Key Takeaways

  • PJM Interconnection asked the US Department of Energy to move data centers to backup generation to ease grid strain.
  • Record heat is pushing electricity and water systems to their limits as AI data center construction accelerates.
  • Lawmakers across the political spectrum are urging stricter oversight of siting, cooling, and resource use.

The United States is in the middle of a severe heatwave, exposing the growing tension between artificial intelligence infrastructure growth and resource constraints. PJM Interconnection's request that the US Department of Energy require data centers to switch to backup generators within 15 minutes of an emergency alert landed as parts of the East Coast expected heat indices above 38.9 degrees Celsius. This directive highlights acute pressure points on the national grid.

A large share of the country's data centers sit in regions that were already monitoring electrical grid capacity closely. According to a recent Gallup survey, 7 in 10 Americans say they do not want a data center built in their community, citing worries about electricity and water usage. The tension aligns with current demand metrics: data centers account for 4% of national power demand today, a figure the US Department of Energy projects could reach 9% by 2030. Background stress quickly transitions into active strain during prolonged extreme heat.

The International Energy Agency estimated U.S. data centers consumed 183 TWh of electricity in 2024, representing more than 4% of national use. A typical hyperscale facility draws between 100 and 300 megawatts, enough to match the load of a small city or hundreds of thousands of homes. The strain extends to water resources. Evaporative cooling systems often do not recirculate water, losing nearly 80% of their potable water input to evaporation. During a heatwave, these cooling systems draw even more heavily on local supplies.

Additional analysis from the International Energy Agency highlights the rapid acceleration of electricity demand linked to AI clusters and large-scale compute buildouts. Federal grid planners note that load forecasts are unlike anything seen in recent decades. For a parallel perspective on infrastructure resilience, several studies surfaced by IEEE examine how thermal stress events complicate reliability across dense compute systems. These assessments indicate that extreme heat acts as a force multiplier for existing infrastructure constraints.

Political reactions to this technical challenge have been unusually aligned. Texas Governor Greg Abbott proposed banning data center construction in rural areas, stating that facilities should generate their own power and reuse water. On the other end of the political spectrum, Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez have called for a moratorium on new data centers altogether.

Utility providers are sending stronger signals regarding capacity limits. Most major grid operators report that electric load is growing after years of plateauing demand, driven by AI data centers, electric vehicles, and new manufacturing plants. In response to climbing load forecasts, some utilities are delaying the retirement of older power plants. This shift raises persistent questions about how quickly new transmission lines can be built in regions where permitting remains a slow process.

Approximately 38% of Americans live within 8 kilometers of one of the 3,000 data centers currently operating. Under typical conditions, cooling represents up to 40% of a data center's electricity consumption, and this share climbs as ambient temperatures rise. A University of Cambridge study found that local heat islands around data centers can push ambient temperatures upward by 2 degrees Celsius on average, reaching up to a 9.2-degree uplift in specific scenarios. While highly localized, these temperature increases exacerbate equipment strain during broader regional heatwaves.

Water usage presents a parallel constraint. Nationally, data centers use an estimated 627 million gallons of water per day. While modest compared with agriculture or heavy industry, the demand is highly concentrated. Two-thirds of new facilities planned or built since 2022 sit in regions already dealing with water scarcity. A single large AI data center can consume up to 5 million gallons of water daily. In municipalities like Corpus Christi, where reservoirs fell below 10% capacity after a 5-year drought, integrating another large-scale facility presents severe resource allocation dilemmas.

Events in California and Virginia illustrate the operational realities of these constraints. Near Lake Tahoe, roughly 50,000 customers were instructed earlier this year to find alternative power providers due to growing demand linked to data centers. In Virginia's Henrico County, home to 37 facilities, local officials recently asked schools to reduce electricity usage. These localized power crunches illustrate the growing competition for constrained infrastructure resources.

Data center operators and cloud providers are attempting to mitigate these resource constraints. Many operators reference ASHRAE thermal guidelines and the Uptime Institute Tier standards when designing new sites, aiming for more efficient cooling or hybrid systems that reduce water reliance. Companies like Microsoft, Google, and Amazon Web Services continue to test new thermal management techniques, though large-scale adoption remains uneven across the industry.

The current heatwave functions as a real-time stress test for utility networks and public policy. The forces driving the expansion of AI infrastructure are fundamentally misaligned with the pace of grid and water system modernization. Future development will largely depend on how quickly infrastructure planners and data center operators can adapt to a baseline environment characterized by persistent resource constraints and extreme weather events.