OpenAI published a National Community Compact and a separate policy blueprint on November 5, 2026, placing electricity bills, grid investment, water use and local accountability at the center of how future AI campuses should be approved. The distinction between the two documents is central: one makes voluntary promises for a defined class of OpenAI projects; the other asks lawmakers and utility regulators to turn similar expectations into generally applicable rules.
In its official announcement, OpenAI says every future campus that it owns and operates will have a locally tailored agreement. That wording matters because it does not, by itself, establish that every third-party or partner-built facility serving OpenAI is covered.
What OpenAI is promising—and what it is not
| Issue | Company commitment or proposal | Key limitation |
|---|---|---|
| Electricity and infrastructure | Pay full campus electricity and associated infrastructure costs; provide assurances against project failure | Voluntary commitment for future OpenAI-owned and operated campuses |
| Water | Site-level water plans, low-water cooling priority and public usage reports | Actual use must be measured against published plans |
| Construction and noise | Publish construction plans and independent noise studies | Disclosure is not itself a permit or local enforcement mechanism |
| Public-official NDAs | Avoid NDAs, except for narrow security and legitimate confidential information | Exceptions still require scrutiny |
| Industry-wide electricity tariffs | Ask state commissions to require large-load customers to pay incremental grid costs | Policy recommendation, not an enacted nationwide rule |
The central financial question: who bears stranded-grid risk?
An AI campus can require new transmission, substations, generation contracts and other infrastructure before it generates meaningful revenue for an electricity supplier. If the campus is delayed, scaled down or canceled, the assets can remain on a utility’s balance sheet. Unless the developer carries enough contractual liability, part of that cost could ultimately be allocated to ordinary electricity customers through regulated rates.
OpenAI says it will provide financial assurances before major spending begins to protect other customers in such cases. Its blueprint calls for large-load tariffs requiring major customers to cover the incremental costs they create. The practical test is the contractual detail: minimum payments, collateral, cancellation charges, the length of supply commitments, and which expenditures the state utility commission allows into general rates.
How this relates to regulated utilities
Utilities still have an incentive to serve large new customers because long-term demand can justify investment and produce earnings growth. Duke Energy, for example, reported third-quarter results on November 5 and continues to discuss significant power-system investment and higher electricity demand across its territories. The company’s official earnings release documents the reporting event, but the precise contract economics and ratepayer treatment of any project should be checked in relevant tariff filings rather than inferred from optimistic management descriptions.
For investors, the distinction is between economic demand and permitted, financed demand. A technology company may have the willingness to buy electricity without securing the regulatory approvals, grid capacity or community acceptance needed to turn a proposed campus into operating load.
Project maturity now matters as much as AI spending guidance
Equity markets often treat chipmakers, data-center developers, cloud operators and regulated utilities as one AI growth theme. They face different bottlenecks. Existing cloud operators may benefit from scarce computing capacity and repricing, while projects still awaiting permits, utility interconnection or financing may face delays and higher costs. Regulated utilities face the separate question of whether investment costs are recoverable from a clearly identified customer or from the broader customer base.
OpenAI’s announcement is therefore a signal about the market’s next stage, not proof of an immediate capex cut. The company has not announced that its commitments represent a federal law or that the commitments have already been implemented at every facility used by its services.
What would prove the compact works?
Investors and residents should look for published site-level power and water data, named large-load tariffs approved by state commissions, disclosed financial assurances, completed independent noise studies, and project-specific evidence of local input before irreversible construction begins. The first meaningful outcomes will likely be measured in contracts, permits and reports rather than press releases.
OpenAI’s blueprint also cites an arrangement at its planned Effingham County, Georgia, campus under which Georgia Power can reduce its delivery by up to 1,000 megawatts during periods of high demand. That provides a useful model for demand flexibility, but it does not establish that every AI project can offer the same dispatchability or that such flexibility eliminates grid costs entirely.