OpenAI spent the first two days of November 2026 doing something no AI company had done in Washington before: writing its own political opening statement. On November 2, the day before the United States midterms, Chris Lehane, OpenAI’s chief global affairs officer, published an essay titled 2028: The AI Election, laying out five questions he says every 2028 presidential candidate must answer. AI compute campuses feature in that argument as a proposed new name for the industry infrastructure that voters have started resisting.
The essay landed on the same weekend that Axios reported OpenAI would open Congress’s lame-duck session, which begins on November 9, with a four-part policy and public relations push. The four planks are mandatory safeguards for the most powerful AI systems, which OpenAI calls speed limits; community compacts and state siting legislation for the data centres it builds; an economic opportunity agenda covering worker disruption, education and business formation; and a push for the United States to shape global AI standards. The fifth element, the one that produced the sharpest reaction, was a rename. OpenAI wants its infrastructure called AI compute campuses rather than data centres. A spokesperson confirmed the name to Business Insider, saying compute campuses better describes the scale and role of the infrastructure being built for AI.
Why AI Compute Campuses Is a Proposal, Not a Rebrand
OpenAI has not announced a formal change of name. What exists is a proposal to be argued in Congress and in statehouses, not a completed rebrand. The distinction matters because the politics of infrastructure are settled in permits, rate cases, zoning boards and ballot measures rather than in vocabulary. The buildings, the power demand, the transmission lines and the water use would remain the same under either label.
Lehane compared the coming political moment to Franklin Roosevelt’s 1932 election, which he described as the last peacetime presidential race held during a change on this scale. He warned that a candidate who backed a data centre moratorium six months from now could see America lose its AI lead to China for want of computing power. He called that prospect possibly the 2028 version of the 2002 vote to authorise the war in Iraq. The essay’s five questions cover safety and human control, taxes, children and jobs, affordability, and the race with China.
Sam Altman’s company is therefore arguing that the country needs a vocabulary for growth as well as rules for restraint. The proposed label carries a claim that these sites differ from the warehouses and server rooms that preceded them, and that their construction and operation represent a new kind of industrial project rather than an ordinary commercial development. That claim will be tested by communities that have to live with the physical consequences.
What Voters Rejected in November 2026
The timing was not accidental. In 40 of 41 local ballot measures on the November 3 ballot, voters rejected new data centres. Democrats won control of both chambers of Congress and the governorships of Texas, Georgia and Ohio. Sixteen states had at least one gubernatorial candidate backing a data centre moratorium, and The Information reported that candidates backing such a position could win up to 15 of the 36 gubernatorial races.
The scale of stalled investment gives the opposition more weight than the ballot counts alone suggest. Data Center Watch counted 45 projects worth 88 billion dollars already stalled by local opposition in the second quarter of 2026. Those three numbers tell different stories. The ballot result measures direct voter rejection of new data centres. The state figure counts candidates who pledged a moratorium, which is a position rather than a completed policy. The stalled project total measures capital held up by local action. None of the three should be treated as the same measure of public sentiment, and together they explain why OpenAI chose to speak before new members are sworn in rather than after.
For OpenAI, that produces an awkward timing problem. Its infrastructure plans depend on decisions by state legislatures, county commissions and utility regulators, several of which now face officials elected on a platform of greater scrutiny. A communications campaign can start in the lame-duck session. Permits and power contracts cannot.
The Pike County Model and Its Limits
The model OpenAI points to is the PORTS-Pike Technology Campus in Pike County, Ohio, on the site of a former uranium enrichment plant. SB Energy builds, owns and operates the facility and leases about 8 gigawatts of IT capacity to OpenAI for 20 years, with at least 10 gigawatts of new generation planned, most of it natural gas. SB Energy has said it will absorb the full cost of the grid upgrades and new transmission lines so those costs do not reach Ohio ratepayers. OpenAI matched SB Energy’s 40 million dollar community fund with 40 million dollars of its own. Those are two separate contributions of equal size, not one larger payment.
OpenAI says the compute campus in Piketon, Ohio alone will need 25,000 skilled tradespeople. The earlier Ohio promise of 2,500 positions, alongside local payments such as 4.8 million dollars for school lunches in Georgia, gives the community compact a scale against which to judge. Compared with planned project costs of more than 30 billion dollars in Georgia, those contributions are modest.
The point of the community compact and compute campus policy blueprint, published by OpenAI on November 5, 2026, is that voluntary commitments are meant to become a template for binding rules across the industry. The documents are voluntary commitments and a proposed policy blueprint, not enacted law. Even so, the scope is limited. The pledges apply to future sites OpenAI owns and operates itself, while much of its capacity is built and held by partners such as Oracle, SoftBank and Foxconn. OpenAI has also promised to stop requiring nondisclosure agreements during negotiations with local authorities, while continuing to protect trade secrets, security information and competitively sensitive material.
Speed Limits Without a Bill
On safety, Lehane’s memo calls for enforceable safety standards, independent evaluations, public oversight and the ability to restrict or shut down systems. He writes that AI is a machine, not a person, and argues that people should retain control over consequential decisions. He cites OpenAI’s own analysis finding likely Chinese Communist Party influence operations posting social media comments claiming that AI buildouts were raising household electricity bills.
The speed-limits plank has no published threshold, no trigger and no named enforcer. There is no bill, no legislative text and no sponsor. TechCrunch reported OpenAI did not respond to requests for further specifics. The proposal is an idea delivered into a Congress whose composition has just shifted against the industry, with Senator Josh Hawley, Senator Richard Blumenthal, Senator Adam Schiff and others pressing questions about power use, safety and accountability. The GRID Act, SB 53 and the New York RAISE Act are among the proposals circulating in that environment, and California’s SB 1047 remains a reference point for what enforceable standards look like in legislative form.
Kate Brennan, who leads AI and corporate power policy at the AI Now Institute, told TechCrunch the New Deal metaphor is an invocation of state capacity written by a private company advocating for its own benefit, ahistorical at best and harmful at worst. She said every proposal is predicated on benefiting from AI’s upside, which has yet to be seen, and noted that in 2024 OpenAI opposed California’s SB 1047, which would have mandated catastrophic risk evaluations and shutdown capabilities.
Industry voices have not been uniform. Watershed and StartupFortune have framed the fight over terms as a contest over public understanding, while Dario Amodei and Anthropic have argued along different lines about the pace of development. METR research on model capabilities and NPR reporting on household bill concerns have given lawmakers material of their own. Business Insider, Politico, The Information and The New York Times have each tracked the industry’s effort to court a more demanding political environment.
What Has to Change Beyond the Label
Affordability is likely to decide the argument. Consumer Reports cited a federal analysis showing residential electricity rates rose 7.3 percent between April 2025 and April 2026, and NPR reported that 70 percent of Americans are worried data centres will raise their bills. Those numbers give the proposed AI compute campus label its central test. A name cannot answer a rate increase, and a compact that applies only to sites OpenAI controls leaves the largest developers outside its scope.
OpenAI’s own commitments offer a partial answer. The company says it will cover the full cost of electricity and grid upgrades at its sites, throttle consumption during peak hours, and disclose construction plans, noise studies and water usage data. It says regular hardware upgrades will create permanent jobs, and that communities should hold companies responsible and accountable for transformative benefits. OpenAI has also said it would be open to independent evaluators embedded in the company, a shift from its earlier position of limiting state regulation.
Whether those promises persuade depends on who verifies them and who enforces them. The Electric Power Research Institute, state regulators and local officials will carry more weight than any press release. The absence of a published threshold, a trigger and a named enforcer for speed limits is the clearest sign that the policy push is an agenda-setting exercise rather than an implementation plan.
For now, the vocabulary campaign arrives before the legislation it hopes to inspire. OpenAI has offered Congress a set of questions, a model site in Pike County, Ohio, and a proposal to call its infrastructure AI compute campuses. Voters have already delivered an answer about labels, local control and electricity bills, and it was not the one the industry wanted. The argument over AI compute campuses will be settled by what gets built, what it costs and who answers for it, not by the name on the application.