Chris Lehane had a phrase for it. The policy blueprint his company published on November 5, he told reporters, takes the pitfalls OpenAI has already stepped in and fixes them into a floor.
The document is short on novelty and long on concession. OpenAI restated that it will pay, in full, the electricity bills for data centers it owns and operates. It committed to consult communities before choosing a site, to publish construction plans, to ensure residents do not pay more for grid upgrades the projects require, and to stop signing confidentiality agreements with local officials. The same blueprint recommends that states write those four commitments into law.
The timing is the story. Two days before the document went out, CNBC counted 41 local ballot measures across 10 states that touched data center development. All but one rejected either the project in front of voters or the provisions that would have made it easier to build. That is not a market signal. It is a political one, and it arrived at the moment the industry needs more power than the grid can supply on the schedule it wants.
The electricity clause is the one worth reading closely, because it addresses the grievance that has proved hardest to answer. A data center draws power like a factory and employs people like a warehouse. The transmission lines and substations built to serve it are typically paid for by every customer on the utility’s books, which means a household sees a line item it did not ask for. Utilities call the practice cost allocation. Residents call it the reason their bill went up, and regulators have struggled to argue with them.
OpenAI’s promise, as written, covers only the sites it owns and operates. Much of the industry’s capacity is leased from third parties, and the power arrangements at those sites are the landlord’s problem, not the tenant’s. The company did not say how its commitment travels to a building it rents, or whether it will insist on the same terms in contracts with developers. That gap will be found by the first county board that asks.
The promise not to sign confidentiality agreements with local officials is a smaller item with a larger effect. Secrecy is what turns a data center from an economic development proposal into a rumor, and rumors are what fill a school gymnasium on a Tuesday night. Utilities and developers have long used nondisclosure agreements to keep load forecasts and tax abatements out of public view until the deal is done. Removing that tool from OpenAI’s own playbook does not remove it from everyone else’s.
Then there is the request for legislation. A company asking states to bind it is unusual, and the motive is not mysterious. Voluntary standards are a marketing expense. Statutes are a compliance cost, and a compliance cost falls on competitors who have not volunteered. If a state writes these four commitments into code, every developer building there pays for community consultation, publishes its plans, and eats the grid upgrade. OpenAI, which can afford all three, gets a rule that raises the price of entry.
Whether legislators take the offer is another matter. The measures that passed on November 3 were not arguments about cost allocation. They were votes against the projects, and a promise to pay the power bill does not answer a neighbor who does not want the building.
Lehane’s phrasing was careful. A floor, not a ceiling, and one built from mistakes rather than foresight. Companies rarely describe their own record that way, and the candor is worth noting because it concedes the premise of the opposition: the industry’s early projects did impose costs on places that had no standing to refuse them.
The grid is where the promise meets its limit. Utilities plan capacity years ahead on forecasts that are built from the requests developers file, and a single campus can double what a county has historically drawn. Those requests are not commitments, and a utility that builds for a project which never breaks ground has spent its customers’ money on steel that serves nobody. Regulators have begun asking developers to post collateral against their load forecasts for exactly that reason, and none of the four commitments in the blueprint addresses it.
The commitments also arrive with no enforcement mechanism attached. There is no audit, no filing, no penalty, and no definition of what consulting a community consists of. A promise to publish construction plans is only as good as the definition of a plan.
What the blueprint does not address is the constraint that no policy can lift. Grid interconnection queues in the regions where these campuses are being built run years long, and the equipment to shorten them is in the same shortage as everything else in the supply chain. A company can promise to pay for a substation. It cannot promise the transformer arrives sooner.
The ballots have already been counted. The next test will be a county commission meeting where a developer arrives with a signed letter from OpenAI, and the room has to decide whether a promise made in a press release is worth more than the one it rejected at the polls.
