Data center capacity in the United States could reach 106 gigawatts by 2035. That number doesn’t land without context: the entire U.S. residential sector currently consumes roughly 1,500 TWh per year. AI infrastructure is on track to consume a material fraction of that. The math was always going to force a reckoning over who pays.
On March 5, 2026, that reckoning got a name. Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI signed the Ratepayer Protection Pledge at the White House, committing to “build, bring, or buy” new power generation for their data centers, fund the associated grid infrastructure upgrades, transmission lines, substations, the works, and negotiate separate rate structures with utilities so that residential customers aren’t absorbing the cost of AI’s power footprint.
President Trump had announced the initiative in his State of the Union address. The formal signing followed days later. Seven companies. One ceremony. No federal penalties for non-compliance.
The scale of the problem the pledge is trying to solve
The numbers behind this story explain why it landed in the White House rather than in a utility commission docket.
U.S. data centers currently account for an estimated 4 to 6 percent of national electricity consumption. The IEA projects global data center electricity demand will roughly double from 415 TWh in 2024 to 945 TWh by 2030, with AI as the primary driver. U.S.-specific projections put data centers at 12 percent of national consumption by 2028 under several modeling scenarios.
The strain is already showing at the grid level. Data center load forecasts accounted for $21.3 billion, 45 percent, of $47.2 billion in capacity costs across PJM’s last three auctions. PJM’s benchmark capacity price increased 833 percent between the 2024-2025 and 2025-2026 delivery years. Residential energy prices in the U.S. rose 6 percent in 2025, with continued increases projected through 2027 and 2028.
Those figures explain the political pressure. Communities and state legislators have been watching data center projects arrive, strain local grids, and leave utility commissions debating how to allocate the costs. Several projects have been cancelled or postponed because of that opposition. The Guardian’s reporting on the signing noted this dynamic directly: the pledge is, in part, a response to organized community resistance that was starting to slow AI infrastructure deployment.
What the pledge actually commits to
The White House text lays out four specific commitments. Signatories will build, procure, or purchase new power generation for their facilities. They’ll fund the grid infrastructure upgrades that connecting that generation requires. They’ll negotiate dedicated rate agreements with utilities and state governments, paying for power capacity whether they use it or not. And they’ll coordinate with grid operators to make backup generation available during emergencies while hiring and training local workers in host communities.
That’s a more structured set of commitments than the early reporting suggested. It’s not simply “we’ll pay our own bills.” It includes grid investment, rate negotiation architecture, and workforce provisions.
What it does not include is an enforcement mechanism. Multiple outlets covering the signing described the agreement as non-binding, with no federal penalties attached to non-compliance. The pledge relies on the reputational and political costs of defection, not legal obligation.
The fragmentation problem no pledge solves
Oregon has already passed state legislation to protect smaller ratepayers from data center power costs. Indiana, Georgia, Missouri, and California are all seeing active legislative or regulatory activity pushing for special electricity rate categories for large commercial users. That fragmentation was developing before the White House pledge. It continues after it.
The pledge’s scope is also limited to new data center builds. Existing infrastructure, and the costs it’s already generating for grid operators, isn’t addressed. Non-signatories aren’t covered. Anthropic, a leading AI lab, was absent from the signing entirely, reportedly due to complications involving a national security dispute with the Pentagon. The pledge’s coverage of the AI industry is incomplete on its face.
Microsoft, Google, Amazon, and Meta alone are expected to spend over $700 billion on capital expenditures in 2026, a 60 percent increase from 2025, with a substantial portion directed to new data center facilities. The companies signing this pledge are the companies driving the demand surge. Their commitment to self-fund the power infrastructure for that expansion is the pledge’s core value proposition.
What this means for infrastructure teams and CFOs
The practical effect of the pledge, if it holds, is a cost model shift. New AI data center projects built by signatories will negotiate their own power agreements directly with utilities and state governments rather than drawing from existing rate structures. That’s a different procurement model than most enterprise infrastructure teams have operated under.
It also sets a precedent that’s worth watching regardless of whether your organization signed anything. If the voluntary framework establishes a norm that large AI workloads carry their own power costs, that norm may eventually migrate into regulatory requirements, especially in states already pursuing rate separation legislation. CFOs building multi-year AI infrastructure budgets should be pricing power procurement as a direct cost line, not a utility overhead assumption, whether the pledge applies to them or not.
The enforceability gap is real. So is the scale of the problem the pledge is responding to. Both things can be true at once, and both shape how this plays out over the next 24 months.