Residential electricity cost 15.9 cents per kilowatt-hour in January 2025. By December it had reached 17.2 cents. Then, on March 4, 2026, seven of the most profitable companies in human history gathered at the White House and promised it would not happen again.
The promise is voluntary. The bills are not.
The pledge that Amazon, Google, Meta, Microsoft, xAI, Oracle, and OpenAI signed commits them to build dedicated power generation, cover infrastructure upgrade costs, and share backup power during grid stress. President Trump announced that prices in communities would "actually come down, and very substantially." Ruth Porat of Alphabet called it an affirmation of Google's "long-held commitment to protect energy affordability." These are the kinds of statements that sound reassuring precisely because they cannot be checked.
Who Covers the Gap Between a Pledge and a Law
Here is what is already on the books, before any of this week's commitments take effect. In the PJM Interconnection territory spanning seven eastern states, ratepayers face $3.1 billion in grid expansion costs tied to data center demand: $647 million for Pennsylvania, $637 million for Illinois, and $1.4 billion for Virginia alone. Those costs exist right now, under existing utility rate structures, regardless of what was signed at the White House.
The pledge does not touch those numbers. It addresses future costs through future negotiations with utilities, structured through the very rate-setting process that produced the $3.1 billion figure in the first place. Companies committed to paying "regardless of usage," which sounds generous until you realize that usage-based billing is exactly how utilities currently avoid assigning grid expansion costs directly to the parties driving them.
To be fair to the companies involved: some, including Microsoft, had already made binding commitments to cover interconnection upgrade costs before this pledge existed. That prior behavior is real, and it matters. But voluntary precedent is not policy. It is the argument companies make to prevent policy from forming around them.
Eric Schmidt, just two days after the signing, urged tech firms to co-locate data centers with their own energy sources rather than drawing from the public grid. That is the structurally honest solution. It removes the ambiguity about who owes what by removing the shared infrastructure from the equation entirely. Only 5 gigawatts of the 16 gigawatts expected online in 2026 are currently under construction. The grid buildout required to serve the rest will run through public utility commissions, meaning ratepayers, unless something legally enforceable stops it.
What the Pledge Buys That the Companies Actually Want
A voluntary pledge signed at the White House serves one underappreciated function: it creates a political record that companies can cite when state regulators propose mandatory cost-allocation rules. "We already committed to this voluntarily" is a powerful argument against binding regulation. It shifts the default from "prove you covered costs" to "assume good faith unless proven otherwise." That shift is worth considerably more than whatever grid expansion costs the companies might absorb.
AI data centers consume 4 to 6 percent of U.S. electricity today, heading toward 12 percent by 2028. The companies spending $350 billion in capital expenditures this year have every financial reason to keep their infrastructure costs socialized across utility ratepayers for as long as the regulatory environment allows them to.
State public utility commissions should not wait for voluntary pledges to fail before acting. They should require direct cost-allocation agreements, signed before grid expansion approvals, that assign data center interconnection costs to the companies requesting the capacity. Not as a condition of the pledge. As a condition of the permit.
The pledge bought goodwill. The ratepayer protection part comes later, if anyone makes it mandatory.
