OpenAI's Community Compact: Paying Full Price for Power, and Asking Rivals to Do the Same

On November 5, two days after voters rejected AI data centers in 40 of 41 local ballot measures across the United States, OpenAI published two documents that read like a direct answer to that verdict: a National Community Compact and a Compute Campus Policy Blueprint. The compact commits OpenAI's future campuses to pay the full cost of the electricity they use and the infrastructure needed to serve them. The blueprint asks federal, state and local governments to make the rest of the industry do the same.

The compact applies to future campuses that OpenAI owns and operates, and its pledges are specific enough to be checked. Each campus gets a locally tailored agreement, built after early engagement with residents through surveys and listening sessions, with a local point of contact for the life of the site and an annual public progress report. OpenAI will publish site-specific plans covering construction traffic, noise and lighting, commission independent noise studies before operations begin and every year after, and disclose site-level water and electricity use. Every campus gets a water plan that prioritizes closed-loop or low-water cooling. And in a clause aimed squarely at the secrecy that has fuelled local suspicion, OpenAI says it will not require public officials to sign non-disclosure agreements, keeping only narrow protections for security information, trade secrets and proprietary data.

The company points to money already committed as evidence that this is more than paper. In Effingham County, Georgia, it has pledged $80 million to community priorities, including a $4.8 million grant funding free school meals for all 14,000 students in the county and $2 million to develop a local park. In Pike County, Ohio, residents named childcare as their priority, and OpenAI and SB Energy jointly committed $6 million to it. The documents build on OpenAI's January 2026 Stargate community commitments and follow the November 2 launch of Grid Protectors, a programme under which future campuses can cut their electricity demand during grid stress; its Georgia Power agreement already allows the utility to reduce deliveries to the Effingham campus by up to 1,000 megawatts at peak times, against 3.2 gigawatts contracted for Project Camellia between 2028 and 2032.

There are two catches, and OpenAI does not hide either of them. The first is scope: the compact covers only sites OpenAI owns and operates, and much of the computing capacity behind its products is rented or developed with partners, where these pledges do not apply. The second is that the whole thing is voluntary, and OpenAI says so itself — the blueprint describes voluntary corporate commitments as "only a starting point" that "cannot give communities lasting assurance at the scale needed." That is why the blueprint's real audience is not OpenAI's own engineers but policymakers: it asks governments to require large electricity users to pay for the grid expansion they cause, to protect other customers when projects are cancelled, to publish the terms of tax incentives and claw them back if developers fail to deliver, and to restrict mandatory NDAs for public officials.

That request is the most interesting part of the package. A company that has spent years resisting regulation is now lobbying for rules that would bind its competitors — which is also, not coincidentally, a way of turning its own concessions into the industry floor. If rivals match the compact, OpenAI's move looks like leadership; if they don't, it looks like a public-relations premium paid by one company while the rest of the industry keeps the old terms. The rebrand tucked into the documents gives the game away: OpenAI now calls its data centers "compute campuses," a name meant to sound like something a town would want next door.

The test, then, is not whether OpenAI keeps its own promises — those are written down, dated and checkable — but whether the standard travels. Voters have already shown they will say no to the old model. The question now is whether the industry's other giants would rather adopt OpenAI's terms voluntarily or wait for the state capitols to impose them. My prediction: by June 30, 2027, at least one other frontier AI lab — Anthropic, Google DeepMind, Meta or xAI — will publish a public community-benefits compact for its data centers comparable to OpenAI's, committing to pay full electricity and infrastructure costs and to disclose site-level water and energy use. I put that at 50%. The pressure is real and now proven at the ballot box, and matching a rival's public standard is cheaper than fighting a moratorium. But these companies have historically preferred quiet local deals to binding public commitments, and none of them has yet shown appetite for OpenAI's level of disclosure. A miss would be informative in its own way: it would mean the compact stays a one-company gesture, and the fight moves to the legislatures instead.