The White House has once again put America’s hyperscalers on notice. In a fresh social media push this week, the administration restated its position that the leading US artificial intelligence companies must build, bring or buy every megawatt of power their data centres consume, rather than allowing the cost to land on household electricity bills.
The message is not new. It revives the language of the Ratepayer Protection Pledge, signed at the White House in March by Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI. But the timing of the reiteration, coming as midterm campaigning intensifies and scrutiny of the pledge’s enforcement grows, suggests the administration is keen to keep the issue firmly associated with its own policy agenda rather than ceding ground to critics.
Under the pledge, signatories agreed to pay the full cost of new generation capacity and grid upgrades tied to their facilities, and to negotiate separate rate structures with utilities and state regulators rather than relying on the socialised cost base that typically spreads infrastructure spending across all customers. Companies also committed to paying for contracted power whether or not they actually draw on it, a provision designed to stop hyperscalers from reserving capacity and leaving the bill with everyone else.
David Sacks, the White House’s special advisor for AI and crypto, described the arrangement at the time as an alternative to more restrictive proposals circulating in Congress, noting that AI firms would be expected to cover new power generation costs rather than pass them on to residential consumers. Amazon Web Services chief executive Matt Garman welcomed the pledge as a baseline for what he called responsible, long-term energy partnerships with the communities hosting new facilities.
Why the White House is doubling down
Public patience with the data centre build-out has been thinning for months. Polling cited by Brookings researchers found that support for new data centre construction in Virginia, long the industry’s most established US market, has fallen sharply since 2023, while opposition in states such as Wisconsin has hardened as voters link rising bills directly to AI infrastructure. Those numbers have clearly registered inside the West Wing, where officials are wary of handing Democratic challengers a ready-made affordability argument ahead of November’s elections.
That political backdrop helps explain why the administration keeps returning to the same script. Energy Secretary Chris Wright has previously said the government wants data centres built and sited quickly, but only on terms that keep the cost off residential bills. Federal Energy Regulatory Commission chair Laura Swett went further in June, when commissioners voted to require large power users, including data centres, to cover the full cost of any grid connections they trigger. She called it action to protect ratepayers from shouldering costs tied to connecting major new loads.
Whether the pledge itself carries any binding force remains an open question. Grid Strategies president Rob Gramlich has pointed out that electricity markets are regulated overwhelmingly at state level, meaning the White House has limited direct authority to enforce voluntary commitments made in Washington. Analysts at Brookings have reached a similar conclusion, arguing that the pledge represents a useful starting point for a new industry consensus but lacks any mechanism to verify whether companies are keeping their word.
What it means for global data centre operators
The core principle now embedded in US federal messaging, that hyperscalers should absorb infrastructure costs directly rather than relying on socialised utility pricing, is already shaping conversations in other markets grappling with grid strain, from Ireland to the Gulf. Anthropic’s own commitment to cover the full cost of grid upgrades tied to its US data centres, announced alongside a $50 billion infrastructure investment earlier this year, reflects a similar logic taking hold industry-wide, regardless of which government is applying the pressure.
There is also a commercial dimension worth watching closely. A recent J.P. Morgan analysis found that more than 60 per cent of data centre capacity scheduled for completion in 2027 had not yet broken ground, based on satellite imagery, with a further 7 per cent already delayed. If power procurement negotiations under the pledge slow project timelines further, as some in the sector fear, the gap between announced capacity and delivered capacity could widen just as AI compute demand keeps climbing.
For now, the signatory companies appear willing to play along publicly. Senator Jon Husted described the pledge as a foundation for building trust with local communities and stakeholders, arguing that new power plants tied to data centres would ultimately support grid stability. Whether that trust survives the state-by-state rate negotiations still to come is likely to be the real test of the policy, rather than any further reiteration of it from the White House press office.
Executives tracking the story should expect more restatements of the same pledge as the midterms approach, alongside continued state-level pushback and further FERC rulemaking on cost allocation for large loads. The direction of travel, towards hyperscalers absorbing infrastructure costs directly, looks increasingly settled. The mechanics of how that gets enforced remain very much unresolved.
RELATED STORIES
- AI power crunch: Trump tells big tech to foot its own energy bills
- Anthropic to shoulder price hikes from its US data centres
- The data centre dilemma: Fuelling the economy and facing local fury
- Data centres continue to guzzle electricity – what’s the industry doing about it?
- The power gap: How data centres can overcome their biggest limiting factor






