OpenAI’s community compact: the real fight is who pays for AI power

On 5 November 2026, OpenAI published a National Community Compact for future compute campuses it owns and operates, and a Compute Campus Policy Blueprint for lawmakers. The soft framing is community voice and shared AI gains. The harder, testable claim is narrower: voluntary campus promises will not settle the US buildout fight; state utility rules on who pays for large new loads will. If commissions keep socializing hyperscale power costs onto households, or peers build without matching terms, today’s compact is branding. If the blueprint’s cost-causation and curtailment rules stick in a few key states, it becomes the template rivals must answer.

What OpenAI actually committed to

For campuses it owns and operates, OpenAI says each site will get a local compact: early resident input that can still change plans, a named local contact for the life of the campus, and yearly public progress reports. The shared floor includes paying the full cost of electricity and of the infrastructure needed to serve the load; posting financial assurances before major spend so other customers are not stranded if the project is delayed, cut, or canceled; a water plan that prefers closed-loop or low-water cooling; publishing site-level water and electricity use; releasing construction plans and independent noise studies before operations and every year after; and refusing broad NDAs with public officials, with only narrow carve-outs for security, trade secrets, or competitively sensitive material.

The company ties this to January Stargate Community commitments and to site work in Georgia and Ohio. In Effingham County, Georgia, residents helped shape a $4.8 million grant for free school meals for all 14,000 students in the county. In Pike County, Ohio, childcare rose as a local priority, leading to a joint $6 million commitment from OpenAI and SB Energy.

Where the numbers bite

The policy half is more revealing than the grant anecdotes. At the planned Effingham County campus, OpenAI’s agreement with Georgia Power lets the utility reduce power delivered to OpenAI by up to 1,000 megawatts when demand is high. That is a concrete curtailment term, not a vibe about “being a good neighbor.”

The blueprint then asks state utility commissions to require large electricity customers to cover the grid costs they add and to protect other customers if a project fails. It also wants easier interconnection for campuses that can safely cut demand when power is scarce. Congress is asked to speed transmission upgrades and to restore and expand clean-electricity tax incentives. On accountability, it wants public reporting on water, energy, noise, and hiring; narrow NDA limits for public officials; and, where tax incentives are offered, public terms, measurable commitments, and clawbacks if a developer misses them. On labor and suppliers, it urges early purchasing plans so local firms can bid, and hiring forecasts shared with unions and community colleges so training matches known jobs, from construction electrical work to long-run power and cooling maintenance.

Steelman and gaps

Steelman. OpenAI is writing the terms it can live with before hostile ordinances or punitive rate designs arrive. Paying own-way on power and posting assurances reduces the main political attack: that AI campuses raise household bills. Curtailment rights give utilities an operational valve. Local grants buy goodwill. Publishing a blueprint lets OpenAI claim industry leadership while still choosing which campuses it owns and operates, and which sit with partners under different rules.

Those points are fair. They do not erase the gaps. The compact is voluntary and scoped to campuses OpenAI owns and operates; partner or landlord sites may sit outside it. The Georgia and Ohio dollar figures are company-reported community investments, not independent audits of net fiscal impact. The 1,000 MW curtailment is an agreement with one utility, not a national rule. The blueprint asks lawmakers to act; as of the post, it is not enacted statute. Peers can free-ride on OpenAI’s political cover while offering thinner local terms. NDAs “narrowed” still leave room for what counts as competitively sensitive. Water and noise reports matter only if regulators or courts can act on them.

Who pays, and what to watch

Households and small businesses near large loads care whether commissions enforce cost-causation or keep burying hyperscale upgrades in the general rate base. Utilities care whether curtailment-capable campuses get faster interconnection without stranded-asset risk. Local governments care whether tax abatements carry clawbacks. Building-trades unions and community colleges care whether hiring forecasts arrive early enough to train people into real seats. Rival campus developers care whether OpenAI’s floor becomes the political minimum in swing counties.

Over the next six months, watch three observables: whether any state commission adopts large-load cost-causation or assurance rules that rhyme with the blueprint; whether OpenAI’s first owned-and-operated campuses publish the promised site-level water, power, noise, and hiring numbers on schedule; and whether a major peer announces a thinner or thicker compact when facing similar local pushback.

Close

OpenAI’s durable sentence is not that AI campuses will be kind. It is that US compute growth now runs through rate cases and county politics as much as through GPU supply, and the company is trying to draft the acceptable bargain before commissions and rivals do. Six months from now, either a few PUCs write cost-causation and curtailment into rules, or this compact stays a well-produced voluntary brand.