Industry observers warn the electrical grid cannot cope with data center buildout, as demand outpaces generation and transmission infrastructure expansion.
As June turned to July, with a heat dome bearing down on the Eastern Seaboard, the US Department of Energy issued an order that would have seemed unthinkable a decade ago. It authorized PJM—the grid operator serving 67 million Americans from Chicago to Washington—to order data centers in Virginia and Maryland to disconnect and run on diesel power if the system came under sufficient strain, freeing capacity so others could keep their air conditioning on. Any facility drawing 50 megawatts or more would have just 15 minutes to comply. It was framed as a last resort before rolling blackouts. Secretary of Energy Chris Wright stressed that keeping power on was "non-negotiable."
The blackouts never materialized. Two days later, PJM forecast a peak demand of just over 166,000 megawatts—breaking a record that had stood since 2006—and the grid held. Yet the near miss itself tells the real story.
For roughly two decades, American electricity demand had been flat: efficiency gains offsetting consumption growth meant utilities built very little. That era has ended. Grid Strategies, a consultancy tracking demand curves, now forecasts American peak demand will rise by around 166 gigawatts by 2030. Just a few years ago, the five-year forecast was 24 gigawatts. The figure has risen sixfold, and the single largest driver is the data center.
The artificial intelligence boom demands relentless electricity consumption from vast processor halls. In 2023, data centers consumed approximately 4.4 percent of American electricity; a federal laboratory projects this will reach between 7 and 12 percent by 2028. On more aggressive projections, data centers alone will account for roughly half of all growth in American demand through the end of the decade.
The real constraint lies upstream: in generation capacity and transmission infrastructure. Over 2,000 gigawatts of new power plants are queued to join the grid—more than the existing fleet—yet only roughly one in seven projects ever gets built, after a typical wait of five years. New transmission lines face even longer delays, held up by permitting processes, litigation, and opposition from communities objecting to pylons in their neighborhoods. Even the gas turbines needed for on-demand generation are backordered into the 2030s.
Data centers destabilize the grid not just through raw appetite but through behavior. Engineered to protect their processors, they instantaneously activate their own generators at the first sign of trouble. One Virginia cluster shed 1,800 megawatts in moments last year. The June order essentially formalizes what these machines already do independently—except the government now decides when they disconnect.
The government has pursued one other significant tool: preventing the closure of aging plants. Since May of last year, the Energy Department has issued over 40 emergency orders under Depression-era statute, forcing coal and oil plants slated for retirement to keep operating. The J.H. Campbell coal station in Michigan exemplifies this approach. Due to close at the end of May last year, it has been kept running by federal order ever since—despite the regional grid operator never requesting this intervention and the state's own regulator flatly declaring no energy emergency exists in Michigan.
Yet keeping half-century-old coal plants online offers no genuine solution to the underlying supply problem. The plants that can be built fastest and that dominate the connection queue are overwhelmingly solar, wind, and battery systems—roughly 19 of every 20 projects—precisely the technologies the current White House favors least.
Securing adequate guaranteed capacity is expensive, and households are bearing the cost, not technology companies. In the PJM region, annual capacity costs surged from $2.2 billion to $14.7 billion in a single year. Complicating everything are increasingly volatile weather patterns. Winter storms that struck Texas in 2021 and the eastern grid in 2022 killed over 300 people combined—the fifth major cold-weather grid crisis in 11 years. Jim Robb, CEO of the North American Electric Reliability Corporation and the man responsible for North American grid reliability, characterized the mounting risks last fall as "a five-alarm fire."
This is not an American problem alone. In Ireland, data centers already consume roughly 22 percent of the nation's electricity—Europe's highest share—forcing Dublin to freeze new grid connections for years. Last summer, as France endured its hottest weather on record, EDF was forced to throttle nuclear reactors because the rivers used to cool them grew too warm. The pattern repeats everywhere: demand races ahead of generation and transmission capacity, and the physical world resists being hurried.
Britain exemplifies the same dynamic in more advanced form, without compensating growth. It has the highest industrial electricity prices in the developed world—roughly four times what American industry pays, a reality quietly decimating what remains of British manufacturing. Its grid connection queue has swollen past 700 gigawatts, four times actual demand, with some projects delayed until the 2040s. The system operator has begun, sensibly, dismantling the queue and rebuilding it. Yet the situation is stark: Britain is competing to attract the very data centers America cannot power, attempting to place them on a grid that cannot clear its own backlog while charging extremely expensive rates.
The diagnosis is straightforward, however uncomfortable. America is not short of power it might generate, but has simply lost the knack of building the things that deliver it. The queue, the pylon, the turbine, and the permit are the binding constraints—each self-inflicted. The remedy is correspondingly unglamorous: prioritize firm power in the connection queue, build transmission infrastructure across regions, and stop retiring dependable plants until something dependable stands ready to replace them.