On March 5, President Trump convened the CEOs and senior executives of America's largest AI companies at the White House and handed them a pen. Amazon, Google, Meta, Microsoft, Oracle, OpenAI, and xAI all signed what the administration is calling the Ratepayer Protection Pledge — a commitment that their data centers' explosive electricity demands will not raise household utility bills. It was a striking photo opportunity. It may also be one of the most structurally complicated promises in the history of U.S. technology policy.
What the Pledge Actually Says
The White House fact sheet describes the pledge in sweeping terms: signatories will "build, bring, or buy" new power generation resources sufficient to cover their data center needs, cover the cost of all power delivery infrastructure upgrades, and negotiate separate rate structures with utilities and state governments — paying those rates whether or not they actually use the electricity.
The companies signing on represent the spine of American AI infrastructure. AWS CEO Matt Garman, Oracle CEO Clay Magouyrk, Google President Ruth Porat, Meta President Dina Powell, Microsoft President Brad Smith, OpenAI COO Brad Lightcap, and xAI's Gwynne Shotwell all attended the signing ceremony. Trump declared that "these companies are committing to provide or pay for all power generation and electricity needed for their AI projects, which is massive."
The pledge is also published in the Federal Register, lending it an air of official policy. But federal publication is not the same as federal enforcement authority.
The Enforcement Gap
Here is the structural problem that energy policy experts raise almost immediately when reviewing the pledge: the U.S. electricity grid is not a federal system. It is a patchwork of 50 states, each with their own public utility commissions, regulatory frameworks, and rate-setting authorities. The rules governing who pays for new power generation and transmission infrastructure are made at the state level — and those states must individually choose to adopt policies that require data center operators to internalize their costs.
"The White House can't do that on its own," said Rob Gramlich, president of Grid Strategies and a former economic advisor to the Federal Energy Regulatory Commission. "It doesn't have any jurisdiction there and of course the technology companies can't do that on their own either."
Trump's own trade advisor, Peter Navarro, had previously said the White House would "force" tech companies to internalize data center costs. But without state-level action — or congressional legislation — the Ratepayer Protection Pledge is a voluntary commitment with no binding mechanism. Democrats were swift to point this out. "A handshake agreement with Big Tech over data center costs isn't good enough," Senator Mark Kelly of Arizona posted. "Americans need a guarantee that energy prices won't soar and communities have a say."
The Scale of the Price Problem
The political urgency behind the pledge is real. U.S. residential electricity prices have risen more than 36% since 2020 — from 12.76 cents per kilowatt-hour to 17.44 cents per kilowatt-hour in February 2026. The U.S. Energy Information Administration projects prices will continue climbing, reaching 19.01 cents per kilowatt-hour by September 2027. Trump had promised to cut electricity prices in half during his first term. Instead, residential prices rose 6% in 2025 alone on average nationwide, according to federal data.
A Goldman Sachs report published in February warned that electricity prices are forecast to rise another 6% through 2026 and an additional 3% in 2028, as data center demand expands faster than new power supply can be commissioned. The political math for midterm elections is obvious: communities are angry, and the anger has a very visible target — the giant, humming, power-hungry campuses that have materialized on the outskirts of cities from northern Virginia to the Arizona desert.
Trump himself acknowledged the optics problem with unusual candor during the signing ceremony. Data centers, he said, "need some PR help."
The PJM Problem Is Structural
The cost surge is most severe on PJM Interconnection, the largest electric grid in the U.S., serving 13 states across the mid-Atlantic and Midwest. PJM is the epicenter of the U.S. data center buildout — home to facilities operated by Google, Anthropic, Amazon, and dozens of smaller operators concentrated in Virginia's so-called "Data Center Alley."
PJM's independent market monitor, Monitoring Analytics, has documented a direct correlation: data center growth has driven $23 billion in additional costs across the PJM system — costs that flow downstream to consumers through a mechanism called the Base Residual Auction, which sets forward prices for power capacity across the grid.
A SemiAnalysis report released this week argues that this market mechanism — not raw data center demand — is doing most of the damage. The Base Residual Auction requires consumers to prepay for anticipated future electricity costs two years in advance, based on simulations that rely on proprietary models. SemiAnalysis found that PJM's models have systematically overestimated future demand, particularly because many planned data centers faced construction delays tied to a chronic memory chip shortage. The result: consumers paid inflated forward prices for capacity that never materialized on schedule.
The contrast with Texas is striking. On ERCOT — the Electric Reliability Council of Texas, the grid that now hosts data center clusters from OpenAI, Anthropic, and Google — residential electricity prices have remained relatively stable since 2022, despite comparable levels of hyperscaler investment. ERCOT's market structure enables direct cost allocation between large industrial consumers and utilities, insulating residential customers more effectively.
Individual Pledges Pre-Date the White House Deal
Several major hyperscalers had already made voluntary commitments to cover electricity cost increases before Trump summoned them to the White House. In January, Microsoft outlined a five-point community plan that explicitly included covering any additional electricity costs resulting from its data center operations. Anthropic followed with a similar commitment in February.
These pre-existing pledges raise a pointed question: if the largest players were already making these commitments voluntarily, what does the White House signing ceremony add? The answer, according to Chris Howard, head of data centers account management at JLL, may be largely political. Such commitments are "particularly important for drawing support from communities that otherwise might oppose projects," Howard told CNBC — especially when accompanied by local investments in jobs and workforce training.
But the financial skepticism runs deep. "The problem is, the industry's not making money, so that puts even more pressure on them," Marc Einstein, research director at Counterpoint Research, said of the hyperscalers. Companies cannot indefinitely absorb grid infrastructure costs that regulators haven't required them to pay, particularly when AI investment is running ahead of monetization timelines.
FERC, Emergency Auctions, and the Governance Gap
The Trump administration has pursued parallel tracks to give the pledge more structural teeth. Energy Secretary Chris Wright wrote to the Federal Energy Regulatory Commission in October requesting that FERC claim jurisdiction over large load interconnections — which would allow the federal government to require data centers to pay the full cost of grid upgrades. The administration and a bipartisan group of governors also called on PJM to hold an emergency capacity auction in January, in which tech companies would bid to bring new power plants online.
Those mechanisms are real. But they are slow. FERC rulemaking takes years. State PUC proceedings take years. The grid connection queue for new data centers already stretches four to six years in primary U.S. markets — and up to a decade in cities like Tokyo. In the meantime, Europe is beginning to test an alternative model: in Dublin, the first microgrid-connected AI data center has come online, able to operate independently of the public grid while also offering dispatchable battery storage back to utilities. The global microgrid market was worth approximately $29 billion in 2025 and is growing rapidly.
The Political Calculus
The Ratepayer Protection Pledge sits at the intersection of two goals that the Trump administration is straining to hold together: enthusiastic support for AI as an engine of economic and national security dominance, and the promise of lower energy costs for American households. Those goals are not inherently incompatible. But they require a genuine restructuring of how electricity markets allocate the costs of large industrial loads — not a voluntary commitment signed in front of cameras.
The political window is narrow. Midterm elections are coming, and community backlash against data centers is intensifying from Virginia to Arizona. Local advocacy groups have staged protests at state capitols. Cities that once competed to attract hyperscaler campuses with tax incentives are now watching those same campuses strain their local power infrastructure and generate noise complaints from neighbors.
The technology companies have every incentive to get ahead of this problem. Communities that feel burned by data center promises become hostile to permitting, zoning variances, and the very local cooperation that hyperscalers need to build at the pace AI development demands. But goodwill gestures and White House photo opportunities cannot substitute for the hard regulatory work of restructuring electricity markets to ensure that the costs of AI infrastructure are paid by the entities that create them — not passed silently onto the utility bills of American households who never asked to subsidize the training of large language models.
The pledge is a start. The grid needs a lot more than that.
