Two days after midterm voters constrained data-center projects almost everywhere they were on the ballot, OpenAI published a National Community Compact and a policy blueprint. The NDA pledge is real—and so is the ownership-scope limit that decides how much of OpenAI’s compute it actually covers.
On November 5, 2026, OpenAI published a National Community Compact and a Compute Campus Policy Blueprint. The company says every future “compute campus” it owns and operates will get a local compact: early community input, a local point of contact, annual public reporting, and a baseline set of commitments on power costs, water, noise, hiring, and secrecy.
The timing is not subtle. Midterm voters across the United States had just delivered a near-unanimous local rebuke of data-center expansion. According to CNBC’s tally, drawing on Ballotpedia and its own results, of 41 local ballot measures across 10 states, voters backed restrictions or rejected pro-development measures in all but one. OpenAI’s announcement lands as political risk management for an industry that still needs megawatts faster than many communities want them built.
What OpenAI actually promised
The compact’s concrete pledges, for campuses OpenAI owns and operates, fall into three buckets.
Energy and ratepayers. OpenAI says it will “pay the full cost of the electricity we use and the infrastructure needed to serve us,” and provide financial assurances before major spending so other utility customers are protected if a project is delayed, reduced, or canceled. That restates a theme the company has pushed through earlier Stargate community commitments and a Ratepayer Protection Pledge. At its planned Effingham County, Georgia campus, OpenAI says its Georgia Power agreement allows the utility to cut delivery by up to 1,000 megawatts when demand is high—a demand-response clause that matters more than slogan language about “paying our way.”
Resources and transparency. Each campus is to get a water plan that prioritizes closed-loop or low-water cooling, with site-level water and electricity use published. Construction plans and independent noise studies are to appear before operations begin and annually afterward. Critically for local politics: OpenAI says it will not use nondisclosure agreements with public officials, allowing “only narrow protections covering legitimate security information, trade secrets, or proprietary or competitively sensitive information.”
Jobs and local spend. The company pledges to publish expected jobs, work with unions and training providers, report annually on hiring and apprenticeships, and make “sustained investments in priorities residents help select.” It cites prior examples: a $4.8 million grant for free school meals for all 14,000 students in Effingham County, Georgia, and a joint $6 million childcare commitment with SB Energy in Pike County, Ohio.
The blueprint goes further than the company compact: it asks state utility commissions to force large load customers to cover incremental grid costs, calls for public tax-incentive terms with clawbacks, and urges limits on when developers can demand NDAs from officials. In other words, OpenAI is inviting regulation that would bind competitors as well as itself—a familiar move when voluntary standards risk becoming a competitive disadvantage.
Why NDAs became the flashpoint
Nondisclosure agreements turned into a proxy for a deeper complaint: that data-center deals were negotiated in private while ratepayers, water districts, and neighbors learned the company name after permits were already moving. Reporting over 2025–2026—from Tucson, Arizona, to South Bend, Indiana, to Louisiana’s Hyperion project—documented officials barred from sharing even basic project identity or resource use. Amazon said in early October 2026 that it no longer uses NDAs with government agencies on data-center projects; Microsoft had already stopped asking local governments to sign them. A bipartisan House bill, the No Secrets for Data Centers Act, and state executive orders in Pennsylvania, Massachusetts, and Delaware had made the practice a legislative target.
OpenAI’s pledge therefore matches a direction the industry was already being pushed. It is meaningful as a formal commitment from a company whose Stargate-linked campuses in Georgia and Ohio are among the largest proposed AI loads in the country. It is not, by itself, a new invention of transparency norms.
The fine print that decides whether this matters
Three limits define the practical reach of the compact.
Owned and operated only. OpenAI is explicit: the compact applies to future campuses it owns and operates. A large share of frontier-model compute still arrives through cloud partners, co-location, and joint development. Data Center Dynamics flagged the gap immediately: it is unclear how much of OpenAI’s leased or partnered capacity sits outside the compact. A transparency regime that covers the campuses on OpenAI’s balance sheet while leaving partner sites under older secrecy practices would solve the political problem unevenly.
“Could” versus “will.” Hard pledges (pay full power costs; no public-official NDAs; publish water and electricity use; noise studies) sit beside softer community-investment language. DCD noted formulations such as “could support housing stability” and “could invest in education.” Soft options are not worthless, but they are not enforceable the way a published meter reading or a ratepayer-protection tariff filing can be.
Rebranding is not regulation. OpenAI is pushing “compute campus” as the label for these sites. Naming can reduce stigma; it does not change interconnection queues, transformer lead times, or ballot-box majorities. Markets noticed the political shift separately: on November 6, Asian chip and tech stocks sold off hard as investors priced slower AI infrastructure buildouts after the U.S. midterms, with SK Hynix, Samsung, TSMC, SoftBank, and related names among the hardest hit in regional trading covered by CNBC and CNA.
What the midterms actually changed
Local land-use politics, not a single federal statute, still decide most data-center outcomes. That is why the ballot scoreboard matters more than any one company’s blog post. CNBC’s 41-measure, 10-state tally—and corroborating reporting from The Verge and TechCrunch on dozens of restrictive local results—shows opposition that is geographic and bipartisan, driven by electricity bills, water, noise, land use, and a broader anxiety about AI’s local costs versus diffuse benefits.
OpenAI Chief Global Affairs Officer Chris Lehane told reporters the blueprint had been in development since August and that the company had briefed dozens of lawmakers and candidates, according to CNBC. That chronology undercuts a pure “overnight pivot” reading: the document was already being shopped as the backlash intensified. The midterms raised the cost of failing to publish it.
CNBC also notes that Meta, Microsoft, Google, and Anthropic have outlined several similar commitments. Convergence on NDA limits and ratepayer language is useful for communities; it also means OpenAI’s package should be judged on specificity, auditability, and scope—not on being first.
Compute ambition versus local consent
OpenAI told investors in February 2026 that it was targeting $600 billion in total compute spend by 2030, CNBC reported. Capex at that scale cannot be built only where neighbors applaud. The industry’s binding constraint is shifting from chip supply theater to permission to connect load: interconnection, generation, transmission, water rights, and zoning. Voluntary compacts are an attempt to buy back social license without waiting for Congress.
They may not be enough. Democratic control of Congress improves the odds for federal ratepayer-protection ideas and NDA limits, even if a nationwide moratorium faces veto politics. States and counties can still impose pauses faster than any blueprint can diffuse. For prior VibeCodAI context on how power contracts and debt shape the same buildout, see our reporting on direct-to-datacenter power contracts and the AI infrastructure debt test.
What to watch next
Three tests will show whether the compact is a standard or a press kit.
First, site-level disclosures: do water, electricity, noise, and hiring reports appear with enough granularity for a county board—or a utility commission—to verify them? Second, partner and leased capacity: does OpenAI extend equivalent NDA and ratepayer terms to projects where it is the tenant or offtaker rather than the owner-operator? Third, clawbacks and financial assurances: are the pre-spending guarantees filed in forms regulators can enforce if a campus shrinks or stalls?
OpenAI’s compact correctly names the political problem: communities will not host AI infrastructure they cannot inspect or refuse to subsidize. The unresolved question is whether owned-and-operated pledges can cover enough of the real buildout—or whether the next round of rules will be written at the ballot box and the public utility commission, with or without a company blueprint.