OpenAI Commits to Full Power Costs and Bans Official NDAs Following Data-Center Election Backlash

OpenAI released its National Community Compact and Compute Campus Policy Blueprint on Thursday, two days after voters across 10 states rejected data-center expansion in 40 of 41 local ballot measures, pledging to pay the full electricity and infrastructure costs at every compute campus the company owns and operates and to stop requiring nondisclosure agreements from public officials. The release, timed to the most decisive electoral rebuke of AI infrastructure expansion in U.S. history, pairs specific site commitments with a legislative blueprint asking federal and state lawmakers to extend comparable requirements across the entire industry.

Voters Sent a Message That Markets and Boardrooms Are Still Processing

The November 3 midterm elections delivered a verdict that money struggled to reverse. Ballot measures spanning bans, moratoriums, and construction restrictions in California, Florida, Illinois, Kansas, Maine, Nevada, New Jersey, North Dakota, Ohio, and Wisconsin moved in an anti-data-center direction save one: in Harwood, North Dakota, voters rejected a measure that would have required independent review of large industrial projects, effectively sparing data centers in that jurisdiction. Data center and AI-acceleration super PACs spent heavily in key races and lost. An exit poll cited by CNBC found that 56% of voters described AI and data centers as a crucial or major factor in their vote for Congress. A Quinnipiac poll cited in CNBC's reporting put opposition to a nearby AI data center at 73% of registered voters.

Democrats flipped seven Senate seats and gained a majority in both chambers—the first time both chambers flipped in a midterm since 2006. They also won governorships in Georgia, Ohio, and Texas. Those three states, together with several other Democratic pickups, now account for 37% of planned U.S. data center capacity, according to data center industry analysis. Incoming Texas governor Gina Hinojosa campaigned on a moratorium on new data center construction; real estate firm JLL had projected Texas would surpass Virginia as the world's largest data center market by 2030. In Ohio, mayors of Piqua and Urbana were recalled specifically over their support for data center projects. By October 2026, more than 100 data center moratoriums were pending at local and state levels across the country, according to Amazon Web Services CEO Matt Garman.

What the Compact Commits OpenAI To

The National Community Compact covers every future compute campus OpenAI owns and operates. Under the core financial commitment, OpenAI will pay the full cost of the electricity its campuses consume and of the infrastructure—substations, transmission extensions, distribution equipment—required to deliver that electricity, insulating utility customers from cost transfers. That commitment is broader than simply paying an electricity bill: grid infrastructure carries fixed costs regardless of whether a large customer is currently drawing power, and traditional rate design spreads those fixed costs across all ratepayers. The compact's pledge eliminates that cross-subsidy for campuses OpenAI controls directly.

The NDA commitment prohibits OpenAI from requiring public officials to sign nondisclosure agreements when negotiating or discussing future campus sites. The ban preserves narrow carve-outs for security information, trade secrets, and competitively sensitive disclosures. OpenAI also commits to disclosing noise levels, water use, and energy demand publicly, participating in demand-response programs, and providing financial assurances to protect ratepayers if projects are delayed, reduced, or canceled.

Site-specific commitments are already in place. At Project Camellia—OpenAI's planned 3.2-gigawatt campus in Effingham County, Georgia—the company has listed an $80 million commitment over the project's life, including a $4.8 million grant for free school meals for all 14,000 county students and $2 million for a local park. In Pike County, Ohio, where OpenAI and SB Energy are developing a planned campus of approximately 8 gigawatts of IT capacity on a former uranium enrichment site, families identified childcare as a priority, resulting in a joint $6 million pledge from OpenAI and SB Energy.

Grid Protectors: Demand Response as a Policy Tool

The compact is the third document in a sequence OpenAI published over three days. On November 2, the day before the election, OpenAI launched the Grid Protectors Program, a formal framework under which campuses it develops will cut electricity consumption when local grids are under strain. The pilot is the Effingham County campus, which carries a 25-year agreement with Georgia Power allowing up to 1,000 megawatts of demand reduction during stress events—roughly one-third of the campus's planned 3.2-gigawatt draw. Under the arrangement, the utility can call a curtailment event specifying depth, speed, and duration; the campus reduces its draw accordingly. Some sites may also return power from on-site generators or batteries. The terms will vary by location and are set to be announced as individual campus arrangements are finalized.

Read more: OpenAI Grid Protectors Program locks demand response into every future AI campus

Georgia Power separately projected in August 2026—after the OpenAI contract was approved by the Georgia Public Service Commission—that its full portfolio of large industrial customer agreements could reduce average residential electricity bills by approximately $950 million annually and roughly $15 per month by 2029. That figure covers OpenAI and other large customers collectively, not OpenAI alone.

A Blueprint Asking Lawmakers to Write the Rules

Alongside its corporate compact, OpenAI released a Compute Campus Policy Blueprint addressed to federal, state, and local policymakers. The document asks utilities to require large electricity customers to cover the full grid costs they add, including infrastructure brought online for their use even when they are not consuming power, and to mandate financial assurances that protect customers if projects fall through. The blueprint calls for narrow statutory limits on NDA requirements for public officials, mandatory public disclosure of water use, energy consumption, noise, and hiring data, and local hiring forecasts tied to both construction and long-term operations. Chris Lehane, OpenAI's chief global affairs officer, said the blueprint had been in development since August and that he had met with dozens of lawmakers and candidates during that period. OpenAI is preparing a policy push for the Congressional lame-duck session, which opens November 9.

OpenAI Was Not First, but May Be Most Specific

OpenAI's compact does not establish a new precedent on NDAs or ratepayer protection—it sets a new high-water mark in specificity. Microsoft committed in January 2026 to drop NDAs with local governments as part of its Community-First AI Infrastructure plan, and said it would review existing NDA arrangements to work toward their closure. Amazon Web Services CEO Matt Garman announced on October 2, 2026 that AWS would stop requiring NDAs with government agencies and committed more than $1 billion to host communities through its Built Together program. All seven major AI infrastructure companies—OpenAI, Amazon, Google, Meta, Microsoft, Oracle, and xAI—signed the White House's March 2026 Ratepayer Protection Pledge, committing broadly to build, buy, or bring new power rather than shifting grid costs to retail customers.

What the OpenAI compact adds to that established baseline is the explicit public construction plan disclosure, independent noise studies, a demand-response requirement with a named pilot and a quantified capacity ceiling, and a legislative blueprint pairing its own corporate pledges with a formal ask of regulators. AI Weekly's editorial summary noted the competitive implication directly: any hyperscaler that does not match the NDA and power-cost floor OpenAI has now published will face the question publicly in state PUC proceedings and upcoming legislative sessions.

The Scope Limitation That Most Reporting Underplays

The compact's most consequential constraint is in its own scope language: "campuses OpenAI owns and operates." That formulation excludes the data center infrastructure on which OpenAI's current AI workloads—including ChatGPT inference—actually run. OpenAI conducts most of its operations on leased infrastructure from Microsoft Azure, to which it is committed in a multibillion-dollar strategic partnership; Amazon Web Services, with which it signed a $38 billion cloud agreement in late 2025; and Oracle Cloud Infrastructure, which underlies Stargate builds in Texas. In each of these arrangements, the data center infrastructure is owned and operated by the hyperscaler—not by OpenAI. The Grid Protectors Program carries the same boundary: it covers only campuses OpenAI develops itself.

For projects developed through SB Energy partnerships—including the planned Pike County, Ohio campus—the ownership structure under long-term development will determine whether the compact's terms eventually apply. AI Weekly flagged the same gap in its editorial analysis: does "owns and operates" quietly exempt the leased capacity and joint ventures where most of OpenAI's actual power draw resides? OpenAI did not publish an answer to that question in the compact released today.

What the New Political Map Changes

The more lasting pressure created by the midterm results is structural. With Democratic governors taking office in January in states holding more than a third of planned U.S. data center capacity, the permitting environment for new facilities will tighten independently of voluntary corporate commitments. Political analysts and incoming members of Congress noted that the new legislative session is more likely to pass legislation protecting households from AI infrastructure costs than the previous Congress was.

OpenAI's national compact and legislative blueprint are designed to shape what that legislation says before Democratic state capitals and a newly empowered Congress write stricter versions. Whether the company succeeds depends on how quickly it converts the compact into binding local agreements—and whether state utility commissions, emboldened by an election in which opposing data centers was a winning position in 40 of 41 contests, hold the company to its own language about which campuses actually count.