The AI Power Bargain: Congress Takes on Who Pays for the Grid

On September 30, 2026, Senators Sheldon Whitehouse (D-RI), Shelley Moore Capito (R-WV), Martin Heinrich (D-NM), and Mike Lee (R-UT) introduced the Bipartisan American Affordability and Jobs Act of 2026. An ambitious package of proposed new laws numbering over 400 pages in total, the Act is intended to reform and increase the efficiency of the federal environmental review and permitting process.

One of the many issues that the Act is designed to address is the impact that our nation’s increased use of artificial intelligence has had on energy demand. AI is the main reason why US electricity demand has been steadily growing since 2000 after more than a decade of flat demand. AI computing models need GPU-packed servers that require far more power per rack than traditional servers. Meanwhile, the rapid development of large-scale AI data centers is placing unprecedented demands on electric power grids.

The draft bill’s most direct response to AI-driven electricity demand is found in Section 2107 entitled “Ratepayer Protection,” which would change how large data-center loads pay for power system capacity and how they obtain service. It covers facilities used for AI training or inference (two main stages of AI model workflows), as well as facilities used for cloud computing, data hosting, and cryptocurrency mining. The Act does not regulate AI models or uses of AI; it only regulates their electric load.

Under the Act, large data centers would bear more of the cost of serving those centers. Proposed Federal Power Act § 228 defines a “covered load” as electric load that "(A) is attributable to 1 or more data centers or high-density computing facilities located at a single site, or at multiple sites under common ownership or control;” and ‘‘(B) equals or exceeds 20 megawatts or, in the case of data centers or high-density computing facilities developed in phases, is reasonably expected to equal or exceed 20 megawatts at full buildout." The bill would prohibit allocating the incremental cost of serving that load to other customers (e.g., retail energy consumers); require recovery of the full incremental cost even if the customer stops buying electricity before those costs are recovered; and require financial assurances or contributions before necessary generation, transmission, or distribution construction or upgrades. For a qualifying AI data-center project, this could increase upfront commitments and the cost of abandoning or downsizing a project.

The Act also permits states to impose additional conditions on covered loads and recognizes state authority to charge covered loads more than embedded average and incremental costs, credit excess revenue to other retail customers, and use competitive solicitations to allocate service. A state could condition service on the load securing sufficient new generation or accepting limits on when or how much electricity it withdraws. It could also give covered loads less favorable rates or service priority than other large industrial or commercial customers. These are options for states, not conditions that the Act automatically imposes on every data center project. However, in practice, a large AI data center developer might need to arrange new supply, accept interruptible or otherwise limited service, or compete for available capacity rather than assume a conventional firm services arrangement. 

These core provisions of the Act reflect that there has been in recent years a real paradigm shift regarding AI development. Two years ago, the question was whether the grid could handle AI; now AI’s growth is capped by what the grid can deliver, and Congress is legislating around that fact. Today, power availability is constraining AI growth, not the other way around. The Act attempts to remove that constraint by offering a balanced approach. It tries to thread the needle by making data centers cover their own costs, so retail consumers and U.S. households do not foot the bill for data center expansion, while fast-tracking the supply side.

The Senate is scheduled to take up the Act following the midterm elections with a cloture vote on the motion to proceed for the Act expected to take place on November 9, 2026. If the bill passes and this balance holds, AI will pay more of the bill for the grid, but importantly it will also get the expansion power it needs faster.