The U.S. fight over AI's electricity demand has moved from forecasts to a regulatory clock. On June 18, 2026, the Federal Energy Regulatory Commission ordered the six regional grid operators under its jurisdiction to explain or change how data centers, factories and other large electricity users connect to the transmission system.

The immediate issue is speed and cost. Data center developers want faster access to power, while utilities, state officials and consumer advocates are watching whether new transmission upgrades get paid for by the large customers that need them or spread across ordinary electric bills.

FERC's orders give grid operators two near-term checkpoints. Within 30 days, around July 18, they must report how they ensure enough generation for existing and anticipated demand. Within 60 days, around August 17, they must either justify current large-load tariffs as reasonable or propose changes.

The order applies to ISO New England, the New York ISO, PJM, MISO, the Southwest Power Pool and California's ISO. Together, those markets cover much of the U.S. population, but not every state or utility customer. That matters because any changes are likely to arrive region by region rather than as one national data-center rule.

FERC identified five areas for grid operators to address: faster study processes, clearer transmission-cost responsibility, treatment of co-located generation, flexible service for large loads that can reduce demand, and study processes for power plants serving nearby large customers. The agency's fact sheet framed the action as a way to support AI and manufacturing while keeping consumer safeguards in place.

Industry coverage since the order has focused on the same tension: data centers can bring investment and jobs, but large, uncertain power requests can also force utilities to plan expensive upgrades before projects are fully committed. IEEE Spectrum noted that the deadlines are ambitious, and Utility Dive reported that FERC officials signaled they could impose solutions if regional operators do not adequately respond.

For readers, the key takeaway is not that a single rule will determine where the next AI campus gets built. It is that electricity planning is becoming part of technology policy. The filings due in July and August should show which regions can connect large loads quickly, which ones will ask those customers to carry more cost, and where local reliability concerns may slow the AI build-out.