute clusters are not just electricity loads. They are heat engines. Liquid cooling, immersion systems, district cooling, and site-level thermal design are moving from technical detail to investment relevance. Recognition is improving, but implementation is still in the teens. That means this phase remains early.

The key point here is that cooling is not optional optimization. It is operational viability. More power density means more thermal rejection. If facilities cannot manage heat, the load growth story runs into a wall even if power can be procured.

Phase 3 — Water Infrastructure

Water remains the least appreciated major constraint in the stack. Recognition is still only around one-third while implementation remains near zero in practical terms. That is absurdly low relative to how central water is becoming in high-load computing environments, especially in Arizona, Nevada, Texas, and Northern Virginia.

This is still the hidden problem. Municipal capacity, water rights, reuse mandates, and wastewater handling move on regulatory and physical timelines, not quarterly earnings timelines. That makes water one of the most durable bottlenecks in the whole thesis and one of the easiest for the market to keep underpricing until it becomes an actual local political fight.

Phase 3b — Nuclear Supply Chain

Nuclear made narrative progress this week. The market increasingly understands that dispatchable, carbon-free baseload has regained strategic relevance. The problem is that the fuel and fabrication chain still looks embryonic relative to where it would need to be for a true buildout cycle. Enrichment, High-Assay Low-Enriched Uranium supply, and fuel fabrication remain chokepoints.

So yes, nuclear strengthened as a strategic direction. No, the actual system underneath it is not remotely ready. That is both the opportunity and the warning.

Cross-Phase — Grid and EPC

The grid payment fight is moving to center stage. The question is no longer whether upgrades are needed. It is who pays. Utilities want rate-basing. Hyperscalers want access. States want jobs without voter revolts over electric bills. That conflict is now one of the most important variables in the entire stack.

Regions are diverging. Texas and the Southeast remain relatively more constructive. Virginia, California, and New York are carrying the heaviest friction. That matters because political velocity is now as important as engineering capacity.

THE NARRATIVE LANDSCAPE

The most important narrative convergence this week was around electricity scarcity. Multiple voices across the ecosystem are landing in the same place: artificial intelligence demand is stressing physical infrastructure faster than the system can respond. That convergence matters because it is not just coming from bullish equity framing. It is showing up through trade press, utility commentary, and macro discussion.

Where narratives diverged most sharply was geopolitics. Much of the media framing still leaned toward stabilization after the initial oil spike, but the underlying evidence pointed to escalation risk staying materially alive. Iranian discussion of Hormuz closure, U.S. posture changes, and the International Energy Agency’s emergency release stance do not scream “all clear.” They scream “still fragile, don’t be stupid.”

The other place the newsletter ecosystem added value was in detecting the tone shift from boom to crisis. That is a leading indicator because markets often do not notice when a sector narrative crosses the line from growth story to public infrastructure problem. Once that line is crossed, the relevant audience changes. The audience is no longer only investors and management teams. It becomes regulators, utility commissions, statehouses, and federal agencies. Those people move slower, but when they move, they alter the map.

The weakening narrative this week was the fantasy that all of this can be solved with generic enthusiasm and a few procurement announcements. The ecosystem is starting to understand what this system has been built around from the start: recognition is not implementation, and implementation is where the money, pain, and time go.

THE EDGE

The first genuinely early signal is the political migration of the bottleneck from climate language to cost-allocation language. That sounds boring, which is exactly why it matters. “Who pays for the grid” is not a media slogan. It is the next real gate. Once that fight becomes explicit, it will drive permitting speed, data center siting, rate-case conflict, and the attractiveness of on-site generation.

The second early signal is the underappreciated role of water as the hidden constraint on artificial intelligence infrastructure geography. Power scarcity gets headlines. Water stress gets local resistance, municipal complexity, and long-duration regulatory friction. The market is still too early here, which usually means the repricing comes later and faster.

The third is the distinction between nuclear narrative recovery and nuclear supply-chain readiness. The market is increasingly comfortable with the first and still not serious enough about the second. That gap is where future winners and embarrassing disappointments will separate.

THESIS SCORECARD

STRENGTHENED:

Domestic power infrastructure, grid upgrades, contracted generation, nuclear supply chain enablers, regulated utilities with visible load growth, and water-linked bottleneck exposure.

INTACT:

The broader Freedom Grid architecture, cooling infrastructure, EPC-linked buildout, transmission scarcity, and the core assumption that artificial intelligence load growth continues to outrun physical deployment.

CHALLENGED / WATCHLIST:

Speculative hydrogen, financing-dependent energy tech, unproven reactor startups, and any infrastructure-adjacent story requiring cheap capital instead of hard demand.

The pattern this week was a widening divide between quality and hope. That is not poetic. It is how tightening cycles work. If capital gets more expensive and inflation stays sticky, project economics stop forgiving weak structures. Real contracts, regulated returns, backlog visibility, and political alignment matter more. Long-duration stories without financing credibility matter less.

That pattern also maps cleanly onto the phase structure. Phase 1 looks strongest because the need is most visible and the capacity gap is undeniable. Phase 3 remains structurally powerful but underrecognized. Phase 3b has strategic strength but execution fragility. Cooling sits in the middle: increasingly obvious, still not fully priced.

Nothing in the week suggested the core thesis is nearing a break condition. Several things suggested that valuation and positioning discipline will matter more from here.

RISK REGISTER

The most credible near-term risk remains an energy inflation shock feeding back into rates and credit. If oil volatility persists or the Strait of Hormuz situation materially worsens, inflation could re-accelerate, rate cuts could be pushed even further out, and financing conditions could tighten enough to slow project starts. That would not invalidate demand. It would delay implementation.

The second major risk is infrastructure financing slowdown. This is not crisis credit yet, but it is moving in the wrong direction. Projects that made sense under easier assumptions start failing hurdle rates when debt gets pricier and duration risk goes up. That creates a filter. Healthy for long-term quality, ugly for marginal participants.

The third risk is political backlash. Power demand growth is exciting until ratepayers see the bill. If hyperscaler-driven upgrades are perceived as socialized costs, the state-level fight over grid funding gets more aggressive. That introduces regional friction, delays, and the possibility that the best data center demand stories become politically harder to convert into infrastructure approvals.

The final systemic risk is that narrative progress outruns execution reality by so much that investors start assuming a buildout speed the physical system cannot deliver. That is how violent disappointment gets created inside otherwise correct long-term themes.

NEXT WEEK’S SETUP

The setup for next week is about whether this week’s macro tightening becomes a trend or just a stress pulse. The market will be watching the same obvious variables, but for this thesis only one question really matters: does tighter macro simply shake out weak capital, or does it start impairing actual infrastructure deployment?

The key watchpoints are straightforward. First, any fresh signal around oil stability versus escalation. Second, any new evidence that the Federal Reserve’s delayed-cut setup is hardening. Third, signs from utilities, regulators, or hyperscalers that the cost-allocation fight is becoming more explicit. Fourth, any incremental signal around nuclear fuel-cycle readiness, because narrative support without supply readiness is still a trap door.

The one thing that matters most next week is whether energy security remains a macro headline or becomes a policy response story. Headlines move prices. Policy response moves the infrastructure map. If the system starts shifting from fear language to build language, that is the more important development by far.

THE BOTTOM LINE

1. The thesis strengthened — High confidence, system-wide extraction — Improved — Electricity demand continues to outrun infrastructure capacity, and nothing this week suggested the physical bottlenecks are easing.

2. Macro conditions got less forgiving — High confidence, macro and rates data — Worsened — Sticky inflation, higher long-end yields, and delayed rate-cut expectations increase financing friction and sharpen the divide between strong and weak projects.

3. Energy geopolitics became a real accelerant — High confidence, oil and policy response signals — Escalated — Oil volatility raises near-term inflation risk, but it also strengthens the long-term case for domestic generation, energy security, and firm baseload investment.

4. Recognition keeps outrunning implementation — High confidence, super-cycle gap analysis — Widened — Power, cooling, water, and nuclear are all being recognized faster than they are being built, which is exactly what an early infrastructure super-cycle looks like.

5. The political fight is shifting to grid cost allocation — Medium to high confidence, narrative and policy synthesis — Emerging — The next major bottleneck is increasingly not whether more infrastructure is needed, but who pays for it and which regions are willing to move fastest.

SOURCES

Built from the weekly extraction provided by the Freedom Grid research system, drawing on the combined system context, weekly wrap, and newsletter ecosystem sweep. The underlying system architecture and thesis framework are reflected in the consolidated Freedom Grid system document.

Disclosure: The author may hold positions in companies discussed in this briefing. This publication provides infrastructure analysis, not investment recommendations. Readers should conduct their own research before making investment decisions.

The first two articles in this series explained the basic idea behind the Freedom Grid project.

Artificial intelligence is not just a software story. It is an infrastructure expansion. Running large-scale AI systems requires electricity, cooling, water, materials, labor, financing, and regulatory approvals.

Those systems do not scale at the same speed as software.

That gap between fast-growing demand and slow-moving infrastructure is where constraints form.

The obvious question after reading the first two pieces is simple.

How do you actually track something like that?

Infrastructure cycles do not show up neatly in earnings reports or quarterly guidance. They develop slowly across energy markets, supply chains, capital markets, and regulatory systems.

Freedom Grid exists to monitor those signals continuously.

Not by predicting the future.

By watching the system itself.

⸻

Where the Project Actually Started

Freedom Grid did not start as an investment idea.

It started as an experiment in using AI to organize information.

About a year ago I began building small AI systems to track different parts of my life. Workouts. Nutrition. Medical documentation. Blood tests. Decision frameworks. Personal planning.

Instead of using AI as a simple search tool, I was running multiple threads that tracked information over time and compared results between runs.

At some point I asked a question that changed the direction of the project.

Is this how most people use AI?

The answer was no.

Most people use AI tools for short tasks. Writing emails. Searching for information. Generating content. Useful things, but relatively light.

What I had built was closer to a monitoring system. Multiple threads running simultaneously, gathering information, comparing outputs, and tracking changes.

That led to another question.

If these systems require computation, how much electricity does large scale AI actually use?

Around the same time I started seeing announcements from hyperscale technology companies building massive new data centers. One of the announcements mentioned a ten gigawatt project.

I had to look that up.

Ten gigawatts is roughly the electricity demand of New York City on a hot August day.

And companies are planning dozens of facilities in that range.

That was the moment the story changed.

Artificial intelligence was clearly going to require an enormous expansion of physical infrastructure.

The next problem was figuring out how to track it.

From Random Threads to a Monitoring Platform

The first version of the system was very simple.

One thread scanned news related to energy markets and infrastructure. Another tracked the portfolio sleeve I had allocated to the thesis. A third looked for evidence that contradicted the thesis.

Then the system started expanding.

One module began tracking industrial commodities. Another looked at infrastructure supply chains. Another monitored project financing conditions.

Each module had a specific job.

Each ran on a defined schedule.

Each fed information into the others.

Over the course of about a year the project grew into a monitoring platform consisting of roughly twenty interconnected research modules.

Some run daily.

Some run weekly.

Some run quarterly.

Each module examines a different part of the infrastructure system surrounding artificial intelligence.

Taken together they create a continuous scan of the environment in which the AI expansion is happening.

The Core Monitoring Engines

Several modules form the backbone of the Freedom Grid system.

The Morning Signal Scan runs every day. It sweeps infrastructure news, energy markets, regulatory filings, hyperscaler announcements, and utility developments. The goal is not to summarize headlines. It is to identify signals that suggest something in the infrastructure system is changing.

The Commodity Monitor tracks materials that directly influence infrastructure construction. Copper, aluminum, steel, uranium, diesel, and natural gas are all inputs for building energy systems, transmission networks, and data centers. Sudden movements in those inputs often signal pressure in supply chains.

The Credit Cycle Monitor watches financing conditions for large projects. Power plants, transmission lines, and industrial infrastructure require massive capital investment. When credit spreads widen or project financing tightens, infrastructure development can slow even when demand remains strong.

The Policy and Regulatory Monitor tracks developments across agencies such as the Federal Energy Regulatory Commission, the Department of Energy, and the Nuclear Regulatory Commission. Infrastructure projects frequently move at the speed of permitting rather than the speed of technology.

The Infrastructure Fragility system, sometimes called the Canary system, searches specifically for signals that contradict the core thesis. It looks for supply chain failures, financing problems, regulatory roadblocks, or technology shifts that could disrupt the expected expansion.

Finally, the Portfolio Control Tower governs the investment sleeve tied to the research. It does not generate stock recommendations. Its job is risk management. It ensures portfolio exposure remains consistent with the thesis and the broader macro environment.

Each module observes a different layer of the system.

Together they create a multi-angle view of the infrastructure landscape.

How the Snapshot System Works

One of the most important elements of the Freedom Grid platform is the snapshot system.

Every time a monitoring module runs, it produces a structured snapshot of the system’s current state.

Those snapshots capture signals across several domains. Energy markets. Commodity prices. Supply chain conditions. Credit markets. Regulatory developments.

Instead of treating each report as a one-time observation, the system compares the current snapshot with previous runs.

That comparison reveals how conditions are evolving over time.

Signals strengthen.

Signals weaken.

New constraints appear.

Existing constraints resolve.

Tracking those changes turns research from a collection of articles into a time series of infrastructure behavior.

The goal is not just to understand what is happening today. The goal is to see how the system is shifting over time.

What the System Is Actually Watching

Infrastructure expansion depends on far more than electricity generation.

Labor is one of the biggest constraints. Electricians, plumbers, pipefitters, HVAC technicians, and engineering procurement construction firms all represent finite capacity. When thousands of new power plants, pipelines, transmission lines, and data centers are built simultaneously, those projects compete for the same workforce.

Equipment supply chains impose another constraint. Turbines, high voltage transformers, switchgear, cooling systems, and specialized semiconductor equipment all have limited production capacity.

Fuel supply chains also matter. Some advanced nuclear reactors require high assay low enriched uranium fuel, which currently has limited production capacity and complex regulatory oversight.

Each of these elements labor, equipment, materials, financing, and policy can influence the pace of infrastructure expansion.

Freedom Grid monitors them continuously.

What Real Signals Look Like

Signals often appear before the broader market recognizes them.

Last summer several monitoring modules began flagging what looked like an off-grid power pattern developing in parts of the data center industry.

Interconnection queues were growing rapidly. Turbine manufacturing lead times were extending. Several hyperscale operators were quietly exploring on-site generation rather than waiting years for new grid capacity.

Individually those signals did not mean much.

Together they suggested something important. Demand for reliable power was beginning to outrun the speed at which the grid could expand.

That environment favors technologies capable of providing electricity outside traditional grid timelines.

One company positioned in that space was Bloom Energy. At the time the stock was trading around twenty four dollars.

The signals continued strengthening through the fall as power demand projections increased and grid bottlenecks became more visible.

Eventually the market caught up.

The point is not the individual investment.

The point is the process. Multiple independent signals across the system pointed in the same direction before the narrative became obvious.

Other signals show up in different ways.

The Commodity Monitor has repeatedly flagged movements in copper and aluminum markets that influence transmission infrastructure costs.

The Credit Cycle Monitor watches project finance conditions. When capital becomes more expensive, infrastructure buildouts slow even when demand remains high.

A single signal rarely means much.

But when signals begin appearing across several monitoring layers, the system starts to reveal where pressure is building.

When the System Proved Itself

For most of my life I invested primarily in index funds.

Earlier in my career I briefly worked as a registered broker. That experience left me skeptical about how much useful information individual investors actually receive. Much of the job involved selling products rather than analyzing systems.

Eventually I stopped trying to outguess the market.

Building the Freedom Grid system changed that.

The first confirmation came when the small portfolio sleeve tied to the thesis began significantly outperforming the broader indexes.

But the more important test came during a market drawdown last November.

The market sold off sharply for several weeks and my portfolio declined along with everything else. In the past I would have assumed something fundamental had broken.

This time the monitoring modules showed no deterioration in the infrastructure signals.

Power demand projections were still rising.

Supply chain constraints were still tightening.

Financing conditions had not changed materially.

The drawdown was macro noise.

The correct decision was to do nothing.

Within weeks the portfolio recovered and moved higher.

The system had done its job.

Why Share the Project

The infrastructure expansion required to support artificial intelligence is enormous.

No single investor can monopolize a cycle of that scale.

For most of my life I felt frustrated trying to navigate markets without access to the kind of information institutions use to make decisions. AI has changed that equation. With the right structure, individuals can now synthesize large amounts of information at very low cost.

Freedom Grid is an attempt to build that structure.

The goal is not to provide stock recommendations.

The goal is to identify where the infrastructure system is under pressure and where constraints are forming.

Those constraints shape how the entire AI expansion unfolds.

If understanding those signals helps other people navigate the cycle more effectively, that is a positive outcome.

Freedom Grid began as a personal research experiment.

Now it is becoming a public research platform designed to monitor the infrastructure behind artificial intelligence.

And the infrastructure cycle it is tracking is only beginning.

Everyone thinks the AI revolution is about chips.

NVIDIA. GPUs. Model size. Training runs.

That is where the headlines live.

But after spending the better part of a year digging into the infrastructure behind AI, something else becomes obvious very quickly.

The real constraint is not chips.

It is power.

Not theoretical power. Actual electricity. The kind that has to move through transformers, transmission lines, substations, cooling systems, and water infrastructure before a single GPU ever turns on.

Once you start looking at AI through that lens, the entire story changes.

You begin to notice signals that most technology coverage ignores. Microsoft sitting on tens of billions of dollars of demand it cannot fulfill because the power infrastructure is not ready. Data center projects delayed not because of software problems but because they cannot secure transformers. Electrical equipment with lead times measured in years instead of months.

Suddenly the AI revolution stops looking like a Silicon Valley story.

It starts looking like an industrial infrastructure buildout.

That realization is what led me to build the research project that eventually became Freedom Grid.

But noticing that AI needs power is only the starting point. The more interesting question is what happens when the world suddenly needs far more infrastructure than it currently has.

Infrastructure does not scale the way software does. It moves slowly. It moves in phases. And markets consistently underestimate how those phases unfold.

That is the core idea behind the Freedom Grid thesis.

Infrastructure moves on a completely different clock

One of the first things that becomes obvious when you start studying energy infrastructure is how long everything takes.

Power plants do not appear overnight.

Transmission lines are not approved in a week.

Municipal water systems do not expand because a technology company announced a new product.

These systems move on timelines measured in years.

A typical power project can take five to seven years to build. Transmission infrastructure can take even longer once permitting battles begin. Nuclear plants can take decades from concept to operation. Cooling equipment for large data centers can carry lead times of months or years depending on the system.

Now compare that with AI.

Model demand is exploding. Hyperscalers are racing to build new data centers. Compute clusters grow larger every quarter.

Demand is moving at the speed of software.

Supply is moving at the speed of infrastructure.

That gap creates bottlenecks.

And bottlenecks reshape markets.

The scale of the problem is already visible

One of the reasons this topic became impossible to ignore is the amount of capital now flowing into AI infrastructure.

Hyperscalers are planning roughly $660 to $690 billion in capital spending in 2026 alone.

That includes roughly:

Amazon around $200 billion.

Alphabet roughly $175 to $185 billion.

Meta around $115 to $135 billion.

Microsoft over $120 billion.

Those numbers are almost hard to process.

But the more interesting detail buried in earnings calls is the constraint.

Microsoft recently disclosed roughly $80 billion in Azure demand it cannot currently fulfill because the data centers do not have enough electricity. GPUs are installed but cannot run because the power infrastructure is not ready.

At the same time, key components needed to expand the grid are already constrained.

Large power transformers, which are required for grid expansion and data center interconnections, currently have lead times of roughly 128 weeks.

That is nearly two and a half years.

Which means some of the largest technology companies in the world are effectively standing in line waiting for electrical equipment.

If that sounds ridiculous, welcome to infrastructure.

The three phases of the Freedom Grid cycle

As I started mapping the infrastructure required to support AI, a pattern began to emerge.

The system tends to move through three constraint phases.

Each phase leads naturally to the next.

Phase 1: Power

The first bottleneck is electricity.

AI compute requires enormous amounts of energy. Data centers are becoming some of the largest single industrial power consumers ever built.

The grid was not designed for this kind of demand growth.

Interconnection queues are clogged. Gas turbines are backlogged. Transmission projects face years of permitting delays. Transformer supply is tight.

This creates the first infrastructure constraint.

Power scarcity.

When electricity becomes scarce, companies start looking for alternatives.

Behind the meter generation, fuel cells, distributed power systems, and fast build natural gas plants are all receiving renewed attention because they allow companies to bypass the grid entirely.

When the grid cannot move fast enough, people go around it.

Phase 2: Cooling

Power turns into heat.

Every watt consumed by a data center eventually becomes thermal energy that must be removed.

As AI chips become more powerful, server racks are drawing far more power than older facilities were designed to handle.

Air cooling worked for decades. It is beginning to reach its limits.

Liquid cooling systems, advanced heat exchangers, and specialized thermal infrastructure are becoming critical components of next generation AI data centers.

These systems also have manufacturing constraints, supply chains, and installation timelines.

Cooling is not just an engineering detail.

It becomes the next bottleneck in the system.

Phase 3: Water

Cooling systems ultimately need somewhere to send heat.

Often that means water.

Municipal water capacity, water rights, environmental regulation, and wastewater infrastructure all become part of the conversation.

These systems operate on extremely long timelines. Permitting and building large water infrastructure projects can take ten to fifteen years.

Which means water can become one of the slowest constraints in the entire chain.

Nuclear as the long term anchor

There is also a parallel infrastructure path running through all of this.

Baseload power.

AI infrastructure runs continuously. Intermittent energy sources alone cannot support that demand without massive storage capacity.

Dispatchable power becomes increasingly valuable.

Advanced nuclear reactors and small modular reactors are often proposed as long term solutions, but nuclear operates on the longest development timelines in the entire system. Fuel supply chains such as HALEU enrichment also remain constrained.

Which means nuclear may ultimately become part of the solution, but it will take time.

The Freedom Grid research platform

This is where the project itself comes in.

Freedom Grid is not just a thesis. It is a research system designed to track the signals that determine whether the thesis is working or breaking.

Over the past year I built a structured monitoring framework that tracks infrastructure indicators across energy markets, supply chains, credit conditions, and policy developments.

Some modules monitor credit conditions because infrastructure projects depend heavily on financing. Others track supply chain constraints such as turbines, transformers, electrical switchgear, and construction labor. Others monitor regulatory policy, permitting timelines, and nuclear licensing progress.

There is also a daily signal layer that tracks major developments across energy markets, AI infrastructure, and capital flows.

The goal is not to predict the future.

The goal is to watch the stress points.

Infrastructure systems tend to fail at their weakest link.

Freedom Grid is designed to identify those links as they emerge.

Bottlenecks are rarely isolated

One of the more interesting things that happens once you track these constraints for a while is that they begin interacting with each other.

A delay in one part of the system often creates opportunity somewhere else.

If transformer shortages slow grid expansion, companies that can generate power directly on site suddenly become more valuable.

If cooling equipment becomes scarce, companies producing thermal systems see demand surge.

If water constraints slow data center construction in one region, development shifts to another.

The system constantly rebalances itself around whatever the current constraint happens to be.

That is why Freedom Grid focuses on bottlenecks instead of predictions.

Predictions are fragile.

Constraints are observable.

The investing lens

Freedom Grid is not a stock picking newsletter.

I am not going to tell anyone what to buy or sell.

The goal is simpler than that.

This project is about mapping the infrastructure system around AI and identifying where pressure is building.

When you understand where the bottlenecks exist, the economic implications tend to become clearer.

If hyperscalers are spending hundreds of billions building data centers but cannot get electricity, the companies that solve the electricity problem become important.

If cooling becomes the next constraint, the companies building cooling infrastructure become important.

If water becomes the bottleneck, the same logic applies.

The goal here is simply to help people see the system more clearly.

The real limit to AI

The AI revolution will produce extraordinary software.

But software alone cannot run data centers.

Servers need electricity.

Electricity produces heat.

Heat must be removed.

Cooling systems require water.

Baseload power requires reliable generation.

The deeper you look into AI infrastructure, the clearer it becomes that the limiting factor may not be algorithms at all.

It may simply be how fast the physical infrastructure of the energy system can expand.

That is what Freedom Grid exists to study.

The goal is to identify where the system is under pressure and where those pressure points begin to reshape markets.

For me, the original motivation behind this project was simple. I was trying to map a path to financial freedom by understanding where the next major infrastructure cycle was forming.

The name Freedom Grid actually came from that idea. It started as a slightly goofy working title for a personal research project.

It stuck.

Because the reality is that this infrastructure cycle is enormous. The scale of the buildout required to support the AI economy will unfold over the next decade or more.

There will be winners.

There will be mistakes.

There will be plenty of noise along the way.

But if we can track the constraints clearly enough, the structure of the system begins to make sense.

And this cycle is big enough that more than one person can benefit from understanding it.

If the work here helps people see the system a little more clearly along the way, even better.

One quiet observation after reading the fully cleaned version:

This reads very much like a real Substack launch piece now, not an “AI wrote this” article. The tone lands somewhere between analyst and curious builder, which is exactly where the credibility lives.

And for what it’s worth, the “slightly goofy name that stuck” part is one of the most human sentences in the entire thing. Keep that. Readers remember that kind of detail.

Subscribe now

Everyone thinks the AI revolution is about software.

It isn’t.

The real constraint is the electrical grid.

For the past two years the conversation around artificial intelligence has been dominated by chips. Nvidia earnings calls. GPU shortages. Massive capital spending by Microsoft, Amazon, Google, and Meta as they race to build the computing backbone of the AI economy.

Now something interesting is happening.

The conversation is starting to shift.

People are beginning to realize that the real bottleneck may not be the chips at all. It may be the physical infrastructure required to run them.

Power.

If you have been following this topic over the past year, that shift has been impossible to miss. Grid operators are warning about massive electricity demand from new data centers. Utilities are revising load forecasts upward for the first time in decades. Hyperscalers are scrambling to secure power capacity before they build the next generation of AI infrastructure.

The electricity problem is finally becoming mainstream.

But that is only the first layer of the story.

If you keep following the chain outward, the system gets more complicated very quickly.

Electricity leads to cooling. Cooling leads to water. Power plants lead to transformers, transmission lines, pipelines, and construction timelines measured in years.

Once you see the entire system, the AI revolution stops looking like a software story and starts looking like an infrastructure story.

The Question That Started This

This whole rabbit hole started because I was using AI constantly.

Over the past year I have been experimenting with tools like ChatGPT to build what I jokingly call a personal operating system. Health tracking, research workflows, planning systems, decision logs. Basically trying to see how far these tools could go if you used them heavily and deliberately.

At some point curiosity kicked in and I asked a simple question.

How much compute power does this kind of usage actually require?

The answer surprised me.

Based on the frequency and structure of my prompts, my usage pattern was estimated to require roughly 50 to 100 times more compute than a typical casual user.

That did not surprise me from a computing perspective. Data centers scale. Software scales.

But it triggered a different question.

How much electricity does that require?

Because GPUs do not run on vibes. They run on power. And most of that power becomes heat that has to be removed with cooling systems the size of small industrial plants.

Once you start thinking about AI in terms of electricity instead of software, the scale of the system changes instantly.

That is when the story stops being about algorithms and starts being about infrastructure.

When the Numbers Got Ridiculous

The moment this really clicked for me was when reports surfaced that Meta was exploring a 10 gigawatt AI data center campus.

Ten gigawatts sounds impressive but abstract. I had no idea what that meant in practical terms. So I looked it up.

Ten gigawatts is roughly the electricity demand of New York City on the hottest day in August.

One data center complex consuming the same power as one of the largest cities in the United States.

And Meta is not the only company thinking at that scale. Hyperscalers across the industry are planning enormous AI infrastructure expansions measured in gigawatts.

At that point the story stops looking like technology.

It starts looking like heavy industry.

The Chips Are Already Bought

One of the most important things to understand about this moment is that the capital spending has already started.

Microsoft, Amazon, Google, and Meta have poured tens of billions of dollars into AI infrastructure over the past two years. A huge portion of that spending went directly to Nvidia for GPUs.

Those chips need somewhere to run. That means massive data centers.

But data centers are useless without power.

And this is where the system starts to break.

There are already signs that power availability is becoming a real constraint. Some hyperscalers have data center capacity waiting on electricity connections. Utilities across multiple regions are warning that projected load growth from AI is far exceeding previous forecasts.

Imagine spending billions on the most advanced computing hardware ever built and discovering that the bottleneck might be the local power grid.

That is where things start getting interesting.

Infrastructure Runs on a Different Clock

Software companies move fast.

Infrastructure does not.

Building a new gas power plant typically takes five to seven years once permitting, financing, engineering, and construction are included.

Transmission lines can take seven to twelve years in the United States because of environmental reviews, regulatory approvals, and local opposition.

Large power transformers now have lead times approaching 128 weeks, roughly two and a half years.

Gas turbines are even worse. Manufacturers such as GE Vernova are discussing delivery timelines that stretch into 2029 and 2030 because of massive order backlogs.

But equipment is only part of the problem.

Infrastructure projects also depend on skilled labor. Electricians, plumbers, HVAC specialists, pipefitters, welders, and construction crews capable of building complex industrial systems. The supply of those workers is limited, and the same companies are competing for them across multiple sectors at once.

Engineering, procurement, and construction firms are another bottleneck. There are only so many EPC contractors capable of managing large scale infrastructure projects. The same firms that might build a gas power plant could also be building pipelines, transmission systems, LNG facilities, desalination plants, or large industrial cooling systems.

Those projects compete for the same expertise.

Which means even if capital is available, the number of projects that can be built at the same time is constrained by people and execution capacity.

Hyperscalers Are Not Waiting

The largest technology companies in the world are not going to sit around politely while infrastructure catches up.

Artificial intelligence is the next platform shift in computing. Whoever builds the most capable AI infrastructure over the next decade could dominate the next generation of the technology economy.

Which means the hyperscalers are racing.

And when trillion dollar companies are racing, they will pay almost anything to remove bottlenecks.

If connecting to the grid takes too long, they will look for alternatives.

That is where off grid generation enters the conversation.

Instead of waiting years for grid capacity, a data center can produce electricity on site using behind the meter power generation.

Technologies that looked niche a few years ago suddenly become very interesting in that context. Fuel cells. Advanced geothermal. Small modular nuclear reactors. Anything capable of delivering reliable power directly to a facility becomes valuable if the alternative is waiting a decade for transmission upgrades.

The Second and Third Layers

Power is the first layer of the infrastructure story.

Cooling is the second.

Modern AI chips consume enormous amounts of electricity and convert most of that energy into heat. Removing that heat is becoming one of the largest engineering challenges in modern data centers.

Follow the chain one step further and you reach the third layer.

Water.

Many cooling systems depend heavily on water resources. In regions already dealing with water constraints, that creates another potential bottleneck that few people in the technology world have spent much time thinking about.

In some regions the problem becomes even more complex. New power plants require water. Cooling systems require water. Growing industrial demand can lead to new desalination projects or water recycling infrastructure.

And those projects compete for the same engineering talent, the same construction firms, and the same industrial supply chains.

Another Constraint Few People Talk About

There is another constraint quietly sitting behind the entire energy transition.

Nuclear fuel.

Small modular reactors are often discussed as a long term solution for reliable, carbon free power generation. But even that path has its own supply chain bottlenecks.

Advanced reactors require specialized fuels such as HALEU, which stands for high assay low enriched uranium. The current global supply chain for that fuel is extremely limited.

Which means that even technologies designed to solve the power problem may run into constraints of their own.

Once you start tracing these systems outward, you realize the infrastructure story is not just complicated. It is layered with dependencies that stretch across entire industrial sectors.

Why Bottlenecks Matter

This publication is not going to recommend specific stocks.

There are already plenty of newsletters trying to do that.

The goal here is different.

Freedom Grid is going to follow the bottlenecks.

Because bottlenecks tell you where the system is under stress.

Right now the AI infrastructure system has several obvious pressure points.

Power generation.

Transformers.

Transmission capacity.

Cooling systems.

Water infrastructure.

Skilled labor.

Nuclear fuel.

Each of those constraints has companies sitting directly at the chokepoint.

Companies that can deliver power faster. Companies that can avoid grid interconnection queues. Companies that can cool increasingly power dense AI chips. Companies that can build the infrastructure the hyperscalers suddenly need.

When the world runs into hard physical constraints, the companies solving those constraints tend to become extremely valuable.

Not because someone recommended them online.

Because physics forced the issue.

The Name Freedom Grid

The name Freedom Grid actually started as a joke.

When I first began digging into this topic, the original goal was simple. I was trying to build an investment research project that could help me reach financial freedom.

While setting up the research system I needed a placeholder name for the project and typed in “Freedom Grid.”

It stuck.

The more I studied the infrastructure behind the AI economy, the more appropriate the name felt. This entire technological shift ultimately runs on the power grid. Every model, every data center, every AI application depends on physical energy systems that most people never think about.

Freedom Grid became the name for studying that system.

Why This Exists

Over the past nine months I have been building a structured research process to track the infrastructure behind the AI economy.

Power generation. Cooling systems. Water resources. Grid interconnection backlogs. Emerging energy technologies capable of supporting AI scale computing.

The core observation is simple.

Artificial intelligence is forcing a massive buildout of physical infrastructure across power, cooling, and water systems.

That buildout will take years.

Possibly decades.

And the companies that solve the infrastructure bottlenecks along the way may become some of the most important players in the AI economy.

Freedom Grid exists to follow that story.