The AI Power Deal Needs a Receipt

AI infrastructure now has a public-accounting problem: who pays when a proposed data center requires new power plants, substations, and transmission lines, then uses less electricity than planned or never opens?

The White House expanded its Ratepayer Protection Pledge on July 23. The administration says the coalition now includes more than 200 utilities, cooperatives, data-center developers, and other organizations, along with participating governors and large technology companies. The central claim is that large data centers will build, bring, or buy the power they need and cover the infrastructure required to deliver it. The White House says participating organizations cover 80% of power delivered to U.S. homes and businesses and 263 million people. Those figures and cited customer savings are administration claims.

The Associated Press described the pledge as voluntary and reported that its effect on household bills remains uncertain. That distinction matters. A signature can establish direction. It cannot show whether a family was protected from a particular upgrade cost.

The useful evidence will appear in slower, less visible places: utility tariffs, interconnection studies, financial guarantees, minimum-payment terms, state commission orders, and monthly bills. If a project causes a utility to build new capacity, the contract needs to explain what happens when the project arrives late, reduces its requested load, or closes. Otherwise, households can inherit infrastructure built around demand that did not materialize.

Federal and state responsibilities divide the verification task. In June, the Federal Energy Regulatory Commission directed six regional grid operators to justify or reform their rules for large loads. FERC asked them to address transmission-cost shifting, cost transparency, flexible-load service, co-located generation, and connection studies. Its materials are explicit that state regulators retain authority over retail electricity rates and retail cost allocation. A national pledge has to survive many regional markets, utility structures, and state proceedings before it changes any specific bill.

Texas shows why request quality belongs in the same conversation. ERCOT reports more than 438,000 megawatts of large-load requests, nearly 89% attributed to data centers. Its new Batch Zero process will study qualified projects of 75 MW or more together, with classifications expected in August. ERCOT also cautions that many requests will never become operating facilities. That queue total can shape planning, but it does not represent a construction forecast.

The demand range is still consequential. Berkeley Lab’s 2025 update models data centers reaching 11.8% of U.S. electricity consumption in 2030 in its reference case, with scenarios spanning 9.5% to 15.3%. These are modeled estimates built from equipment, utilization, and cooling assumptions. They show the scale of the planning problem while preserving uncertainty about the outcome.

For communities, regulators, and operators, the practical unit should be a project-level receipt showing who pays what. That record would include requested and contracted load, required upgrades, who financed them, minimum-payment obligations, exit terms, curtailment rules, public incentives, local water commitments where relevant, and the rate effects eventually observed. It should update when the project changes.

That receipt would make disagreement more useful. A data center may bring tax revenue, construction jobs, digital capacity, and new generation to an area. It may also create land, water, reliability, or affordability pressure. Those effects vary by place and project. A shared, inspectable record gives supporters and critics something firmer than a press release or a queue total.

H.R. 9340, introduced in June, would require state authorities and nonregulated utilities to consider standards that recover full incremental upgrade costs from large-load customers of at least 100 MW and require financial assurances before upgrades begin. It remains introduced legislation. Congress has not enacted it.

Verification bottleneck

Verification is becoming the scarce institutional function.

What moved faster: Data-center proposals and power requests expanded faster than conventional grid studies, generation buildout, and public cost records.
Who now has to verify: Utilities, grid operators, state commissions, consumer advocates, local governments, developers, and residents need to trace project-driven costs from request through operation.
Where the bottleneck sits: Voluntary commitments cross federal transmission rules, state retail regulation, utility contracts, and local permitting, with no single record showing the full cost allocation.
What to watch next: Large-load tariffs, financial-assurance requirements, FERC regional filings, ERCOT’s August Batch Zero classifications, and evidence that promised household protections appear in actual rate outcomes.

Opportunities

Where value may appear for builders and operators:

A public large-load receipt generator that turns tariffs, commission orders, and contracts into a consistent project record.
A queue-quality tracker that distinguishes a request, a studied project, contracted load, a construction start, and an operating facility.
A local due-diligence service that maps power, water, tax, workforce, and exit obligations before a vote or permit decision.
A bill-impact monitor that compares promised rate protection with approved costs and observed customer rates over time.

These are research and operating ideas, not legal, engineering, financial, policy, or investment advice.

Count the infrastructure promise only after the cost allocation can be inspected.

The AI Power Deal Needs a Receipt

Verification bottleneck

Opportunities

Sources