Energy Policy Inversion

The utility was always going to build. The only thing the hyperscaler PPA changes is who pays.

The Department of Energy has spent the second half of 2025 and the early months of 2026 advancing a specific claim about the relationship between large data centers and American electricity prices. Energy Secretary Chris Wright has stated it from CNBC, Fox Business, an Iowa stump speech in Palo, and most recently from a Cato Institute graphic that has circulated widely in conservative energy policy circles. The claim, in its most compressed form, runs as follows.

The argument has three working parts. First, that hyperscaler operators (Microsoft, Google, Amazon, Meta, and their second-tier peers) fund their own generation and transmission upgrades through power purchase agreements and a forthcoming "electricity auction" mechanism, preventing cost socialization onto residential ratepayers. Second, that high load factor industrial customers running 24/7 spread fixed grid costs across more megawatt-hours and mechanically reduce per-kWh average cost. Third, and most aggressively, that hyperscaler capital is what gets new dispatchable capacity built, the Three Mile Island restart, the Constellation-Microsoft SMR commitment, the Amazon-X-Energy deal, the broader gas combined-cycle wave, and that this capacity would not otherwise exist.

This third leg is where the claim collapses. The case that collapses it most cleanly is Texas.

I. The Captive Service

The argument requires the reader not to think about how regulated utilities actually make money. So let's think about it.

A regulated investor-owned utility delivering electricity to retail customers operates under cost-of-service regulation. The state public utility commission sets a rate base, the depreciated value of approved capital investments, and authorizes a rate of return on that rate base. Authorized returns on equity currently sit between roughly 9.0 and 10.5 percent across most jurisdictions. The utility recovers its operating costs through rates and earns its return on every dollar of approved rate base.

This is not a market business. It is a financial product secured by a captive customer base. Electricity is a captive service. No residential customer can opt out of consuming it. No business can choose not to be served by its local distribution utility. The product is not at risk of going unsold.

Under this structure, the utility's commercial problem is not "will customers buy." The commercial problem is "what is the largest prudent capex program the PUC will let me put in rate base this cycle." Capex is the earnings mechanism. Every approved dollar generates authorized ROE for the next 30 to 40 years over the asset's depreciation life.

Economists have known about this incentive structure for sixty years. The Averch-Johnson effect, named for the 1962 paper by Harvey Averch and Leland Johnson, is the formal demonstration that rate-of-return regulation produces over-capitalization. The utility prefers capital-intensive solutions to operating-cost-intensive ones, because it earns return on capital but only recovers operating cost. The utility would rather build a new substation than tune the existing one, rather build new transmission than negotiate a demand response program, rather build new generation than enroll customers in time-of-use pricing.

The relevant implication for the Wright claim: utilities are not investors waiting for someone to guarantee demand. They are investors maximally seeking approval to deploy capital, on the basis of forecast demand that they themselves prepare for their own PUC filings. The load forecast is evidentiary fuel for the capex program, not a marketing risk to be hedged with offtaker contracts.

The empirical record matches the theory. Investor-owned utility annual capital expenditures, tracked by the Edison Electric Institute, have grown from roughly $60 to $70 billion in the mid-2000s to over $170 billion in recent years. The trajectory is smooth. It does not exhibit a hockey-stick at any point coinciding with hyperscaler entry. The capex was happening anyway.

But this is an inferential argument. A more direct test is available.

II. What the Claim Requires

For Wright's "unlock" claim to be true, the following counterfactual must hold: in the absence of hyperscaler offtake commitments, generation capacity at the scale and composition now being added would not be financed and built.

This is a falsifiable statement. The test: does there exist a market structure in which large-scale generation capacity has been added at significant volumes without hyperscaler PPAs serving as the financing backbone?

If such a structure exists, and if the capacity addition under that structure was large and sustained, then Wright's claim is disproven by existence. No counterfactual modeling required. No elasticity estimates required. Just point at the working example.

The working example is the Electric Reliability Council of Texas.

III. The Falsification

ERCOT is the cleanest test case in the United States for the proposition Wright is advancing. It is structurally different from every other regional grid in two ways that matter.

First, ERCOT operates an energy-only wholesale market. Until the very recent Performance Credit Mechanism reform, approved by the Public Utility Commission of Texas in late 2023 and still being implemented, ERCOT had no capacity market. Generators in PJM, MISO, ISO-NE, and NYISO earn revenue from both energy sales and from capacity payments designed to compensate them for being available even when not dispatched. ERCOT generators historically earned revenue only when they actually produced energy, plus modest ancillary service revenue. This means ERCOT generators bear scarcity risk that generators in capacity-market regions do not.

Second, generation in ERCOT is overwhelmingly merchant. The 1999 restructuring of the Texas electricity market under Senate Bill 7 unbundled generation from regulated transmission and distribution utilities. Generation was opened to competitive entry. The traditional regulated utility rate base earnings model does not apply to ERCOT generators. They do not earn an authorized ROE on capex. They earn whatever the market delivers.

These two facts mean ERCOT is the most market-exposed major grid in the country. Generators take demand risk and price risk. There is no rate base shelter, and historically no capacity payment floor. If Wright's claim were correct, ERCOT should show the cleanest demonstration of the principle, since generation in Texas faces the most acute commercial pressure to require offtaker commitments before committing capex.

What actually happened in ERCOT bears no resemblance to Wright's framework.

From the early 2000s through 2024, ERCOT added more than 40 GW of wind generation capacity, making Texas the largest wind-producing state in the United States by a factor of roughly three over the second-place state. Utility-scale solar grew from negligible levels in 2018 to producing 45 TWh of energy in the first nine months of 2025, a fourfold increase since 2021. Battery storage grew from 833 MW at the end of 2021 to over 6.5 GW by 2024. Natural gas combined-cycle and peaking capacity was added throughout the same period. Total ERCOT installed capacity now exceeds 145 GW.

This buildout happened without hyperscaler PPAs serving as the financing structure for the vast majority of the capacity. Two facts make this concrete.

The first concerns financing structure. The dominant offtake mechanism for ERCOT wind has not been a traditional power purchase agreement with a corporate or utility offtaker. S&P Global Market Intelligence has documented that approximately two-thirds of ERCOT wind capacity is hedged through bank fixed-volume price swaps rather than physical PPAs. The project owner sells energy into the wholesale market as a merchant, while a separate financial swap with a bank counterparty fixes the project's revenue per megawatt-hour of expected generation. There is no offtaker buying physical power. There is a bank providing a financial hedge.

This structure is the cleanest possible falsification of the unlock theory. The financing did not require a Microsoft signature. It did not require a Google PPA. It did not require an industrial offtaker committing to take physical power. It required a bank willing to provide a hedge against wholesale market exposure. Tens of gigawatts of generation got built that way.

The second concerns policy mechanism. Texas wind was driven by a combination of three factors, none involving hyperscaler demand. The federal Production Tax Credit provided a per-kWh payment for the first ten years of operation. State Senate Bills 7 and 20, passed in 1999 and 2005 respectively, established and expanded the Texas renewable portfolio standard, mandating 5,880 MW of renewable capacity by 2015 and continuing expansion thereafter. The Competitive Renewable Energy Zone (CREZ) transmission build-out, completed between 2008 and 2014, deployed roughly $7 billion of transmission from West Texas wind resources to load centers, paid through a ratepayer surcharge. These three mechanisms, federal tax credit, state RPS mandate, and ratepayer-funded transmission, were collectively sufficient to drive the largest wind buildout in American history.

Hyperscaler load did not become a material category in ERCOT until roughly 2022. By that point, ERCOT already had more than 35 GW of installed wind and 10 GW of utility-scale solar. The buildout preceded the framing.

The implication for Wright's argument is exact. The largest single-state generation expansion in modern American history happened in a market without traditional capacity payments, without rate-base regulation on the generation side, and without hyperscaler offtake. The capacity was built because wholesale price signals, federal tax credits, state policy mandates, and ratepayer-funded transmission were collectively sufficient.

The claim that hyperscaler PPAs are the financing structure required to unlock new generation is falsified by the existence of ERCOT's pre-hyperscaler buildout. The disproof is empirical, not theoretical.

IV. The Predictable Counters

Three counter-arguments deserve attention before they are deployed.

The first is that ERCOT received subsidies, so it was not a "pure market" test. This is true and irrelevant. The Wright claim is not that generation gets built without any policy support. The claim is that hyperscaler PPAs specifically are the unlock mechanism. ERCOT demonstrates that other support structures (tax credits, mandates, ratepayer-funded transmission) are sufficient. The existence of those alternatives is itself the disproof.

The second is that ERCOT is now experiencing capacity adequacy strain, and that the Performance Credit Mechanism was approved specifically because the energy-only structure failed to procure adequate firm dispatchable capacity. This is also true and also irrelevant. The Wright claim is about whether capacity gets built at all without hyperscaler offtake. ERCOT proves it does. The question of whether the buildout's composition (heavy in renewables, lighter in firm dispatch) is the right mix is a separate composition argument, addressed below.

The third is the strongest version of the counter and gestures at a real composition question. The steelmanned Wright claim would be: hyperscaler PPAs are uniquely positioned to finance new dispatchable firm capacity (nuclear, gas) because hyperscalers themselves demand 24/7 firm power. This narrows the original claim considerably. It is no longer "data centers drive down electricity prices" as a general proposition. It becomes "hyperscaler offtake is required to finance new firm dispatchable capacity at the composition the grid needs."

Even at this narrowed claim, the falsification holds. ERCOT added more than 25 GW of natural gas combined-cycle and peaking capacity over the 2000 to 2020 period without hyperscaler offtake. The capacity was financed by merchant developers, by independent power producers, and by utility affiliates operating in the competitive market, all on the basis of wholesale price signals and ancillary service revenue. Plant Vogtle Units 3 and 4 in Georgia, the most recent new nuclear capacity completed in the United States, were built under traditional regulated cost-of-service ratepayer financing with no hyperscaler involvement. The composition of the buildout responds to the price signals on offer. When firm dispatch is scarce, scarcity prices spike, and developers respond. When the price signal favors intermittent generation, intermittent generation gets built. The market mechanism works. Hyperscaler PPAs are not a precondition for it functioning.

V. The Political Function

If the claim is falsifiable, and if it is falsified, then the question becomes: why is it being made?

The most charitable answer is that the Department of Energy is attempting to manage a political problem. Residential electricity prices are rising. The Consumer Price Index electricity component grew faster than overall inflation through 2025. PJM's 2025-2026 base residual capacity auction cleared at $269.92 per MW-day, roughly nine times the prior year's clearing price, and PJM itself attributed the spike to load growth forecasts dominated by data centers. Senator Katie Britt acknowledged in her own questioning of Wright that wholesale prices surged 250 percent in data-center-heavy regions. Voters are noticing. The Energy Secretary needs an answer that places blame somewhere other than the administration's signature industrial policy initiative (the AI infrastructure buildout) and the administration's preferred fossil generation mix.

The answer is to argue that data centers are not the cause of rising prices but the solution to them. The argument requires turning the actual causal direction inside out. Rising prices are real and caused, in significant part, by hyperscaler-driven load growth. The argument restates this as: hyperscaler-driven load growth is what will eventually lower prices through new supply. The same load that is causing the price increase is repositioned as the eventual fix for it.

The less charitable answer is that the framing serves a specific cost-allocation outcome. Under existing FERC and state PUC cost-allocation methodologies, transmission and generation upgrades caused by data center load growth are largely socialized across the ratepayer base. Standard load-share allocation methods divide the cost of system upgrades among all ratepayers in proportion to their share of total system load. A data center taking 500 MW from a particular load zone causes a transmission upgrade that gets allocated across all customers in that zone, including residential customers who did not cause the upgrade.

The "Ratepayer Protection Pledge" announced in 2025 is voluntary. The "electricity auction" mechanism that would force hyperscalers to bear cost-causation has not been defined. The FERC rejection of the AEP-Talen-Amazon co-location amendment at Three Mile Island in November 2024 demonstrates that when regulators look at the actual mechanism by which hyperscalers attempt to escape system cost allocation, they recognize the cost-shifting and shut it down.

In this context, the Wright framing has a specific function. If hyperscalers are characterized as benefactors of the grid (creating supply that wouldn't exist) rather than as large industrial customers (causing demand that requires capacity expansion), the political case for socializing their costs onto residential ratepayers is much easier to defend. The Cato slide is not making an argument for an audience of electricity economists. It is producing a legitimacy artifact for an audience of voters, who are being prepared to accept rising rates as the necessary cost of an investment that will eventually pay them back.

The investment, in the framing, is not a cost-allocation outcome favoring concentrated capital. It is a national infrastructure program. The data center is not a customer. It is a benefactor.

VI. Why the Pledge Stays Toothless

The reader who has accepted the falsification will reasonably ask the next question. If the framing is wrong, is the Ratepayer Protection Pledge the actual protection? Wright is on record that hyperscalers "very much want" to pay for necessary infrastructure upgrades rather than have those costs "socialized onto the backs of ordinary American ratepayers." Voluntary or not, does that not address the cost-shifting problem?

It does not. To see why, consider the building you live in.

Fourteen new tenants move in paying higher rent than you do. At lease renewal, the landlord raises your rent. The landlord points to the building improvements made to accommodate the new tenants. The landlord does not, and is not required to, use the higher rent the new tenants pay to subsidize yours. The differential becomes profit. Your only protection against the rent increase is rent control, which is either binding or it isn't.

That is the polite version. The actual mechanism in utility ratemaking is sharper. The new tenants do not pay higher rent than you do. They pay less per square foot, on a discount tariff structure built decades ago for steel mills and aluminum smelters that could relocate to other states if rates rose too high. Hyperscalers inherited that discount even though they cannot relocate the way the discount structure presumes. Dominion's data center tariff class has an effective per-kWh rate well below the residential class rate. The new tenants get the volume discount built for a different kind of tenant who could actually leave. The building gets rebuilt around their requirements at your expense. And there is no rate-design mechanism by which any of the differential they pay flows back to you, because cost-of-service ratemaking has no cross-class redistribution mechanism.

This is the architectural problem the Ratepayer Protection Pledge cannot solve. The pledge can promise that hyperscalers will pay for "their" upgrades. It cannot redirect the differential to reduce what residential customers pay, because the existing regulatory structure has no mechanism for that redirection. Whatever excess revenue is generated by industrial customer rates becomes shareholder earnings under the authorized ROE, not residential bill credits. The pledge is asking voters to imagine a redistribution the system is not built to perform.

There are at least six structural reasons the pledge stays toothless, and they compound.

The first is jurisdictional. Retail rate-setting is state PUC authority under Federal Power Act Section 201. The federal executive has no direct enforcement mechanism over retail rates. DOE can pledge whatever it wants. The rate case is in Richmond, Columbus, Springfield, not Washington. The pledge is structurally incapable of reaching the rate case.

The second is the chosen instrument. "Pledge" is the policy form. A rule goes through APA notice-and-comment, gets enforcement provisions, becomes judicially reviewable. A pledge gets a press release. The decision to use a pledge as the vehicle is the decision not to give it teeth. The form is the substance.

The third is regulatory capture by design. State PUCs are funded by utility assessments and staffed by people who rotate between PUC and utility employment. Their structural incentive is to find a defensible compromise that allows the utility to recover its costs. Real teeth would require PUCs to act against utility cost recovery at scale, which they structurally do not do.

The fourth is administration coalition geometry. Hyperscalers have direct administration access through the Sacks, Andreessen, Musk, Thiel, and Vance network. Any teeth proposal routes through those channels before it sees a rulemaking docket. The teeth get filed off in the room before they are announced.

The fifth is the internal contradiction the framing exists to manage. Conservatives want lower energy prices for voters and unconstrained AI infrastructure buildout. The Wright framing reconciles the contradiction by claiming both are simultaneously true. Teeth would force the administration to choose. They will not choose. They will keep the framing.

The sixth is technical complexity as defense in depth. Cost-causation in a meshed grid is genuinely difficult to allocate cleanly. Load-share allocation is the standard method because it is tractable, not because it is correct. Any teeth framework would have to define what counts as hyperscaler-caused cost versus system-benefit upgrade, which lawyers will fight in rate cases forever. The complexity is itself a defense against enforcement.

The pledge does not fail to get teeth by accident. It was designed without teeth, and the design will be defended on each of these six grounds depending on who is asking.

There are conversations the administration wants to have about this. Charge causation directly to the new customer. Raise all rates and subsidize residential back. These are the reformist options that work in principle inside the existing investor-owned utility framework. They require political conditions that do not currently exist and a pledge framework specifically designed to never accumulate them. They are the conversation the administration wants to have because they preserve the architecture and produce indefinite delay. The conversations the administration does not want to have are about whether the architecture itself is the problem. Those are for another piece.

VII. The Cost Allocation Reality

The honest description of what is happening is this. Utilities and hyperscalers are co-beneficiaries of a load-growth narrative that makes both sides' preferred outcomes more approvable at the public utility commission. The utility wants rate base growth and earns 9.0 to 10.5 percent on every dollar of approved capex. The hyperscaler wants firm low-cost power, queue priority for interconnection, expedited siting, and political cover for the cost-allocation rules that currently socialize their causation onto residential ratepayers. The PUC wants reliability, a defensible rate-case record, and political cover for the rate increases that follow.

None of these parties has an interest in stating the cost-allocation question plainly. The narrative that hyperscalers are unlocking capacity that wouldn't otherwise exist serves every interest at the table except the residential ratepayer's. The ratepayer is not at the table. The Cato graphic is part of the project of ensuring the ratepayer does not show up.

The utility was always going to build. The PUC was always going to approve substantial capex. The load forecast would have grown regardless. The hyperscaler PPA does not change the build. It changes who pays for it. The framing exists to obscure that fact.

The falsification matters not because it embarrasses Wright. It matters because it strips away the legitimating argument for a cost-allocation outcome that is currently being implemented in real time, in real rate cases, in real load zones, in real residential bills.

The Cato slide is the slogan. The rate case is the policy. The ERCOT record is the disproof. The residential ratepayer is the one paying the bill.