Earlier this year, federal energy officials were bracing for a brutal summer. The nation’s largest power grid, PJM Interconnection, which delivers electricity to roughly 65 million people across 13 states from New Jersey to Illinois, faced projections showing that a wave of artificial intelligence data centers could push demand past safe limits during the hottest weeks of the season. Emergency powers were invoked. Reliability warnings were issued.
Then something unexpected happened: the data centers didn’t show up on schedule.
Interconnection delays, permitting bottlenecks, equipment shortages, and financing hurdles have slowed the pace at which AI-linked facilities are actually drawing power from the grid. The result, heading into June 2026, is a summer outlook that is still tight but meaningfully less dire than the worst-case scenarios regulators were planning around just months ago.
Federal officials prepared for the worst
The clearest sign of how seriously Washington took the threat came from the Department of Energy. Earlier in 2026, the DOE issued PJM Interconnection Order No. 202-26-02 under Section 202(c) of the Federal Power Act, a rarely used emergency provision that allows the Energy Secretary to compel generators to keep running when grid reliability is at risk. The provision has historically been reserved for genuine crises, such as extreme weather events or sudden plant failures, making its preemptive use for summer planning a notable escalation.
The Federal Energy Regulatory Commission reinforced the concern with its own 2026 summer energy market and reliability assessment, which flagged supply adequacy risks during peak demand periods. That assessment built on warnings the North American Electric Reliability Corporation had been issuing since its 2024 Long-Term Reliability Assessment, which placed parts of PJM’s territory in an elevated risk category due to shrinking reserve margins and accelerating coal and gas plant retirements.
Taken together, these actions amounted to a clear signal: the people responsible for keeping the lights on were not confident the grid could handle what was coming.
Why AI data centers stalled
The load growth that alarmed grid planners was real. Over the past two years, hyperscale developers tied to companies like Microsoft, Amazon, Google, and Meta flooded PJM with interconnection requests, many of them for facilities designed to train and run large language models. Northern Virginia, already the world’s densest data center market, saw a particularly sharp surge, along with emerging clusters in Ohio and Indiana.
Those requests reversed a decade-long trend. U.S. electricity consumption had been essentially flat since the mid-2010s, according to the Energy Information Administration. Suddenly, PJM’s planners were modeling load growth curves that looked more like a developing economy than a mature grid.
But requesting a grid connection and actually pulling power are very different things. PJM’s interconnection queue, which had already swollen to more than 2,500 projects totaling hundreds of gigawatts by late 2024, became a bottleneck in itself. Studies that once took months stretched into years. Meanwhile, developers ran into practical obstacles: high-voltage transformers with 18- to 24-month lead times, local permitting fights over land use and water consumption, and financing conditions that tightened as interest rates stayed elevated.
The net effect is that a significant share of the data center megawatts utilities expected to serve by summer 2026 have slipped into 2027 or beyond. Each megawatt that doesn’t materialize this year is a megawatt the grid doesn’t have to cover during a July heat wave.
How much relief, and how certain?
The directional trend is clear, but the precise scale of the reprieve is harder to pin down. PJM has not published a granular comparison of projected versus actual data center megawatt additions for 2026 in a format that allows independent verification. Utility-level load forecasts, which would show how individual service territories have adjusted their summer peak expectations, remain largely internal planning documents.
What is publicly visible is the gap between the emergency posture regulators adopted in early 2026 and the somewhat calmer tone in more recent industry commentary. The DOE’s Section 202(c) order and its extension materials remain in place, but the order functions as a backstop rather than an active deployment. FERC’s summer assessment still flags risks, yet the most alarming scenarios, widespread rolling blackouts driven by an AI load surge, have receded from the near-term conversation.
It is also worth noting what has not changed. The structural pressures on PJM’s grid persist. Coal plants continue to retire on schedule. New natural gas generation faces its own permitting and financing headwinds. Transmission expansion, the kind of large-scale line construction that would unlock new supply corridors, remains a multi-year process even under the best circumstances. The data center slowdown has bought time, but it has not solved the underlying supply-demand imbalance.
The delayed demand doesn’t disappear
Grid planners face an uncomfortable reality: data center projects that miss their 2026 targets do not vanish from the queue. They roll forward, potentially concentrating years of demand growth into a compressed window. If a large batch of delayed facilities energizes simultaneously in 2027 or 2028, the supply pressure could return in a form that is harder to manage than a gradual ramp.
Distinguishing genuine project cancellations from temporary delays is difficult with the data currently available. Some developers have publicly scaled back ambitions or shifted to regions with more available power. Others are simply waiting for equipment deliveries or permits. Until PJM publishes updated queue withdrawal and completion statistics, the split between dead projects and delayed ones will remain an open question.
For policymakers, this creates a planning dilemma. Building new generation, storage, and transmission to meet demand that may arrive in 2027, or may slip again to 2029, requires committing billions of dollars on uncertain timelines. The federal emergency tools currently in place were designed as bridges, not permanent solutions, and regulators will need to decide how long those bridges can hold.
What households and businesses should expect
For the roughly 65 million people living in PJM’s footprint, the practical takeaway is cautiously positive. The risk of rolling outages tied specifically to an AI-driven demand spike is lower this summer than it appeared when emergency orders were first being prepared. That does not mean the risk has vanished. A prolonged heat wave, an unexpected plant trip, or a transmission failure could still push the system to its limits on any given afternoon.
Utilities and grid operators will still call for conservation during extreme heat. Voluntary demand response programs, including smart thermostats that nudge temperatures up a few degrees during peak hours, remain a frontline tool for managing tight conditions. Businesses on interruptible rate programs should expect continued pressure to shift usage away from late-afternoon peaks, when air conditioning load collides with declining solar output.
The broader picture is one of a grid that caught a break it did not plan for. The AI construction boom that was supposed to collide with a sweltering 2026 summer has instead stumbled over the same permitting, supply chain, and financing obstacles that slow nearly every large infrastructure project in the United States. That stumble has given grid operators a narrow window of additional headroom, but the demand is still coming. The question is no longer whether AI will reshape the power grid, but whether the grid can build fast enough to meet it when it finally arrives.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
