AI data centers are turning grid stress into a year-round problem

The US power grid used to have a predictable stress calendar. Summer heat waves and winter cold snaps were the danger zones, and the months in between offered some breathing room.

That breathing room is disappearing. Booming electricity demand from artificial intelligence means supply constraints now extend beyond the two traditional peak seasons, turning grid strain into a year-round affair.

The numbers behind the squeeze

The clearest signal comes from the North American Electric Reliability Corporation, or NERC, the body that monitors the reliability of the continent’s bulk power system. Its January 2026 Long-Term Reliability Assessment forecasts summer peak demand growth of 224 gigawatts over the next decade.

That figure represents a 24% increase from previous projections.

Winter looks even more demanding. NERC projects winter peak demand growth of 246 GW, outpacing the summer figure.

The pressure is especially concentrated in PJM Interconnection, the regional grid operator whose territory stretches across a large swath of the eastern US. PJM is expecting an additional 30 GW of demand from data centers alone by 2030.

That surge has come with rising congestion costs in the PJM region. Congestion happens when transmission lines cannot carry enough cheap power to where it is needed, forcing grid operators to rely on pricier, closer generation.

Washington steps in

On October 19, 2026, the Federal Energy Regulatory Commission issued directives to six regional grid operators, pushing them to reform tariffs that govern how data centers connect to the grid.

FERC asked operators to justify their existing tariffs for these large loads. The concern is that current rules have created bottlenecks in connection timelines.

Those bottlenecks are not minor. Utilities have cited waits of 5–10 years for grid-connected service because of congestion.

The Department of Energy has also intervened. In fall 2026, it authorized emergency operation for certain generation units in the PJM region, giving them more flexibility to respond to demand spikes and maintenance challenges.

Why the old seasonal playbook is breaking

Data centers do not follow the traditional seasonal rhythm. They draw large amounts of power continuously, which raises the baseline demand the grid must carry every day of the year.

Grid operators have voiced widespread concerns about reliability risks that now extend beyond traditional seasonal peaks.

What this means for utilities, developers, and ratepayers

For AI developers and data center operators, the connection queue may matter as much as chip supply. A 5–10 year wait for grid service effectively caps how fast new capacity can come online, regardless of how much capital is available.

The regulatory track is the other key variable. FERC’s October directives could eventually streamline how large loads connect, but the outcome depends on how the six grid operators respond and what reforms survive the process.

Watch for three things in the months ahead: how grid operators answer FERC’s tariff challenge, whether DOE emergency orders become a recurring tool rather than an exception, and how NERC’s next assessment compares with its already steep 2026 forecast.