Who Pays for the Power Behind American AI?

1 The fight hiding inside the AI boom

Every AI model runs on electricity, and the facilities housing it — data centers, per Wikipedia buildings of IT infrastructure for processing and AI training — consume power at a scale utilities were not built for. The question is who pays: the owners, or the customers who share the grid. The White House took a position in March 2026: its national AI legislative framework, per the release, states ratepayers should not foot the bill for data centers and calls on Congress to streamline permitting so they can generate power on site.

2 The current default: socialized costs

Under traditional utility regulation, large new loads are folded into the general rate base — the cost of new generation and transmission is spread across all customers. Applied to AI data centers, that logic breaks: beneficiaries are concentrated, costs diffuse. A residential bill can rise to fund infrastructure serving a facility its owner will never use. That dispute is playing out in state utility commissions nationwide.

3 What the framework would actually change

Two proposals carry the load. Ratepayer protection: data centers should bear the costs they cause, meaning commissions adopting tariffs that assign infrastructure costs to the facilities triggering them. Supply: streamlined permitting so operators can build on-site generation — gas turbines, fuel cells, behind-the-meter renewables — reducing load on the shared grid. The release describes both as calls on Congress: the framework is a proposal, not enacted law.

4 How costs would be measured

Allocation runs on measurement, and the contested number is cause. Data centers draw continuously: flat demand is cheaper to serve per megawatt, but a continuous multi-hundred-megawatt load still forces generation buildout that would not otherwise occur. Under a beneficiary-pays tariff, the auditable quantities would be the incremental megawatts and capital cost attributed to the facility; under today's default, they are spread into general rates, visible only through rate-case filings.

5 The on-site generation wager

On-site generation would insulate neighboring ratepayers from AI's demand but shifts siting, emissions, and reliability burdens onto the facility itself — and depends on permitting that is currently slow, which is why the framework pairs protection with streamlining. The wager pays off only if streamlined permitting produces dispatchable power faster than the grid expands to serve the same load.

6 What is proposed versus what exists

Enacted today: state utility regulations, existing tariffs, and the general rate base. Proposed: the federal ratepayer-protection principle, permitting reform, and any uniform national standard — all awaiting congressional action, and contested by states that argue energy siting is theirs to regulate.

7 Bottom line

The framework's answer to who pays is: the beneficiary, enforced through tariffs that assign AI-driven costs to AI facilities, with on-site generation to shrink the shared bill. None of that is enacted. Until it is, the meter runs on whoever state commissions say it runs on.