OpenAI pledges to pay its own power bills. It wants lawmakers to make that the rule for everyone.
OpenAI says it will pay for the power its own campuses use and will not require officials to sign NDAs. Those pledges are voluntary, and the rules it wants written into law are not law yet.
On November 5, OpenAI published two documents about the data centers it builds. A National Community Compact lists what communities can expect at campuses OpenAI owns and runs. A Compute Campus Policy Blueprint asks federal, state and local governments to write similar rules into law for the whole industry. The compact’s pledges are voluntary, and OpenAI says it will report on them once a year.
Both build on what OpenAI announced in January as its Stargate Community commitments, plus specific deals in Georgia and Ohio. Only part of the announcement page was readable to me, so the detail below comes from the two PDFs. I also read CNBC’s report on the launch.
Does OpenAI really pay its own electricity bill?
The announcement page says OpenAI will “pay the full cost of the electricity we use and the infrastructure needed to serve us.” Wording for the infrastructure part changes in the compact. There it promises “the incremental costs of the infrastructure required to serve our sites and the full cost of the electricity we use.” Nothing in the compact defines “incremental.”
My read: that wording gap is the first thing to check. “Incremental” is a narrower word than “the infrastructure needed to serve us,” and a reader should ask which version a utility contract actually uses.
The Georgia deal has the most numbers attached. OpenAI says its agreement with Georgia Power covers the full infrastructure cost for its Effingham County campus. It also “committed up to 1,000 MW of flexible demand response,” which lets Georgia Power reduce the power delivered to the site at times of high demand. In the blueprint, Georgia Power says typical residential customers should see savings of at least $15 a month beginning in 2029. That is Georgia Power’s claim, and I could not check it.
It also promises financial assurances “before major spending begins,” so other customers are protected if a project is delayed, reduced or canceled. It does not say what form those assurances take or how large they are.
What is voluntary, and what gets checked?
OpenAI’s own blueprint calls these commitments voluntary. It says they are “an important starting point,” but “voluntary measures alone don’t meet the scale of this moment or provide durable confidence to communities.” Nothing in either document makes the compact enforceable in law. The same question came up with the September White House AI accord, which six AI chief executives signed as a voluntary pledge.
Reporting is self-reported too. Its text says OpenAI will “report publicly each year” and “report annually to the public on progress.” Neither document names an outside auditor. Noise studies are called independent. OpenAI commissions them.
My read: the NDA line is the easiest pledge to test. The compact says OpenAI “will not require public officials to sign NDAs as a condition for discussion or negotiation,” with narrow exceptions for legitimate security information, trade secrets and competitively sensitive information. The blueprint adds that OpenAI voided NDAs in Effingham in July. That is OpenAI’s statement, and the documents alone cannot confirm it. A local reporter could test it against county records.
What OpenAI does itself, and what it wants the law to say
| Topic | OpenAI’s pledge for its own campuses | What the blueprint asks lawmakers to require |
|---|---|---|
| Electricity | Pays the full cost of the electricity it uses, plus infrastructure costs (the compact says “incremental”) | Large-load tariffs so data centers pay the incremental costs they create, and codify the national Ratepayer Protection Pledge |
| Financial protection | Financial assurances before major spending begins | Utilities secure financial assurances before major spending |
| Public officials | No NDAs, except narrow security and trade-secret cases | Bar developers from requiring NDAs from public officials |
| Water | A water plan per campus, closed-loop or low-water cooling, water use published by location | Mandatory water-use reporting by location, including cooling and water-management practices |
| Noise | Independent studies before operations and yearly after, commissioned by OpenAI | Independent studies before operations and regularly after, with results public |
| Energy | Publishes on-site electricity use, including power it generates | Large users report total facility electricity use and power usage effectiveness on standard methods |
| Tax incentives | No general pledge. In Effingham, OpenAI says it agreed to pay at the pre-reduction tax rate | Public disclosure of incentive deals, measurable commitments, and clawbacks if they are not met |
| Jobs | Workforce strategy, annual reports, local hiring goals where labor conditions allow | Public workforce strategies and annual progress reports, with local hiring goals where labor conditions allow |
On the left, OpenAI’s own pledges. On the right, what it wants legislatures to require from every developer. My read: only the right column would bind a company that does not want to be bound.
Why the timing matters
CNBC reports that voters backed restrictions or rejected pro-development measures in 40 of 41 local ballot measures across 10 states on November 3, citing a Ballotpedia analysis and its own count. CNBC also reports that OpenAI’s chief global affairs officer, Chris Lehane, said the blueprint has been in the works since August and that he has met dozens of lawmakers and candidates since then. Lehane told reporters the aim is “a very high floor in terms of how people move forward.” OpenAI told investors in February that it is targeting $600 billion in total compute spend by 2030.
My read: a company with that spending plan has a clear reason to want one set of rules in place before the next Congress. Lehane’s own timeline points there. He described the work as getting done in advance, so it is ready when Congress returns in late January. Read the blueprint as a policy pitch, because that is what it is.
This is the second recent OpenAI campus story. The compute campus rename and the Grid Protectors program at Effingham covered the naming and the Georgia Power grid deal. This one covers the commitments and the rules OpenAI wants written down.
What to check next
Three things would tell a reader more than the launch documents. The first is the first annual report, and whether it gives the measured numbers the compact promises, such as water use by location and on-site electricity. The second is the cost language in the utility agreements behind the Georgia and Ohio sites, to see whether “incremental” or “full” appears. The third is whether any state takes up the large-load tariff idea in the blueprint before its next session.
Is it enough for a reader?
My read: the NDA pledge and the financial assurances are specific enough to test, so they are where to hold OpenAI first. The rest of the compact depends on reports OpenAI writes itself, and the blueprint is still a request to lawmakers. Until a legislature passes something, the compact is a list of stated intentions for OpenAI’s own sites, not a standard that applies to anyone else.
Note on sources: I read both PDFs in full. The compact runs eight pages and the blueprint 18. The announcement page was only partly readable to me, so I cite it only for the commitment and blueprint lines I could see. The PDFs' document metadata says "Stargate Policy Blueprint," though the covers read "Compute Campus Policy Blueprint." The Georgia Power savings claim, the $4.8 million school-meals grant, the $6 million Pike County childcare commitment and the 1,000 MW figure all come from OpenAI's own documents. I did not check them against county, utility or SB Energy records. The NDA voiding in Effingham is OpenAI's statement. The ballot-measure count is CNBC's, citing Ballotpedia.