There is exactly one example of a large industrial economy decarbonizing its electricity in a hurry, and it happened before anyone was trying to decarbonize anything.

In March 1974, five months after the oil embargo exposed how much of France ran on imported petroleum, Prime Minister Pierre Messmer went on television and committed the country to building nuclear reactors at a pace no democracy had attempted in peacetime. The plan was not debated in parliament. It was not put to referendum. EDF, the state utility, ordered sixteen reactors that year alone, doubling the national order book in a single procurement.

Over the next fifteen years France built fifty-six of them. The fleet went from fifteen reactors in operation at the start of the 1980s to fifty-five by the end. By the time it was finished, France produced roughly three-quarters of its electricity from nuclear fission, and it has stayed there for forty years.

No large democracy has matched that pace before or since; even China’s current nuclear buildout, enormous in absolute terms, has not replicated France’s per-capita sprint. If you want to know whether the firm, low-carbon grid the climate models assume can actually be built on a compressed timeline, France is the only completed experiment you have. It is worth understanding precisely what it proves — and, more importantly, what it doesn’t.

Because the most instructive thing about the French nuclear program is not that it succeeded. It is that France, having done the hardest version of this once, can no longer do it at all.

For the Rust Belt specifically, this is not an abstract energy debate. The region still holds much of the country’s industrial memory — steel, forgings, turbines, welders, river logistics, rail, machine shops, the habit of building heavy things meant to last longer than a software cycle. If America is serious about nuclear, the question is not only which reactor wins. It is whether the industrial belt that once made the components of national power can be rebuilt to do it again.

The lesson comes in three parts. France proved nuclear works when a country standardizes the machine and finances it like national infrastructure. France also proved that the capacity to build can decay until even a single reactor becomes a fiasco. America’s recent record, from Vogtle to NuScale, shows the same failure in a different form — not bad physics, but bad institutions and bad capital structure. And the new Westinghouse push matters because it is the first American attempt in decades to solve the actual problem, which was never engineering.

Nuclear’s modern problem is more accurately a question of capital structure: who absorbs the first-of-a-kind risk, who carries the interest during the years of delay, and who gets to build the second unit cheaper than the first.

What France actually proved

The standard telling treats the Messmer Plan as a triumph of engineering or national will. It was neither, exactly. It was a triumph of standardization and finance — and those are the two variables that matter for anyone trying to learn from it.

France did not build fifty-six different reactors. It built a handful of designs, repeatedly, mostly licensed from Westinghouse and adapted domestically. Build the same thing fifty times and the fiftieth is dramatically cheaper than the first: the workforce learns, the supply chain matures, the regulatory questions get answered once instead of every time. This is the learning curve, and it is the entire reason the buildout was affordable. EDF moved to larger standardized designs specifically to capture economies of scale.

The financing was the other half. A state utility borrowing at sovereign-adjacent rates, building on a national mandate, does not face the cost-of-capital problem that kills nuclear everywhere else. Nuclear plants are almost all upfront cost — the fuel is cheap, the construction is brutal — which means the interest rate during construction is not a detail, it is the ballgame. France financed its fleet at terms no ordinary American utility project could replicate, and it never had to relitigate each plant in front of a hostile rate commission.

So the benefits were real and they were large. Four decades of some of the lowest-carbon electricity in the developed world. Electricity prices that stayed low and stable while neighbors swung with the gas market. Genuine energy independence — the original goal, climate being a happy accident the French backed into fifty years early. An export business in both power and reactor technology. And a live demonstration of exactly the attributes the rest of this decade is rediscovering it needs: power that is dense, dispatchable, domestic, and always on.

That is the case for nuclear, and France is the proof of concept. But a proof of concept is only useful if you understand the conditions it ran under. And the conditions are where the story turns.

How France lost the machine

Three problems came for the French fleet, and all three are instructive because they are not arguments against nuclear — they are arguments about what you are actually signing up for.

The first is that standardization cuts both ways. The same design fifty times means the same flaw fifty times. In late 2021, inspectors found stress-corrosion cracking near welds in certain reactors, rooted in a design geometry EDF had adapted from the original — and because the fleet was standardized, the discovery could not be contained to one plant. By the summer of 2022, thirty-two of France’s fifty-six reactors were offline. National nuclear output fell to 279 terawatt-hours, the lowest since 1992, and France, the model of energy independence, became a net electricity importer for the first time in decades — during the European energy crisis, at the exact moment nuclear was supposed to prove its resilience against the gas shock. A standardized fleet carries standardized failure risk. The warning arrived on schedule.

The second problem is demographic, in the way that fleets have demographics. Build fifty-six reactors in fifteen years and you have bought yourself fifty-six retirements in roughly fifteen years, four decades later. The grand carénage — the refurbishment program to push the fleet past forty years — runs into the tens of billions and climbing. The sprint that built the fleet so cheaply guaranteed the bill would come due all at once. You do not avoid the capital-stock turnover problem with a compressed buildout; you time-shift it.

The third problem is the one that should stop anyone who thinks the answer is simply “do what France did.” France can no longer do what France did.

Flamanville 3, the reactor France began in 2007 to prove it could still build, was supposed to cost €3.3 billion and open in 2012. It connected to the grid in December 2024 — more than twelve years late — at a construction cost of €13.2 billion. The French Court of Auditors, counting the financing, put the all-in figure near €23.7 billion. At that number, the auditors calculated, the electricity would have to sell at roughly €138 per megawatt-hour to earn a four-percent return. That is not competitive power. That is a monument.

The country that built fifty-six reactors in fifteen years took seventeen years to build one. The capability did not transfer across the generational gap. The workforce that knew how to pour the concrete and weld the circuits had retired. The supply chain had atrophied. The standardization advantage was gone, because you cannot have a learning curve on a fleet of one. Everything that made the 1970s buildout cheap was absent from the 2000s rebuild, and the cost reflected it precisely.

The American mirror

If you want to know whether the United States can simply decide to build nuclear, the answer is already in the record, and it reads almost identically.

Plant Vogtle in Georgia, Units 3 and 4, are the only new reactors completed in the United States in more than thirty years. They used the Westinghouse AP1000, a new design, built for the first time. They were projected to cost $14 billion and open in 2017. They were finished in 2024 — seven years late, fifteen years from start to finish — at a final cost of $36.8 billion. Westinghouse went bankrupt partway through. The twin project in South Carolina, the V.C. Summer expansion, started at the same time using the same reactor, was abandoned in 2017 after billions had been spent and nothing was finished; some of the executives involved went to prison over the cost misrepresentations.

Read the French and American stories side by side and the symmetry is the whole point. Two of the most capable industrial states on earth, with every incentive and considerable political will, have struggled to build large nuclear reactors on a schedule or budget that can serve as a repeatable model. France lost the capability it once had. The United States arguably never industrialized it — Vogtle and Summer were first-of-a-kind builds precisely because America never did the standardized fleet France did.

This is why the lazy version of the pro-nuclear argument fails. “France did it, so we can too” ignores that France did it under conditions neither country can reconstruct: a single state utility, a standardized design built dozens of times, sovereign-cost financing, and a political system that could override local opposition and never reopen the question. Strip those conditions away and you get Flamanville. You get Vogtle. You get expensive power with real operating value — sixty years of firm, clean electricity — but a construction record too poor to use as a template.

The binding constraint on American nuclear was never the physics. The reactors work; Vogtle is generating clean power right now and will for sixty years. The binding constraint is institutional and financial — the cost of capital, the absence of a standardized design, the regulatory relitigation of every plant, and a utility model that earns its return on getting projects approved rather than on building them cheaply. The regulatory weight is real and worth understanding on its own terms. After the 1971 Calvert Cliffs decision forced nuclear licensing to take environmental review more seriously under NEPA, the permitting process became slower, broader, and easier to litigate. At the same time, the ALARA principle — radiation exposure kept “as low as reasonably achievable” — created a safety logic with no obvious stopping point: if another layer of protection can be justified, the plant can be asked to carry it. Some studies attribute a meaningful share of the 1970s and 1980s nuclear cost escalation to this regulatory ratchet, much of it before Three Mile Island. Then came the accident itself, only days after The China Syndrome had put a fictional meltdown into theaters. The result was not just higher cost. It was a cultural fusion of civilian nuclear power with catastrophe.

There is also a political economy question underneath the cost history. Nuclear threatened incumbent fuel systems, especially coal, oil, and gas, because it offered dense baseload power without a continuing fuel market of the same kind. That does not prove a coordinated campaign against it. The regulatory ratchet, local opposition, environmental law, and post-Three Mile Island fear had plenty of organic force on their own. But an honest account of why America stopped building should still ask who benefited from delay, because the economics of the delay were not neutral. These are the same institutional questions that run through every account of why the American grid costs what it does. Nuclear just states them in their most expensive form.

Two bets, and the one that isn’t

So the country needs more firm, dense, carbon-free power — the kind that survives the winter peak, that holds when the wind drops, that the prior essays in this series have argued the retirement schedule wrongly assumed away. And the one technology that unambiguously provides it cannot currently be built at acceptable cost by the two nations best positioned to build it. The way out is not a new machine. It is a different way of paying for the machine that already works.

The popular answer is the small modular reactor — shrink the machine until you can build it in a factory, identically, hundreds of times, and buy the French learning curve through manufacturing repetition instead of state command. It is an appealing thesis. It is also where most of the hype has attached to the shakiest structures, and the distinction between the two tracks of the nuclear revival is the most important thing to get right.

Start with what the institutional argument actually predicts. If the binding constraint is the cost of capital and the absence of a repeatable design, then the credible builders are the ones with a licensed, operating design and a balance sheet — or a state — large enough to absorb first-of-a-kind losses. That is not a clean-sheet startup. It is an incumbent.

Watch where the real money went. In October 2025 the U.S. government, Brookfield and Cameco announced a strategic partnership intended to accelerate at least $80 billion of new Westinghouse AP1000 deployment, with the government helping organize the financing and permitting framework and the regulator extending the AP1000 design certification to 2046. Strip away the politics and look at the structure: it is the only Generation-III+ design actually built and operating in America — the Vogtle reactors — slated to be poured again, repeatedly, with the cost-of-capital problem addressed the way it has most often been addressed in successful nuclear buildouts: by moving part of the construction risk toward a public or sovereign balance sheet. That is the France formula, transposed. Whether the second AP1000 fleet comes in dramatically cheaper than the first is the entire bet — and it is the only bet with a completed American reactor underneath it. Large reactors are not the part to give up on. They are the part the institutional logic actually points to — for utility-scale baseload, behind a utility or a state that can carry a thirty-year asset.

The genuinely promising small reactor follows the same rule, not the opposite one. GE Vernova Hitachi’s BWRX-300 — a simplification of a boiling-water design that has operated for decades — is the first SMR under construction in North America, building now at Ontario’s Darlington site behind a public utility paying out of its own balance sheet, with the Tennessee Valley Authority pursuing the same design at Clinch River in Tennessee. An incumbent vendor, a proven design lineage, a creditworthy public offtaker. Even here the honest caveat holds: the BWRX-300’s marketing price of roughly $2,250 a kilowatt is not what the first Darlington units are actually costing, and first-of-a-kind is exactly the phase where Flamanville and Vogtle went wrong. But the structure is sound. This is the small reactor’s real lane — not as a replacement for the large fleet, but for the data-center campuses and industrial loads that are co-located, that have a deep-pocketed offtaker, and that the grid cannot serve fast enough on its own.

Then there is the speculative tier, and it should be named as such. NuScale holds the only U.S.-certified small-reactor design, which sounds like a lead until you read the rest of the record: its flagship project in Idaho collapsed in 2023 when the per-megawatt cost climbed past what its municipal subscribers would pay and they walked. Its public-market structure rewards design milestones, customer announcements, and memoranda of understanding long before the hard validation arrives — a built, operating, repeatable plant. That is not a knock on the engineering. It is the argument of this entire essay restated a third way: the incentives are pointed at the stock, not the steel, and a company without a balance sheet to absorb the first-of-a-kind loss is the wrong institutional vehicle to solve a problem that was institutional all along. The reactor is not what fails. The capital structure is.

Who pays for the first one

To understand why the $80 billion structure looks the way it does, you have to start with why the state suddenly cared. For twenty years electricity demand in America was flat, and a flat market does not produce strategic panic. That ended. Artificial intelligence turned electricity into the input that determines who wins the most important technology race of the era. One industry description captures the logic: a data center takes the most expensive form of energy, electricity, and turns it into intelligence. And on the metric that matters, the United States is behind: China now generates more than twice as much electricity as America does, self-sufficient through coal and renewables, while the U.S. faces the scale-up and deployment constraints the prior essays in this series have documented. When power becomes the binding constraint on national competitiveness, power stops being a market and becomes a matter of state.

Bridgewater has a name for the broader shift, and it is the right one: modern mercantilism — the return of the state to the center of strategic industries, prioritizing national strength, supply-chain security, and self-sufficiency over the free-market consensus that governed the prior forty years. Their energy-security work makes the corollary explicit: cheap, abundant, domestic power is now treated as a precondition for winning the AI race and everything downstream of it. Once you see the nuclear revival through that lens, the financing structure stops looking strange and starts looking inevitable.

Recall the Green Premium — Bill Gates’s framing that the clean option costs more than the dirty one, and that the transition is fundamentally the work of driving that premium to zero through innovation. It is a useful idea with a blind spot. For nuclear, the premium is not mostly a technology problem. A reactor is overwhelmingly upfront capital; the fuel is cheap and the plant runs for sixty years. That means the cost of the electricity is set, more than by anything else, by the interest rate during construction and the years of delay before the asset earns a dollar. Flamanville’s auditors said it plainly: at €23.7 billion all-in, the power needs to sell near €138 a megawatt-hour to clear a four-percent return. Much of that premium is not the reactor. It is the financing of the reactor — plus the delay risk that financing magnifies.

Which means there are three ways to kill the premium, not one. You can innovate it away, which is Gates’s bet and a slow one. You can equitize it — raise speculative capital and hope the learning curve arrives before the subscribers leave, which is the bet that just failed in Idaho. Or you can finance it away: have the state supply the one input only the state owns below market — sovereign, patient, low cost-of-capital — and let the private builder supply the design and the operations.

The $80 billion AP1000 structure is the third option made concrete. The government helps organize the financing and permitting framework, takes a slice of the upside through a profit-participation interest — 20% of Westinghouse’s cash distributions above a $17.5 billion threshold, convertible to an equity stake of up to roughly 20% if the company goes public above $30 billion by 2029 — and, by facilitating the financing of long-lead-time items, moves part of the construction risk toward the sovereign balance sheet. The premium does not have to be innovated down to zero, because the most expensive part of it — the cost of capital — is being shifted to where it is cheapest.

The mercantilist lease — not legally, but economically. Look at where the capital actually comes from and the arrangement names itself. The stack is being shaped partly through allied money. Japan’s $550 billion U.S. investment commitment has been reported as a potential vehicle for nuclear projects involving Westinghouse, with Japanese firms positioned in the supply chain. Korea’s broader industrial package, and firms like Doosan, point in the same direction: allied capital and allied heavy industry being organized around American strategic capacity. Allied surplus capital is being steered toward a domestic strategic asset that the American state helps finance, partly underwrites, and stands to share in — while the hyperscalers line up to offtake the power. Each party supplies what it holds most cheaply: the allies their capital surplus, the state its balance sheet and permitting power, Westinghouse its one proven design, the AI buildout its demand. That is not a subsidy in the old sense of a check written to lower a price. It is the state leasing out its cost-of-capital to a sector it has decided is too strategic to leave to the market — the four tenets of modern mercantilism expressed as a term sheet.

What that sovereign capital actually buys, in the physical world, is not a reactor. It is a supply chain.

The deal’s own language is explicit that Westinghouse will lean on the industrial base “established during the construction of Vogtle units 3 and 4” — which is the tell. The thing that makes the second reactor cheaper than the first is not a cleverer design. It is a workforce that has poured the concrete before, a fabricator that has passed nuclear-grade inspection before, and a forging shop that has cast a reactor pressure vessel before. The industry has a name for the destination: nth-of-a-kind, or NOAK — the cost level a design reaches once the learning curve has been climbed and the supply chain is warm. France’s fleet was the purest NOAK machine ever built. Vogtle, Flamanville, and the cancelled Summer plant were all FOAK — first-of-a-kind — and FOAK is where the money dies. The entire $80 billion bet is an attempt to buy the country a passage from FOAK to NOAK, and that passage runs through physical plant. When the four Vogtle and Summer reactors placed their major component orders simultaneously, the order-book volume alone cut the price of some key materials by a quarter to a third. Scale and repetition are not financial abstractions here; they are forgings and crews.

This is where the financing structure becomes a regional question — and where this publication keeps its attention: not the reactor as a finished object, but the critical components inside it, and who can still make them. The heaviest of those components — reactor pressure vessels, steam generators, pressurizers — are forged at a handful of facilities on earth, most of them in East Asia, which is why Korea’s Doosan supplied Vogtle’s. Korea is the comparison that should focus the mind. Its recent nuclear build costs are commonly estimated at a fraction of U.S. costs, while Vogtle became one of the most expensive nuclear construction projects in the world. The numbers are not perfectly like-for-like, and the regulatory systems are different. But the direction is not mysterious: Korea kept a nuclear supply chain warm, and the United States did not. American nuclear components ordered cold after a thirty-year gap are expensive precisely because the capability atrophied. Korean components are cheaper because Korea never stopped making them.

So the choice arrives in its sharpest form. A serious American nuclear program is a decision about whether those forgings get made here again, in shops that can handle a 17,000-tonne press and the metallurgical tolerances a reactor demands — or whether the country keeps renting the critical components from the allies who held onto the capacity, which is cheaper today and is part of what the mercantilist lease actually buys. That is not a Silicon Valley capability. It is a heavy-industrial capability — the kind the Rust Belt once specialized in. The mercantilist lease is ultimately a bet that sovereign capital can rebuild not just a reactor fleet but the heavy-industrial base that fleets require — the machine shops, the forge lines, the welders certified to nuclear standard, the river and rail logistics to move components that cannot go by road. The financing structure is the visible part. The supply chain is what it is really trying to resurrect.

This is also why the earlier distinction holds so cleanly. The structures that work are the ones aligned with where the cheap capital actually lives: the sovereign behind the AP1000 fleet, the creditworthy public utility behind the BWRX-300. The structure that failed tried to fund the most capital-intensive asset in the economy with the most expensive money in the economy — retail equity chasing a milestone narrative. Gates’s framework cannot quite see this, because it treats the premium as a property of the technology. The mercantilist sees it for what it is: a property of the financing, and therefore something a state can simply decide to absorb. Whether that is wise industrial policy or a slow socialization of enormous construction risk is a real question, and the honest answer is that we will not know until the follow-on reactors are poured. But it is the bet the country has placed, and it is a coherent one.

What the record actually recommends

The case for American nuclear does not rest on optimism, and it should not pretend to. It rests on a process of elimination, and now on a decision the state has already made. The winter peak has to be met by something firm. The AI buildout is not waiting. The retiring coal fleet provided dispatchable power that intermittent sources do not replace one-for-one. Something dense and always-on has to fill that space — the list of carbon-free options that qualify is short, and the state has now begun putting its balance sheet behind the only one of them that is also firm.

France proved the machine works and the grid it produces is worth having. France also proved that the machine is only as buildable as the institutions around it — that the reactors were never the hard part, the standardization and the financing and the political durability were, and that a country can lose those faster than it loses the engineering.

That is the actual lesson for the United States, and it is not “build reactors,” nor is it “wait for the factory to save us.” It is that the country has to rebuild the conditions under which reactors are buildable: a standardized design poured more than once, financing that does not strangle a thirty-year asset at the construction phase, and a regulatory posture that answers the safety questions thoroughly and then stops reopening them. That is what the AP1000 partnership and the BWRX-300 are each, at different scales, trying to buy — a proven design behind a balance sheet that can survive the first one. And it is what the speculative pure-plays cannot, because they are trying to solve a financing problem with the wrong kind of financing.

The honest forward-looking caveat is that none of it is proven yet. The repeat AP1000 fleet has not poured its first follow-on unit. The BWRX-300 will not run until the end of the decade. TerraPower’s Gates-backed Natrium reactor breaking ground at a retiring coal plant in Wyoming is the cleanest single image of where this decade is heading, and it is also first-of-a-kind, which is the phase that has humbled everyone who entered it. The difference between this round and the last is not certainty. It is structure: proven designs, behind balance sheets, built more than once. That is the only configuration that has ever worked.

France built it once and forgot how. America built it twice — Vogtle and Summer — and nearly swore off it. The question now is whether either country can do the one thing neither has managed since the 1980s: build the same proven reactor a second time, and a third, cheaply enough that the third is the argument for the fourth.

A software economy scales by copying code. An industrial economy scales by preserving crews, suppliers, permits, financing channels, and muscle memory. Nuclear is the place where America finds out whether it still has the second kind of memory — and the Rust Belt, which holds most of what remains of it, is where the answer will be found or lost.

The reactors were always the easy part. They were the easy part in 1974, and they are the easy part now. What France had and lost, and what America never built, was the machine that builds the machine.

Subscribe now

Share

WORKS CITED & SOURCES

Grouped by claim. Primary and canonical sources favored; some outlets are named where the original is paywalled.

France — the Messmer Plan and the buildout

How France achieved the world’s fastest nuclear buildout, Works in Progress — https://worksinprogress.co/issue/liberte-egalite-radioactivite/

History of France’s civil nuclear program, Wikipedia — https://en.wikipedia.org/wiki/History_of_France%27s_civil_nuclear_program

The beginning of nuclear energy in France: Messmer’s plan, IAEA INIS — https://inis.iaea.org/records/czje0-6cv23

France — the 2022 outages

Nuclear Power in France, World Nuclear Association — https://world-nuclear.org/information-library/country-profiles/countries-a-f/france

RTE (French grid operator) — https://www.rte-france.com/en

Flamanville 3 — cost and schedule

Cour des comptes report, via Connaissance des Énergies (€20.4bn in 2015 euros / €23.7bn in 2023 euros; original €3.3bn) — https://www.connaissancedesenergies.org/afp/programme-depr-la-cour-des-comptes-publie-un-rapport-critique-et-recalcule-le-cout-de-flamanville-3-250114

Energy Intelligence, on the Cour des comptes €110–120/MWh estimate — https://www.energyintel.com/0000017b-a7db-de4c-a17b-e7db0a550000

Vogtle & V.C. Summer

Plant Vogtle Unit 4 begins commercial operation, EIA — https://www.eia.gov/todayinenergy/detail.php?id=61963

Two Years After Completion, Plant Vogtle Still Looms…, Inside Climate News ($36.8bn final, 7 years late) — https://insideclimatenews.org/news/10052026/plant-vogtle-nuclear-debate-two-years-after-completion/

What Was Learned from Building New Nuclear Reactors?, POWER (15-year schedule) — https://www.powermag.com/what-was-learned-from-building-new-nuclear-reactors/

Vogtle project update, AP via ANS Nuclear Newswire (Summer abandonment; cost milestones) — https://www.ans.org/news/article-3949/vogtle-project-update-cost-likely-to-top-30-billion/

Regulation, culture, and political economy

It’s the Regulation, Stupid, The Breakthrough Institute (Calvert Cliffs, regulatory ratchet) — https://thebreakthrough.org/journal/no-20-spring-2024/its-the-regulation-stupid

Why Does Nuclear Power Plant Construction Cost So Much?, Institute for Progress (Eash-Gates study) — https://ifp.org/nuclear-power-plant-construction-costs/

Bernard Cohen, The Nuclear Energy Option, ch. 9 (regulatory ratcheting) — http://www.phyast.pitt.edu/~blc/book/chapter9.html

Jack Devanney, Nuclear is too Expensive (ALARA; Korea vs. U.S. cost) — https://jackdevanney.substack.com/p/nuclear-is-too-expensive

Three Mile Island accident was eerily foreshadowed by a Hollywood blockbuster, WHYY (The China Syndrome, 16 March 1979; TMI, 28 March 1979) — https://whyy.org/articles/three-mile-island-accident-was-eerily-foreshadowed-by-a-hollywood-blockbuster-days-before/

Korea vs. U.S. build cost

Heavy Manufacturing of Power Plants, World Nuclear Association (forging capacity; Doosan) — https://world-nuclear.org/information-library/nuclear-fuel-cycle/nuclear-power-reactors/heavy-manufacturing-of-power-plants

Jack Devanney, Nuclear is too Expensive (Korea ~$2,000/kW; Vogtle >$10,000/kW) — https://jackdevanney.substack.com/p/nuclear-is-too-expensive

The Westinghouse / AP1000 partnership ($80B)

United States Government, Brookfield and Cameco announce transformational partnership, Cameco (28 Oct 2025) — https://www.cameco.com/media/news/united-states-government-brookfield-and-cameco-announce-transformational-partnership

Cameco Form 6-K, SEC (partnership structure: 20% of distributions above $17.5bn) — https://www.sec.gov/Archives/edgar/data/0001009001/000119312525253372/d935274dex991.htm

US Government Announces Historic $80 Billion Nuclear Partnership…, K&L Gates — https://www.klgates.com/US-Government-Announces-Historic-80-Billion-Nuclear-Partnership-with-Westinghouse-Electric-Company-Cameco-Corporation-and-Brookfield-Asset-Management-to-Construct-AP1000-Reactor-Fleet-10-30-2025

US partners with Westinghouse, Cameco and Brookfield on $80B nuclear deployment, Utility Dive — https://www.utilitydive.com/news/westinghouse-cameco-brookfield-nuclear/803999/

SMRs — BWRX-300, NuScale, TerraPower

GEH BWRX-300 SMR Approved for Construction at OPG’s Darlington Site, Neutron Bytes (construction May 2025; first unit ~2030) — https://neutronbytes.com/2025/05/17/geh-bwrx-300-smr-approved-for-construction-at-opgs-darlingtion-site/

BWRX-300, Wikipedia (OPG timeline; final investment decision May 2025) — https://en.wikipedia.org/wiki/BWRX-300

NuScale / UAMPS Carbon Free Power Project cancellation, 2023 — reported by Reuters and the U.S. Department of Energy (search “NuScale UAMPS Carbon Free Power Project terminated 2023”)

Supply chain & FOAK to NOAK

Feds in $80 billion Partnership with Westinghouse to Build AP1000s, Neutron Bytes (leverage of the Vogtle-era supply chain) — https://neutronbytes.com/2025/10/28/feds-in-80-billion-partnership-with-westinghouse-to-build-ap1000s/

NSI report addresses supply chain bottlenecks, ANS Nuclear Newswire (25–30% order-book savings) — https://www.ans.org/news/article-7870/nsi-report-addresses-supply-chain-bottlenecks/

Nuclear supply chain under pressure in a fragmented world, Damona (East-Asian forging concentration) — https://www.damona.co/nuclear-supply-chain-under-pressure-in-a-fragmented-world/

Financing frame — modern mercantilism, the AI-power thesis, Green Premium

Bridgewater Associates / Greg Jensen, We’re All Mercantilists Now — bridgewater.com (Research & Insights)

Bridgewater Associates, Power Politics: Energy Self-Sufficiency in a Modern Mercantilist World (Oct 2025) — bridgewater.com (Research & Insights)

U.S. Energy Information Administration, international electricity data (China vs. U.S. generation) — https://www.eia.gov/international/data/world

Bill Gates, How to Avoid a Climate Disaster (the “Green Premium”) — gatesnotes.com

A note on the figures referenced in the charts: dollar and euro totals are rounded; the AP1000 and BWRX-300 economics are forward commitments, not realized costs. The “mercantilist lease” is an analytic characterization of the deal’s structure, not a legal description of ownership.

SUBSCRIBE

If you’ve read this far, you’re exactly who The Rustbelt Reader is written for — people who want to understand the physical economy as it actually works: the capital, the supply chains, the institutions, and the industrial memory that decide what gets built and what doesn’t.

Subscribe to get every essay in your inbox. Free subscribers get the full pieces. It’s the best way to make sure the next one finds you.

And if this clarified something, forward it to one person who’d argue with it.

In Tucker County, West Virginia, the last working coal mine closed this winter. Two hundred jobs in a county of 6,572 people — three percent of every man, woman, and child.

McDowell County, in the southern coalfields, was the largest coal-producing county in the United States in 1950. It held 98,887 people that year. By 2020, it held 19,111 — an eighty-percent population collapse over seventy years. The county that fueled the steel of the American industrial century now has a median household income of $30,000 and a poverty rate of thirty-one percent. Mingo, Logan, Wyoming, Boone — every Central Appalachian coal county has absorbed losses on similar proportional scales.

The wealth extracted from these mountains helped build the American century. Pittsburgh steel, Cleveland steel, Gary steel, the rails that carried it, the ships and tanks that fought two wars — all of it ran on coal pulled from these mountains. Most of the wealth did not stay. What remained in the host counties was acid mine drainage that does not age out of the watershed, an aging labor force, and a tax base that depended on extraction the geology and the policy could no longer sustain.

Since the start of 2025, West Virginia and Eastern Kentucky together have lost another 1,600 coal jobs. The arc continues.

But the country has stopped asking what the arc is actually costing — not in jobs alone, which would be enough, but in operational capabilities the system is removing before it has built the replacements.

Why Coal Was Crowned

Coal was not crowned by politics. It was crowned by physics.

A coal-fired generating station occupies roughly 1,500 W/m² of land — the upper register of any dispatchable power source humans have built. Vaclav Smil’s work catalogs the comparison: natural gas combined-cycle 1,200 W/m², nuclear 800 W/m², solar PV around 12 W/m², wind around 2 W/m². The exact values vary by site and technology, but the order-of-magnitude separation is structural. A coal plant could power a city on a footprint no diffuse source can match.

It burned when you put it in the firebox. It could be stored in a pile next to the plant. And it was American — the Pittsburgh seam, the Illinois Basin, the Powder River Basin. The country that built itself on coal did not need to negotiate with foreign powers for the energy of its industry.

By 2026, coal’s share of US electricity generation, near fifty percent as recently as 2005, had fallen to roughly fifteen percent. The retirement schedule was built in an environment of flat demand, $2 natural gas, and the assumption that liquid hydrocarbon supply was reliable. The trade was defensible on those terms.

The terms have changed.

What Coal Specifically Does

The case for keeping coal capacity online during the transition window is not a case for coal. It is a case for not removing capabilities the system actually uses before the replacements are built and operating.

Five specific attributes of the existing coal fleet matter to the grid as it actually functions in 2026. None of them is captured by the levelized cost framework that drove the retirement schedule. All of them are physical attributes that flow from how the fuel actually works.

Fuel on the ground. A typical coal plant maintains thirty to ninety days of fuel inventory on-site as a physical pile of rock. The fuel does not transit a maritime chokepoint. It does not depend on pipeline pressure. Natural gas plants carry hours of fuel in pipeline pressure. Wind and solar carry none. The coal pile is the closest thing the US grid has to a strategic reserve for the electricity sector.

Synchronous inertia. A coal-fired generating unit has a heavy steam turbine spinning at thousands of revolutions per minute, mechanically coupled to the grid — a massive flywheel. When demand surges or supply drops, the kinetic energy stored in those spinning masses physically resists frequency change for several critical seconds, long enough for governors to respond and reserves to come online. Inverter-based resources do not provide this.

Surge dispatchability. The grid is running coal because at the margin of system stress, coal is the dispatchable inventory available. A coal plant can run flat-out for weeks at a time, without weather dependency, without fuel resupply. When PJM forecasts a 147 GW winter peak and the wind drops and gas pipeline pressure falls, the coal plants still operating run hard for the duration of the event. That is the dispatchable inventory the system still has.

Geography that matches the buildout. The retired and operating coal sites are concentrated in PJM, MISO, and ERCOT — the territories where data center load is growing fastest. They carry grid interconnections, transmission corridors, water rights, and rail access that took decades to build and cannot be quickly replicated. This is path dependency, in the technical sense: the value of these sites is not just the plant — it is the institutional infrastructure that grew up around them. The replacement system has to either inherit that infrastructure or rebuild it. Retiring the host sites before the inheritance is captured loses optionality the country cannot quickly recreate.

Domestic integration. Coal is the most domestically integrated dispatchable fuel in the US electricity system. The Pittsburgh seam, the Illinois Basin, and the Powder River Basin are American resources moved on American railroads to American plants. Mining equipment, rail capacity, and metallurgical supply chains are not perfectly domestic, but they are more so than the alternatives. Natural gas crosses pipelines that increasingly serve LNG export markets competing with domestic delivery. Uranium fuel-cycle services depend on a global supply chain that includes Russian enrichment. In a post-Hormuz world where energy security has returned to the policy table, the domestic integration of coal is worth more than the framework assumed.

None of these attributes is an argument for new coal capacity. None of them denies that coal has real costs the alternatives mostly do not. Together, they suggest that the existing fleet is providing operational capabilities the system has not yet replaced.

The Fleet Has Already Decided

The most reliable evidence about which coal plants the system needs is not found in policy documents, levelized cost models, or retirement projections. It is found in the EPA’s Continuous Emissions Monitoring database — hourly generation data every coal unit reports under federal law, cross-referenced with EIA’s plant-level retirement status flags.

That operating record is the empirical fact.

From 2017 through 2024, EIA Form 860M forecasts of coal retirements were directionally accurate. Planned retirements and actual retirements matched within reasonable error each year. Policy and operations were aligned.

In 2025, that alignment broke. The forecast missed by more than 7 GW. Of the approximately 8 GW of coal-fired capacity scheduled to retire in 2025, only 2.6 GW actually retired — the lowest annual total since 2010. Three distinct mechanisms drove the override: DOE Section 202(c) emergency orders directing specific plants to stay online for reliability (Campbell in Michigan, Schahfer and Culley in Indiana, Craig in Colorado, Centralia in Washington); a PJM Reliability Must Run agreement keeping Brandon Shores in Maryland operating through 2029; and utility-driven delays citing data center load (Oak Creek, Comanche, others).

This is what happens when a policy framework collides with physical constraint. The structural break is not a single-year event. It is a regime change in how the retirement schedule will function as long as demand growth continues to outrun replacement capacity.

This is the strongest evidence in this essay, because it does not depend on projections, prices, or policy interpretation. It is hourly federal data showing what the grid actually dispatched in 2024.

The coal plants the DOE intervened to keep online had been running at capacity factors substantially above the fleet average through 2024 — above 60% in winter. The utility-delayed plants had been running similarly hard. The plants that actually retired in 2024 had been operating at 20–40% capacity factors for years before retirement — they were already not serving the dispatchable function, and their retirement did not produce operational disruption.

The grid has been bifurcating the fleet operationally for years.

The 2025 federal interventions confirmed what the operating record had already revealed.

The retirement schedule selected plants by age, emissions profile, and economic competitiveness against gas. But the operating data shows that capacity factor — how hard the grid actually runs each plant — was a better predictor of whether a retirement would succeed without operational disruption. The framework was not asking that question. The 2025 interventions are the result.

The Terms Have Changed

US electricity generation grew 3% in 2024 and 2.8% in 2025 — the first sustained acceleration in more than two decades. Data center demand alone is projected to climb to roughly 600 terawatt-hours by 2030, approaching 12% of total US electricity consumption.

The supply side cannot respond on the required timeline. The PJM interconnection queue, closed in 2022, did not fully reopen until April 2026. New gas plants entering the queue today are unlikely to deliver power before 2034. Combined-cycle turbines from the three global heavy-duty manufacturers are committed through the end of the decade. The PJM Base Residual Auction cleared at $28.92 per megawatt-day for delivery year 2024/25, $269.92 for 2025/26, and $329.17 for 2026/27 — a 9× to 11× move in two years. This is what physical scarcity looks like.

The Iran war added a layer the schedule had not priced. The closure of the Strait of Hormuz disrupted as much as 13 million barrels per day of oil flows and threatened comparable volumes of natural gas. Every kilowatt of dispatchable generation the country still operates became more valuable the day that chokepoint closed.

The retail evidence is already in the rate base. West Virginia, at 86.9% coal generation, has residential electricity rates roughly 22% below the US average. Kentucky has the lowest residential rate east of the Mississippi. The Rocky Mountain states with coal-and-wind grids — Wyoming, Montana, North Dakota — sit at the bottom of the national rate ranking. California, which has moved fastest away from dispatchable thermal, now pays the highest residential rate at 34.26¢/kWh.

The comparisons carry confounds, but the directional pattern is difficult to ignore: the states still running coal at scale are charging their residents less for electricity than the states that retired it. The Kentucky Public Service Commission’s own analysis puts the mechanism plainly: coal’s delivered price is less exposed to the global gas market than any other dispatchable fuel. That stability is what the retirement framework treated as expendable.

Energy transitions take time the policy framework does not allow for. Vaclav Smil’s documentation of US electrification is the empirical anchor: the transition took 26 years to connect 10% of US dwellings and 75 years to reach 99% — longer than the average American life expectancy when it began, despite federal coordination and a product so superior that farming families gave up 28% of their annual income to live in an electrified home.

The current policy framework assumes the coal-to-renewables transition can occur in 15–20 years. For infrastructure systems of this magnitude, the historical record does not support that assumption.

What the Country Has Earned the Right to Ask

The retirement schedule was built in good faith inside a climate policy framework that deserves honest engagement. The physics of CO₂ is well-established. The climate concern is scientifically grounded.

What the schedule did not price was the second cost: what happens to electricity rates, industrial competitiveness, and basic reliability when capacity is removed before replacement is built. The framework optimized for climate cost under the assumption that demand was flat, gas was cheap, and the hydrocarbon supply was reliable.

The assumptions changed. The framework did not.

The honest framing is not “save coal.” Coal will not rule again. It is: keep what works until what replaces it actually works, and pay the host communities for the service they are providing while the rest of the system catches up.

The fleet has already decided which plants the country needs. The policy framework is catching up.

Subscribe now

Share

Leave a comment

There are two energy crises happening right now, and most analysts are treating them as separate stories.

The first is domestic. The United States spent forty years optimizing its power grid for a world where electricity demand was flat. From 2003 to 2023, generation grew at less than 1% per year. The grid tightened reserve margins, retired aging capacity faster than it replaced it, allowed interconnection queues to backlog, and watched the skilled-trades pipeline age out. Then, in 2024, demand grew 3%. In 2025, another 2.8%. AI compute, reshoring, electrification — all arriving at once, on a system that was not maintained for growth.

The second is global. The US-Iran war and the effective closure of the Strait of Hormuz have removed 13 million barrels per day of oil from world markets and 140 billion cubic meters of natural gas — by the IEA’s own measure, larger than the 1973, 1979, and 2022 crises combined. Brent crude moved from $72 to $120 in six weeks. The IEA released 400 million barrels of strategic reserves, its largest emergency drawdown ever. Forty energy assets across nine countries were severely or very severely damaged. The agency’s executive director has called it “the greatest global energy security threat in history.”

These are different crises. Different causes, different geographies, different policy responses. If the Strait of Hormuz reopened tomorrow, the US grid capacity gap would remain. If the PJM interconnection queue cleared tomorrow, the global oil shock would remain.

But both expose the same underlying reality: the global economy was built for a world of cheap, abundant, geographically concentrated energy — and that world is ending. The two crises arrived simultaneously, and each amplifies the other.

That convergence — not the events themselves — is what matters.

What the Data Shows

We analyze US Energy Supply-Demand through a Dashboard that tracks eight panels organized into four sections: demand, supply, bottlenecks, market signals. Read together, the panels tell a single story.

The demand panels show the twenty-year flat line breaking. US electricity generation grew at less than 1% CAGR from 2008 through 2021 — a fourteen-year plateau confirmed by the EIA’s own characterization. The breakout in 2024 (+3%) and 2025 (+2.8%) is the steepest sustained acceleration in over two decades. Data center demand alone is projected to grow from 147 TWh in 2023 to 606 TWh by 2030, reaching nearly 12% of total US power demand.

The supply panels show why the system cannot respond. From 2014 through 2024, fossil retirements exceeded new gas and nuclear additions every single year. The deepest year was 2024 at roughly negative seventeen gigawatts — meaning the country shrank its reliable, dispatchable generating capacity at an accelerating pace while demand was about to surge. Renewables grew through the same period, but solar and wind cannot, by themselves, run a steel mill at three in the morning or a data center through a January cold snap.

The bottleneck panels show why supply cannot catch up. PJM Interconnection — the grid operator covering Pennsylvania, New Jersey, Virginia, and eleven other states — closed its interconnection queue in 2022 because the backlog had become unmanageable. The queue did not fully reopen until April 2026, when 220 GW of new generation applications arrived almost immediately, dominated by gas and storage. Expert testimony in Maryland regulatory proceedings concluded that gas plants in this cycle won’t come online until 2034 or 2035. Heavy-duty gas turbines from the three global manufacturers are committed through the end of the decade. Large power transformers carry lead times of 40 months and rising.

The market panels show what reliability now costs.

PJM capacity auction clearing prices ran around $29 per megawatt-day for delivery year 2024/25. The 2025/26 auction cleared at $270. The 2026/27 auction cleared at $330. A 9× to 11× move in two years — the system paying a structural premium for any generation that can actually be relied upon during peak. This is what physical scarcity looks like in a regulated market.

The dashboard tells a coherent story across all four sections: demand is breaking out, supply structurally cannot respond, the bottlenecks are physical and binding, and the market has already started pricing the result. This was true before the Iran war began.

The Other Half of the Picture

The IEA’s quantitative claim about the current crisis can be tested. The 1973 oil embargo removed roughly 5 million barrels per day from global supply. The 1979 Iranian Revolution removed roughly the same. The 1990 Gulf War took out about 4.3 million barrels per day. Russia’s 2022 invasion of Ukraine removed 75 billion cubic meters of natural gas from European markets.

The current crisis has removed 13 million barrels per day of oil and 140 billion cubic meters of natural gas — larger than 1973 and 1979 combined, with the gas loss almost twice the size of the 2022 European shock.

The market response confirms the scale.

Brent crude jumped 55% in six weeks. The IEA executed its largest-ever coordinated reserve release — 400 million barrels, drawing on the strategic petroleum reserves of all 32 member countries simultaneously. The IEA’s emergency architecture has been activated only six times since 1974. The current release is the largest. Even so, the agency’s own framing is that the release only “reduces the pain.” It is not a cure. The cure is opening the Strait of Hormuz.

The supply shock is not just crude oil. Europe gets about 75% of its jet fuel from refineries in the Middle East — and that is now effectively zero. Petrochemicals, fertilizers, sulfur, and helium are all disrupted. Vital arteries of the global economy that no one talks about until they stop flowing.

Why They Converge

Two different causes. One shared conclusion.

The US capacity crisis was caused by capital allocation — forty years of underinvestment in physical infrastructure while financial capital chased software and asset-light returns. Tech and hyperscaler capital expenditure rose to 2.1% of US GDP by 2026; oil and gas upstream sits at 0.85% of GDP, half of its 2014 peak; mining sits below 0.4% where it has been since 1990. The economy redirected its productive capital toward digital infrastructure and away from physical infrastructure for an entire generation.

The Iran war crisis was caused by geographic concentration — a hundred years of building the global energy system around the assumption that Middle Eastern oil would flow reliably through specific maritime chokepoints. Twenty percent of the world’s daily petroleum consumption transits the Strait of Hormuz. When that flow stops, no other geography can absorb the redirection on the timeline the world economy requires.

Both expose the same truth: the global economy was built on cheap, reliable, abundant energy that came from somewhere specific, on infrastructure that was either not maintained or not diversified. When the demand environment changed — AI compute on one side, geopolitical fragmentation on the other — the physical system could not absorb the shock.

The deeper alignment becomes visible when you read both at once. The Iran shock accelerates electrification. When oil supply is disrupted at this scale, the long-term capital response is to reduce dependence on liquid fuels for transport, industrial process heat, and power generation. Nuclear restart. Renewables expansion. Electric vehicle acceleration. Grid hardening. The strategic playbook for post-Hormuz energy security is electrification.

But every one of those responses requires the US power grid to absorb dramatically more load. Electrified transport runs on electricity. Heat pumps run on electricity. Reshored manufacturing runs on electricity. Data centers run on electricity. The geopolitical response to the Iran war pushes more activity onto the same grid that is already failing to absorb the AI buildout.

The Iran shock does not solve the US grid problem. It compounds it.

What Gets Built

The world is going to spend the next decade rebuilding physical energy infrastructure at a pace not attempted since the 1970s.

Industrial Info Resources tracks 2,883 data centers valued at approximately $2.43 trillion planned over the 2026-2030 period. Add fossil retirements being replaced, transmission expansion, nuclear restart, battery storage at scale, and the rerouting of liquid fuel logistics around chokepoints. The cumulative capital requirement runs into the multiple trillions.

The lead times are non-negotiable. The labor is finite. The trades aged through the post-2003 underinvestment period. The workforce now competing for AI data center construction is competing with reshored manufacturing and federal infrastructure spending for the same finite pool. The crises arrive simultaneously and compete for the same workers. The supply chain that builds the buildout is the supply chain that the buildout is trying to replace.

This is not a forecast. This is a description of the conditions that already exist.

What This Means

The financial economy spent fifteen years pricing the world as if physical boundaries no longer mattered. The physical layer is now reasserting that they do — across multiple dimensions simultaneously, on timelines that do not negotiate.

Both crises were inevitable in the sense that the conditions that produced them were already in place. The capital allocation pattern that starved physical infrastructure for forty years was going to produce a physical infrastructure crisis eventually. The geographic concentration of oil supply that made the Strait of Hormuz the world’s most important chokepoint was going to produce a chokepoint crisis eventually. The specific timing of each event was contingent. The arrival of both was not.

The valuation gap is now extreme.

Integrated energy companies now generate forward free cash flow yields above 15%. The hyperscaler basket has compressed toward zero as AI capital expenditure swallows operating cash. The two sectors crossed in 2021 and have diverged steadily since. The current setup is one of the widest valuation gaps between the physical and digital economies in decades. The Iran war is the accelerant. The structural reallocation is the fire.

What the next decade will look like is the rebuilding of the physical layer at a pace the financial system has not yet priced. Capacity auction prices. Electricity rates. Transformer queues. Skilled-trades wages. Capital expenditure as a share of GDP. All of these are early indicators of a multi-year capital intensity cycle in the very sectors that have been starved for one.

Physical scarcity does not remain trapped inside infrastructure markets. It eventually appears in electricity bills, insurance premiums, freight costs, and housing affordability. The household-level consequences arrive on a slower clock than the wholesale signals, but they arrive.

The easy-energy era trained markets to believe scarcity was temporary and infrastructure optional.

The next decade is going to relearn the cost of both.

Sources & Methodology

1. US grid and capacity data. US Energy Information Administration, Electric Power Annual (October 2025); EIA Monthly Energy Review, Section 7 (June 2025); EIA Annual Energy Outlook 2025 reference case; PJM Interconnection Base Residual Auction summary documents; PJM Inside Lines, “Generation Interconnection Reform Progress” fact sheets (2025–2026); North American Electric Reliability Corporation, 2025 Long-Term Reliability Assessment (released January 2026); NERC Level 3 Alert on data center loads (May 2026); US Department of Energy, Evaluating the Reliability and Security of the United States Electric Grid (July 2025).

2. Demand-side projections. McKinsey & Company, How Data Centers and the Energy Sector Can Sate AI’s Hunger for Power (September 2024); BloombergNEF data center demand forecasts (December 2025); Lawrence Berkeley National Laboratory data center load scenarios; Industrial Info Resources, Global Market Intelligence Industrial Manufacturing Project Database (2,883 announced data center projects, $2.43 trillion over 2026–2030); Grid Strategies LLC, Review of NERC’s 2025 Long-Term Reliability Assessment (March 2026).

3. Global oil and gas shock data. International Energy Agency Executive Director Fatih Birol public statements at the Atlantic Council (April 13, 2026), the National Press Club Canberra (March 23, 2026), CNBC CONVERGE LIVE Singapore (April 23, 2026), and the Norges Bank Investment Management “In Good Company” podcast (April 1, 2026); IEA coordinated 400 million barrel strategic reserve release (announced March 11, 2026); Reuters Factbox, Biggest Global Oil Supply Disruptions in History (March 13, 2026); Al Jazeera analysis of the 1973 embargo and 2026 Hormuz closure (March 24, 2026); Brookings Institution Energy Security and Climate Initiative analysis (Samantha Gross, March 2026).

4. Market and pricing data. Brent crude futures spot prices from Trading Economics (March–May 2026); CNBC, Fortune, and Al Jazeera reporting on the Brent price trajectory; Goldman Sachs 2026 Brent forecast updates; PJM Base Residual Auction clearing prices ($29/MW-day for 2024/25; $270/MW-day for 2025/26; $330/MW-day for 2026/27); EIA average residential electricity rate series.

5. Interconnection and lead-time data. PJM Interconnection reformed queue intake (April 2026; 220 GW of new generation applications including 106 GW gas, 67 GW storage, 18 GW nuclear, 15 GW solar, 9 GW solar-storage hybrid, 5 GW wind); Maryland Public Service Commission expert testimony on gas turbine timelines (May 2026); GE Vernova, Siemens Energy, and Mitsubishi Heavy Industries 2026 order-book disclosures.

   The Dashboard

   The US Energy Supply-Demand Dashboard referenced throughout this essay tracks eight panels across four sections — demand, supply, bottlenecks, and market signals — with data-quality tags identifying which series are anchored to primary EIA/NERC/PJM data versus approximations versus forward projections.

   The interactive dashboard is available to paid subscribers of The Rustbelt Reader.

Subscribe now

Share

Leave a comment

Modern civilization runs on electricity. Every server query, every hospital ventilator, every steel furnace, refrigeration unit, water pump, and rail signal depends on energy moving through an enormous system of turbines, transformers, substations, wires, and fuel. The electric grid is not an abstraction. It is one of the largest machines ever built.

That fact matters more than ever because the current excitement around artificial intelligence has a way of making the material world disappear. The story is usually told in terms of software, models, chips, and productivity gains. But intelligence may be digital while power is physical, and physical systems obey arithmetic. AI can scale only to the extent that the electrical system beneath it can scale, and that system is governed not by hype but by engineering limits, capital intensity, and time.

The first principle is simple: electricity must be delivered instantly. Unlike oil or coal, it cannot be cheaply stored at the scale required for a modern economy, and so the system must remain balanced in real time. At every second, generation must match demand, frequency must remain stable, and voltage must stay within narrow tolerances. This is why grid planners do not design around average conditions. They design around the worst hour of the year—a freezing winter evening or a sweltering summer afternoon, when demand peaks and the margin for error disappears. Failure is not theoretical. It is a blackout.

Public debate tends to focus on how electricity is generated, as if the cost of fuel or the elegance of the technology settles the matter. But for most consumers, the largest expense is not producing electricity. It is moving it. The electric bill is, in large part, a bill for infrastructure.

A simplified cost stack often looks like this:

• Generation: ~30–40%

• Transmission: ~10–15%

• Distribution: ~40–50%

That distribution share is the key. Poles, transformers, substations, service lines, maintenance crews, vegetation management, and storm repair are expensive, labor-intensive, and geographically fixed. They do not disappear when wholesale power gets cheaper. They must be maintained whether demand is rising or flat, whether electricity is produced by gas, nuclear, wind, or solar. Much of the cost of modern electricity is not the energy itself, but the system required to deliver it reliably to millions of endpoints.

This is where artificial intelligence changes the scale of the problem. AI does not simply add incremental demand. It introduces a new type of demand—large, concentrated, and continuous. A hyperscale data center can require hundreds of megawatts at initial build, with expansion paths toward gigawatts. That is the electrical equivalent of a small city, but with a flatter, more relentless load profile and far less tolerance for interruption.

Because these facilities must run continuously, they bring the question of energy density back into focus. The issue is not ideological. It is physical. A data center that must operate 24 hours a day cannot depend on a power source that is abundant only when conditions are favorable, unless it is paired with enough storage and backup capacity to eliminate intermittency. In practice, this pushes the system toward dispatchable generation—sources that can run when needed, not just when available.

Natural gas satisfies that requirement in the near term, offering continuous output and an established supply chain. Nuclear satisfies it even more completely in the long term, combining extremely high energy density with sustained, uninterrupted generation. These characteristics matter because industrial systems require continuity, grids require stability, and advanced economies require predictability. Intermittent resources can contribute energy, sometimes very effectively, but they do not by themselves solve the problem of firm capacity for always-on infrastructure.

These constraints are no longer theoretical. They are already shaping how new infrastructure is being built. In Western Pennsylvania, the redevelopment of the Homer City site in Indiana County offers a clear example of what it now takes to power the digital economy.

The System, Rebuilt on an Old Platform

Construction is underway on a co-located system: a new natural gas power plant built alongside a hyperscale data center campus. The developer group is backed by Knighthead Capital Management, which has committed significant equity—reportedly over $1 billion—and controls interests tied to the former coal facility. By its own description, the overall commitment represents one of the largest private capital investments in Pennsylvania.

The logic of the site is not speculative. It is geographic, industrial, and historical.

The region sits atop the Marcellus Shale, one of the most productive natural gas basins in the world. Fuel, in other words, is local. That reduces both cost and supply risk, avoiding dependence on long-distance transport as a primary constraint.

The site also lies within the footprint of PJM Interconnection, providing access to an existing high-voltage backbone and a liquid wholesale market. Even with co-location, interconnection matters—for synchronization, backup, and the ability to sell excess generation. At Homer City, developers have indicated that a portion of output (including power not consumed on-site) will be sold into PJM, while a defined tranche is directed to the data center campus.

The land itself reflects an older industrial logic. The coal-fired plant that operated here for decades shut down in July 2023. Much of it has now been demolished, leaving behind what earlier generations built to move energy: substations, transmission corridors, rail access, and a large, already-disturbed industrial footprint. The project is not being built on empty ground. It is being built on an inherited system.

What It Takes—In Practice

A project like this is not a single build. It is a stack of interdependent systems:

Fuel. A new steel natural gas pipeline—roughly 5.8 miles—must connect the site to existing infrastructure, crossing streams, wetlands, and floodways. That triggers state permitting, environmental review, and conditions on construction practices to protect water quality.

Generation. A combined-cycle plant—gas turbines, heat recovery steam generators, steam turbines, cooling systems, emissions controls—must be engineered for continuous operation and rapid response.

Interconnection. Substations and high-voltage links tie the facility to the regional grid, even if the goal is to minimize reliance on it.

Load. The data center campus brings its own infrastructure: power conditioning, redundancy, backup generation, and large-scale cooling—often as complex as the plant itself.

Capital. Equity, debt, offtake structures, and fuel hedging must align to support multi-billion-dollar construction and decades of operation.

Permitting and time. Environmental approvals, public hearings, and agency conditions extend timelines and add uncertainty—particularly for pipeline work crossing sensitive waterways.

Seen together, the system is not merely large. It is integrated. Each layer must function for the whole to work.

Bypassing the Wires

What the configuration reveals is a shift in how the grid is being used. Instead of relying on distant generation, congested transmission, and a shared distribution network, the project compresses the supply chain. Generation sits next to load. Fuel is secured upstream. Transmission exposure is reduced. Reliability is internalized.

This is not the grid failing. It is the wires being bypassed.

For large, capital-rich users, the economic logic is straightforward. The most expensive part of the system is often not the power plant, but the network required to deliver electricity across distance and through thousands of connection points. By co-locating generation and consumption, developers can avoid some of the most costly and uncertain layers of the system—particularly transmission congestion and distribution infrastructure built for dispersed users.

The Question That Follows

But the fixed costs of the shared grid do not disappear. Poles still need maintenance. Transformers still need replacement. Substations still need upgrading. Storm damage still needs repair. Universal service obligations remain.

The question is who pays.

If hyperscale users increasingly self-provision to avoid rising infrastructure costs, those costs are redistributed across a smaller base of customers. Households and small businesses, who cannot easily bypass the system, remain tied to it. What begins as an engineering solution becomes a problem of cost allocation—and ultimately a political one.

Energy transitions are often described as substitutions. In practice, they are accumulations. New systems are layered onto old ones; new demands arrive before old infrastructure is retired; new technologies introduce new costs even as they promise efficiency.

Artificial intelligence does not change these realities. It intensifies them.

The machine endures. But who it serves—and who pays for it—is beginning to change.

Subscribe now

Share

Then wildfire risk met jury rooms—and it became a fight over who pays.

In the old utility world, the pitch was simple: steady and dull.

A regulated monopoly earns an allowed return, builds long-lived infrastructure, and pays a predictable dividend. Equity gets bond-like stability. Credit gets near-sovereign comfort.

Warren Buffett’s Berkshire Hathaway Energy (BHE) was built for that world.

Natural monopolies. Long-duration capital investment. Regulated returns.

In 2019, that assumption collided with reality.

PG&E — the largest utility in California — filed for bankruptcy after the 2017–2018 megafires turned contingent liability into an existential funding problem. The event mattered far beyond one company. It forced California to confront a brutal truth:

If wildfire losses remain uncapped, the grid becomes unfinanceable at regulated returns.

That bankruptcy is why the state rebuilt the wildfire model around certification, pooled backstops, and securitization — not to make fire risk disappear, but to keep capital markets open when it inevitably appears.

Then the wildfire era arrived — and it stopped being a California-only problem.

Why This Matters Now

The wildfire story is no longer just an environmental problem or a regional utility issue.

It is rapidly becoming a capital markets story.

Over the next decade the United States must invest trillions of dollars to electrify transportation, expand transmission networks, and harden aging infrastructure against extreme weather.

Those investments require one thing above all else:

cheap, reliable access to capital.

If wildfire liabilities make regulated utilities unfinanceable, the consequences extend far beyond California or Oregon.

They reach directly into:

• the cost of electricity

• the pace of grid expansion

• and ultimately the feasibility of the entire energy transition.

The Question Beneath the Entire Sector

Before diving into PacifiCorp, California, or the bond market, one question sits beneath the entire utility sector:

When a catastrophic wildfire occurs — who ultimately pays?

Ratepayers?

Taxpayers?

Insurance markets?

Equity investors?

Or bondholders higher in the capital stack?

For a century the answer was mostly invisible because disasters were manageable.

Wildfire risk changed that.

Today the western utility sector is increasingly priced around one uncertainty:

Does the next fire stay in equity — or migrate up the capital stack into credit?

1) PacifiCorp’s Reality: Mitigation Is Real — Perfection Is Not

PacifiCorp has not been asleep at the switch.

Its wildfire mitigation program spans multiple states and includes the full modern playbook:

• grid hardening

• covered conductors

• advanced relays

• weather monitoring systems

• vegetation management

• public safety power shutoffs (PSPS)

Yet exposure persists.

Because tail risk is not linear.

The scale of the legal exposure is already staggering.

PacifiCorp has reached roughly $2.2 billion in wildfire settlements and faces estimates of as much as $55 billion in potential claims tied to western US fires, according to reporting on the company’s litigation exposure.

Oregon’s 2020 Labor Day fires produced a major jury verdict in 2023 and ongoing class certification fights that could broaden the claim set.

Ratings pressure has followed. PacifiCorp’s credit rating has fallen to one notch above junk territory, a remarkable development for what was once considered one of the most stable regulated businesses in the country.

Financing flexibility has become, in part, a regulatory question.

This is the uncomfortable truth:

Risk management reduces probability.

It does not eliminate maximum loss.

S&P summarized the mismatch in a recent investor FAQ.

Investor-owned utilities were designed a century ago to deliver reliable service with modest equity returns (roughly 9–11%) and relatively high leverage (around 50% debt).

That capital structure was never built to absorb multi-billion-dollar catastrophe liabilities.

One extreme event can devastate credit quality.

Perfection is not possible. That is not ideology. That is math.

2) Why Berkshire Sold — and Why Manulife Is in the Deal

Seen through that lens, Berkshire’s move reads less like a philosophical pivot and more like risk management and balance-sheet defense.

When a conglomerate owns a regulated monopoly, wildfire liabilities are not just operational risks.

They threaten the consolidated fortress through contingent liabilities.

Selling a “cleaner” asset can function as liquidity management while larger legal exposures remain unresolved elsewhere.

Now zoom in on the buy side.

Portland General Electric’s logic is straightforward:

• acquire rate base

• acquire customers

• acquire long-duration infrastructure growth

Manulife solves a different problem.

How do you finance a large equity check without:

• over-levering the balance sheet

• diluting shareholders

• concentrating catastrophic exposure?

The answer is structural.

Capital markets will fund growth — but a large minority partner helps absorb first-loss volatility, ring-fence risk, and keep the broader enterprise financeable when the next event arrives.

Manulife is not a legal shield.

It is a capital stack design choice.

3) California’s Redesign: PG&E’s Bankruptcy Forced a New Model

California responded to the PG&E bankruptcy not by eliminating wildfire risk, but by engineering a financial architecture to contain it.

The redesign rests on three pillars.

I. Operating Certification

Utilities must file formal wildfire mitigation plans and obtain annual safety certifications.

Operational discipline becomes the gateway to financial protection.

II. A Pooled Wildfire Backstop

California created a large wildfire fund to prevent each fire season from triggering a capital-markets shutdown.

Ratings analysts have been explicit: for high-exposure utilities, a robust wildfire fund is central to maintaining investment-grade credit.

Without it, the capital stack cannot support uncapped catastrophe risk.

III. Securitization

This is the bridge.

Wildfire liabilities are converted into rate-recovery streams, financed through securitization structures in capital markets.

Instead of collapsing balance sheets overnight, catastrophe costs become long-duration regulated obligations.

California did not solve fire.

It solved finance after fire.

4) The New Tell: Credit Markets Have Set a New Floor

Was there a complete credit-market shutdown after PG&E?

No. But markets repriced the sector.

Spreads later compressed — but they remain higher than the pre-wildfire era for now.

The market moved in stages:

1. Shock repricing

2. Policy stabilization

3. A search for equilibrium

Western utilities now carry a wildfire premium.

Consider a recent Southern California Edison financing:

• Yield to maturity: 4.825%

• Spread: ~103 basis points over Treasuries

Credit markets are open.

But price based on a structured wildfire finance regime.

5) Equity Markets: Where the Risk Appeared First

Equity markets were the first place wildfire risk surfaced.

Some companies stabilized.

Others haven’t fully recovered.

The market began distinguishing between utilities with:

• containable wildfire exposure

• and open-ended catastrophe risk

6) The Human Flaw: Equity’s Normalcy Bias

Credit markets think in terms of term limits, collateral, and recovery hierarchy.

Equity does not.

Utility equities historically behaved like slow-compounding bond substitutes.

But wildfire liabilities do not arrive smoothly.

They arrive as step-function shocks:

wind event

ignition

litigation

jury verdict

Suddenly the steady dividend machine becomes a fight over who pays.

Which returns to the only question that matters:

Is the risk socialized or not?

And if it is — through whom?

Ratepayers.

Taxpayers.

Insurers.

Equity investors.

Policy is attempting to prevent two outcomes simultaneously:

• a grid starved of capital investment

• rate shock from citizens refusing to pay

That tension now defines the western utility sector.

7) The Forward Signal: Certification as a Credit Covenant

In California, wildfire safety certification is quietly becoming something resembling a credit covenant.

Approval of mitigation plans enables utilities to access:

• regulatory protections

• wildfire funds

• securitization structures

Without certification, those protections weaken dramatically.

Which leads to the most important regime shift of all:

The cost of capital is now tied directly to fire governance.

The Market’s Verdict

Utilities perceived to have containable exposure retain traditional multiples, while others trade at persistent discounts.

It prices it in:

• earnings multiples

• dividend yields

• and the probability that catastrophe migrates up the capital stack.

The Takeaway

Berkshire did not sell because utilities stopped being “steady and dull.”

It sold because a new variable entered the model:

uncapped contingent liability that refuses to respect regulated returns.

When that variable becomes large enough, an allocator build to last protects the fortress.

California’s response is now the template:

certification

wildfire funds

securitization

A financial architecture designed to keep capital flowing into a grid that must expand, harden, and electrify.

Credit markets reflect this reality.

First came panic.

Then repricing.

Now a search for an new equilibrium.

A premium that says:

“We will lend — but we are not pretending.”

And equity remains the crumple zone.

At least until the next fire forces the system to decide again who ultimately pays.

That is the underwriting now.

If you found this analysis useful, consider subscribing to The Rustbelt Reader.

This publication focuses on the intersection of infrastructure, credit markets, energy systems, and the long economic cycles shaping the American economy.

Utilities, grids, and capital markets may seem technical — but they determine how fast the country can build, electrify, and grow.

Subscribe now

Share

Leave a comment

At first glance, Berkshire Hathaway’s $1.9 billion sale of utility assets in Washington looks routine. Portland General Electric is buying wires, wind farms, a gas plant, and roughly 140,000 customers from PacifiCorp. Another regulated deal in a sleepy sector.

The detail that matters is the partner. Manulife Investment Management will take a 49% stake in the acquired Washington utility business.

That has inspired a clean, appealing narrative: pair a utility with a capital partner tied to insurance and forestry expertise, and you get a next-generation owner built for wildfire risk. Grid meets forest. Risk pricing meets land management.

It’s a satisfying story. It may also be misleading.

Look past the marketing and this transaction looks less like a visionary climate-adaptation partnership and more like liquidity management under legal stress—with important implications for how investors should think about utilities as an asset class.

A Distressed Sale, Not a Philosophical Pivot

Start with the seller. PacifiCorp has been explicit about why it is selling: liquidity pressure from wildfire litigation, particularly tied to Oregon’s 2020 Labor Day fires. Management has pointed directly to credit strain and the need to stabilize finances.

Buffett has also been unusually direct about what’s changing in the utility business. In Berkshire’s latest annual letter he wrote that “it will be many years” before the company knows the final tally of Berkshire Hathaway Energy’s wildfire losses—and warned the implications could be industry-wide. Some utilities, he suggested, may no longer “attract the savings of American citizens” and could be pushed toward a public-power model.

That framing matters because it changes how this deal should be read. Berkshire isn’t exiting Washington because the grid there is uniquely unmanageable. It appears to be selling a performing asset to raise cash while it fights a far larger legal battle next door. This looks less like a strategic retreat from the Pacific Northwest and more like an amputation to protect the balance sheet.

If this were purely about “fire country,” Berkshire could have walked away wholesale. Instead, it carved out cleaner assets to help fund the messiest exposure.

The Buyer’s Logic Is Financial First

On the buy side, the story is similarly prosaic.

PGE is buying rate base at a defined multiple, with dispatchable gas capacity alongside renewables, and a long runway for capital spending regulators are likely to approve—at least initially. This is classic utility math.

The Manulife partnership helps solve a simpler problem: writing a very large equity check without blowing out leverage or diluting shareholders. Whatever the truth is behind the narrative about forestry expertise, the immediate function of a 49% partner is to make the capital stack work.

That doesn’t make the deal reckless. But it does suggest that the “forest management” story may be overstated. Commercial timberland management and utility wildfire risk are not the same business. Forestry is centralized and contiguous. Utility vegetation management is fragmented, adversarial, and political—narrow rights-of-way crossing thousands of parcels, negotiated homeowner by homeowner, under constant regulatory scrutiny.

Which brings us to the most underappreciated flaw in the bullish case.

The ‘Geographic Firewall’ Myth

A common defense of this transaction is that Washington offers a safer legal regime than Oregon—that PGE is buying into a different liability universe. Look closely and that firewall dissolves.

Investors fixate on inverse condemnation because California made it toxic: if utility equipment starts a fire, strict liability can attach even without negligence. But Oregon’s PacifiCorp crisis didn’t require that doctrine. In the landmark 2023 trial over the 2020 fires, the jury rejected inverse condemnation claims outright—and PacifiCorp still faced catastrophic exposure through standard tort law: negligence, trespass, and nuisance.

That’s the real danger. Once a jury crosses the threshold into gross negligence, damages aren’t limited to replacing burned property. They expand into non-economic and punitive awards that can dwarf physical losses. The math becomes violent.

Washington does not shield utilities from that dynamic. Like Oregon, it evaluates wildfire liability under negligence standards. If a jury decides a utility should have de-energized lines, trimmed more aggressively, or acted sooner during a wind event, the equity is exposed to the same jury calculus. Washington even carries its own statutory multipliers—such as treble damages for timber trespass—that aggressive plaintiffs’ attorneys will inevitably test.

Strip away the borders and the regimes look structurally similar: if your wire sparks a fire, you are at the mercy of a jury.

Which leads to an uncomfortable conclusion. Berkshire didn’t sell Washington because the law is worse there. It sold Washington to raise cash to feed the legal beast and manage contingent liability in Oregon.

PGE and Manulife aren’t buying because Washington law protects them. They’re betting—explicitly—that under their management they can prevent ignition, because once the courtroom phase begins, state lines don’t cap the downside.

What This Means for Utility Investors

This deal quietly forces a rethink of how utilities should be underwritten.

Historically, utilities were treated as bond-like equities: capped upside, stable returns, and near-sovereign credit. Wildfire liability breaks that model because the shock is non-linear and instantaneous. A single event can erase years of earnings. Rate recovery lags. Politics intrudes.

Equity is the crumple zone. Investors are holding a capped-return asset with effectively uncapped downside. That is not a safe yield play.

Credit has held up better, protected by the assumption that regulators will never let a utility fail. But that protection may prove conditional. Once liabilities strain liquidity and ratings, borrowing costs rise, capital spending becomes harder, and the feedback loop turns vicious. At the extreme, risk is socialized—not because policymakers want to, but because the alternative is a broken grid.

That is the quiet lesson of this transaction.

Berkshire saw a tail risk large enough to threaten the fortress balance sheet and stepped back. PGE stepped in, with a partner, inside a tighter box, seems to be betting that prevention will prevent potential litigation.

For investors, the takeaway is blunt: Western utilities are no longer simple income vehicles. They are complex bets on environmental risk, tort law, regulatory behavior, and political tolerance for rate shock.

And the most important question is no longer how much rate base a utility can build—but whether the next fire stays an equity problem, or migrates up the capital stack and becomes a credit problem.

That is now the real underwriting.

Share

Leave a comment

“Energy is the only universal currency.”

— Vaclav Smil

If high-voltage transmission lines are the interstate highways of the grid, the substation is the off-ramp.

It’s where electricity slows down, changes lanes, gets checked for weapons, and then gets allowed into the city.

And right now, a growing share of America’s onshoring story is running into a hard truth:

You can announce factories.

You still have to step the voltage down.

This is a fast tour of what a substation actually does, why it becomes a chokepoint, and why Phoenix is the clearest case study of industrial policy meeting copper and steel.

Part I — What a substation actually is

A substation looks like a fenced junkyard because it’s built around one unforgiving fact:

Electricity is always trying to take the shortest path to ground.

Substations exist to make that path useful instead of catastrophic.

Every substation does three jobs:

• Transform voltage

• Route power

• Protect the grid

Miss any one of those, and the project isn’t real.

The transformer: the muscle of the grid

The transformer is the star.

It manipulates the relationship between voltage and current.

• Near generation, it steps voltage up — often into the 115 kV to 765 kV range — so power can travel long distances efficiently.

• Near cities and factories, it steps voltage down again to levels the next network can handle.

Order of magnitude matters.

A single large substation transformer might be 50–100 MVA — enough to serve tens of thousands of homes, or a small number of extremely power-hungry industrial or data-center loads.

Because the grid runs on three-phase power, substations use either one large three-phase transformer or three single-phase units banked together.

At these voltages, electricity is constantly trying to escape.

Current passes through bushings — conductors wrapped in ceramic insulation. When insulation fails, you get flashover: a violent arc to ground.

That’s why substations look overbuilt.

They are.

This is also why availability matters. Large power transformers now come with multi-year lead times. When CBRE talks about “power availability delays,” this hardware is what they mean.

AIS vs. GIS: same off-ramp, different packaging

Most substations are open-air yards (AIS):

• big footprint

• visible equipment

• cheaper

• needs land

Dense or fast-moving nodes increasingly rely on GIS:

• compact

• sealed enclosures

• less land

• more expensive

Firms like Burns & McDonnell have been blunt about this trade-off:

When schedules matter, GIS can be the only way to keep a project alive.

Same function.

Different geometry.

Same bottleneck.

Routing power: buses and feeders

Redundancy here allows operators to reroute electricity around failures—but adds complexity that puts substations on the critical path.

After voltage is stepped down (often 35 kV to 138 kV), power doesn’t just leave.

It enters the substation’s internal routing:

• Transformer = pump

• Bus = main header pipe

• Feeders = garden hoses going to neighborhoods and facilities

Utilities design redundancy here so power can be rerouted around failures.

Reliability requires options.

Options add complexity.

Complexity puts substations on the critical path.

That’s why Engineering News-Record now treats grid access — and substations in particular — as a front-of-project issue.

The napkin math: why we crank voltage so high

Losses scale like I²R.

Double current and losses quadruple (as heat).

So the grid avoids pushing amps.

It raises voltage.

Because P = V × I:

• Move 100 MW at 10 kV → ~10,000 amps

• Move 100 MW at 230 kV → ~435 amps
  (simplified single-phase equivalent; three-phase further reduces current)

That’s why high voltage exists.

It turns “melt the line” into “move power.”

Substations are the machines that perform that trick.

Protection: the “weapon check”

A “weapon” in the grid is anything trying to jump the rails to ground:

• lightning strikes

• short circuits

• equipment failures

If a crane clips a line or lightning hits a tower, breakers and relays try to localize the fault — one line, one bus section — instead of letting it cascade.

This is the weapon check.

And it’s why substations are chokepoints: they are conversion, routing, and defense.

Part II — Phoenix: cluster economics meets the grid

Phoenix isn’t landing one project.

It’s building a manufacturing cluster.

Green Street documents the multiplier effect:

• anchor announcements by TSMC, Intel, and LG

• 50+ suppliers

• 9+ million square feet of committed space

• much of it not yet operational

Here’s the thesis, plainly stated:

In Phoenix, the limiting factor on turning announcements into jobs is not land, labor, or buildings.

It’s how fast you can pour and wire new off-ramps.

This framing is now mainstream.

JLL calls it “power as the new real estate.”

CBRE notes that projects requiring new substations face materially longer timelines.

Different words.

Same bottleneck.

The outside-the-fence tell: “not yet”

This isn’t theoretical.

In Phoenix, the local utility has already started telling some large customers “not yet.”

Not because the desert ran out of land — but because substations and feeders can’t be built fast enough.

Engineering firms like Black & Veatch have described developers taking power infrastructure into their own hands — self-building substations — after utility timelines stretched multiple years.

When a real estate developer becomes a utility-scale power engineer just to stay on schedule, the off-ramp is full.

Loop 303 vs. Chandler: two load shapes, same constraint

• Loop 303 (West Valley): fewer, larger facilities → corridor stress

• Chandler / Gilbert (East Valley): many smaller nodes → network stress

Different geometry.

Same outcome.

Substations decide the pace of reality.

Phantom capacity

Green Street’s most important insight is timing.

Much of the “available” industrial power in Phoenix is already allocated to projects under construction.

On spreadsheets, it looks like capacity.

In reality, it’s spoken for.

That’s Phantom Capacity.

Musical chairs where the music hasn’t stopped — but the chairs are gone.

Electrical obsolescence

Industrial real estate used to be about location, height, docks, and labor.

Clusters add new filters:

• does the local substation have spare MVA?

• is there a path for a second feeder?

• is the transformer you need actually available — or stuck in a 2–4 year procurement window?

That’s physical obsolescence now.

Not old.

Electrically wrong.

And yes — the same step-down logic repeats all the way to the gray can on your street, until the grid finally hands you 120/240 volts.

The Phoenix lesson

Phoenix is what it looks like when a country tries to build a 21st-century industrial cluster on a 20th-century pace of substation construction.

Announcements move at the speed of politics.

Off-ramps move at the speed of steel, copper, and transformer lead times.

Thanks for reading

If this helped you understand how the grid actually works, subscribe.

How Shit Works is about the physical systems underneath the economy — the infrastructure, constraints, and bottlenecks that decide what’s real and what’s just a press release.

No hype. Just the machine.

Subscribe now

Leave a comment

Works Cited

• CBRE. “High Demand, Power Availability Delays Lead to Record Data Center Construction.” CBRE Insights, Sept. 14, 2023. (cbre.com)

• JLL. “Data Center Availability Crisis Deepens as Vacancy Hits Historic Low.” JLL Newsroom (North America Data Center Report – Midyear 2025), Aug. 18, 2025. (jll.com)

• Engineering News-Record. Gottlieb, Bryan. “Grid Access, Not Land, Emerges as Bottleneck for Data Center Construction.” Engineering News-Record, Dec. 18, 2025. (enr.com)

• Black & Veatch. “AI Data Centres: Black & Veatch on the Infrastructure Crisis.” Data Centre Magazine (whitepaper/interview feature), Aug. 7, 2025. (datacentremagazine.com)

• Black & Veatch. “Leading Data Center Provider Turns to Substations to Solve Fast Expansion Needs.” Black & Veatch (project/case study). (bv.com)

• Burns & McDonnell. “Beat the Clock: GIS Substations Are Your Ally for Tight Deadlines.” Burns & McDonnell Blog, Aug. 14, 2024. (blog.burnsmcd.com)

• TSMC. “TSMC Intends to Expand Its Investment in the United States to US$165 Billion.” TSMC Press Release, Mar. 4, 2025. (pr.tsmc.com)

• “Data Centers Continue to Push Phoenix Area’s Power Grid to the Limit.” KBTX, Nov. 11, 2025. (kbtx.com)

• Engineering in Plain Sight. Hillhouse, Grady. Engineering in Plain Sight: An Illustrated Field Guide to the Constructed Environment. No Starch Press, Sept. 2022, 264 pp.

• Green Street. Industrial Outlook — Phoenix Case Study (p. 20). (Subscriber research / report excerpt provided by author.)

You flip a switch and the room lights up like the universe owes you an explanation.

It doesn’t.

What you’re actually doing is tapping into a synchronized machine that spans a continent — a live system where supply and demand must match every single second, or the whole thing starts to wobble. The electrical grid isn’t just “wires.” It’s physics, history, control rooms, and a thousand quiet compromises that keep your life from going dark.

And if you live in the Rust Belt, that machine isn’t background scenery — it’s the plot.

This region was built by industries that don’t run on vibes. Steel, glass, chemicals, aluminum, machine tools — they were all power density businesses. When the grid was cheap and expanding, the Rust Belt could scale. When power got constrained, expensive, or politically tangled, the industrial base didn’t just “slow down.” It lost its oxygen.

The grid in one sentence: it’s a real-time balancing machine built on the logic of heat, distance, and stability — and the Rust Belt rises or falls on whether that machine has spare capacity.

The physics: why the grid looks this way

Electricity is easy to make. The hard part is moving it.

When power travels through a wire, some of it becomes heat. That loss is brutally simple:

Loss = I²R

(current squared × resistance)

It’s waste heat. You’re paying to warm up the sky.

If current doubles, losses quadruple. That is the first constraint the grid was built to obey.

Engineers get around it with another relationship:

Power = Voltage × Current

Step voltage up, and you can step current down while delivering the same power — meaning far less loss to heat. That’s why transmission lines run at extreme voltages. In the U.S., the biggest long-distance lines can go as high as 765 kV.

High voltage isn’t a flex. It’s how the math works.

Transformers are the quiet heroes here. They let the grid “change gears”: raise voltage for long-haul travel, then lower it again for safe use.

Rust Belt reflection: if you want modern industry to come home — EV supply chains, robotics, electric arc furnaces, data centers, advanced manufacturing — you don’t start with speeches. You start with the math of I²R and the physical reality of transformers.

The machine you’re connected to (from turbine to toaster)

The grid is a ladder that moves power from bulk production to fine delivery.

Most generation — coal, gas, nuclear, hydro, wind — ultimately turns a turbine. A rotating magnetic field induces electrical current. From there, electricity climbs into the transmission system, then steps down into local distribution, then steps down again into your building.

The “drop” is the part you recognize: the pole-top cylinder in older neighborhoods, or the green pad-mounted box in newer ones. That transformer is where the grid turns dangerous long-distance voltage into something that belongs in a kitchen.

If you want the grid in one line: big lines move bulk power, small lines deliver it, and transformers make both possible.

Rust Belt reflection: the factories and mill towns weren’t placed randomly. They clustered where power was reliable, fuel was available, and the network could deliver huge loads without blinking. Industrial geography is electrical geography.

Why transmission lines run in threes

Look at high-voltage towers and you’ll usually see conductors grouped in threes.

That’s three-phase power — three waves offset from one another so delivery stays nearly constant instead of pulsing. It’s more efficient for long-distance transmission and ideal for running large motors.

This reveals the grid’s DNA: it wasn’t built around phone chargers. It was built around steel mills, factories, and the heavy machinery of the 20th century.

You’re living inside that inheritance.

Rust Belt reflection: when we say “bring back industry,” we’re really saying: can we still run the kind of loads that built the Arsenal of Democracy — only now it’s server halls, advanced fabrication, and electrified production?

The grid’s real job: balancing, instantly

Here’s the part most people miss:

Electricity cannot be stockpiled in the wires.

It has to be used at the exact moment it’s generated.

When you turn on an air conditioner, a generator somewhere must work a little harder right now — or another load must ease off right now. That’s why grid operations are less like plumbing and more like aviation: constant monitoring, constant adjustment, tight safety margins.

If the balance slips, the system doesn’t merely get “a bit weak.” It becomes unstable. Voltage can sag. Frequency can drift. Protective systems trip lines to prevent physical damage.

A blackout is often the grid doing exactly what it was designed to do:

sacrifice parts to save the whole.

Rust Belt reflection: “reindustrialization” isn’t just a policy story. It’s an operations story. You can’t will a blast furnace, an EAF, or a data center into existence if the grid operator can’t reliably balance the load behind it.

How cascades happen (and why tight coupling cuts both ways)

Most major outages aren’t one thing breaking. They’re sequences.

A line trips — storm, overload, fire, equipment failure. Power reroutes to adjacent lines. Those lines heat up, overload, and trip. Generation and load fall out of sync. Operators and automated protections isolate segments to stop the bleeding. If too much trips too fast, you get a cascade.

This is the trade:

Tight coupling makes the grid efficient.

Tight coupling also makes it capable of failing quickly.

Rust Belt reflection: the same interconnection that lets power flow where it’s needed also means the grid is a shared battlefield of constraints. When the system is tight, everyone is competing for the same reliability margin.

The future: the grid becomes a conversation

The old grid was built for one-way flow: large plants feeding passive consumers.

The new grid is becoming two-way and volatile.

Solar is intermittent. Wind surges and fades. EVs add load — but can also act as flexible demand.

Meanwhile, data centers are arriving as a new class of hyper-industrial load — demanding flat, 24/7 reliability in a system increasingly powered by intermittent weather.

“Smart grid” isn’t just a gadget. It’s the inevitable evolution of the control layer: sensors, automation, forecasting, and fast response. The grid is becoming a cyber-physical system where information is as important as copper.

Rust Belt reflection: this is the new “mill siting” question. Not where is the river or where is the rail yard — but where can you get firm power, fast interconnection, and stable reliability? The next wave of industrial investment will cluster around those answers.

The Rustbelt takeaway

If the Rust Belt is serious about a new industrial century — advanced manufacturing, electric arc steel, robotics, data centers, electrified logistics — then the question isn’t “do we have ideas?”

It’s:

• Do we have power?

• Do we have interconnections that can actually clear new load?

• Do we have transformers and substations — the middle layer — in enough quantity?

• Do we have the capacity to run the machine without living on the edge?

Because the grid is the first industrial platform. Everything else rides on it.

And that’s the twist: the Rust Belt doesn’t just need “jobs.” It needs amperage. It needs transformers. It needs permission to connect. It needs infrastructure that can carry the next century’s load the way it carried the last one.

Next in How Shit Works: Transformers & Substations — the chokepoint hiding in plain sight.

If you want more of this series — ports, rail, waterways, steel, semiconductors — subscribe. I’m building a practical library of the systems that actually run modern life.

No fluff. Just the machine.

Subscribe now

Works Cited

Bottomley, L. J. (2022). Electrical Engineering for Everyone [Course Guidebook]. Chantilly, VA: The Teaching Company.

Hillhouse, Grady. (2022). Engineering in Plain Sight: An Illustrated Field Guide to the Constructed Environment. San Francisco, CA: No Starch Press.

Ressler, S. (2015). Everyday Engineering: Understanding the Marvels of Daily Life [Course Guidebook]. Chantilly, VA: The Teaching Company.