Shortly after midnight, as Great Britain headed into its third heat wave of the year, the national grid operator sent out an urgent request. The National Energy System Operator, known as Neso, asked electricity generators for extra supplies to cover the surge in demand from households switching on fans and air conditioners. That July 2026 warning captures a reality grid planners now face everywhere. Extreme heat pushes electricity demand up while it drags generation down, and the rapid buildout of AI data centers adds a heavy new load right in the middle of that squeeze.

A heat wave squeezes the power grid from both sides

Neso’s notice pointed to forecasts of “tight electricity margins” during the Thursday evening peak. The cause was telling. Extreme temperatures across Europe were reducing the availability of some generation at the exact moment millions of people turned on cooling at home. Parts of southern England were heading toward 34 degrees Celsius, only weeks after a provisional June record of 37.7 degrees was measured in Norfolk.

The supply side of the grid suffers in hot weather in ways that often go unnoticed. Across the Channel, the French utility EDF warned that output could be curbed at up to five of its nuclear plants. The reactors rely on river water for cooling, and when that water gets too warm, production has to slow down. Since France regularly exports power to the UK and Germany, a heat wave in one country quickly becomes a grid problem for its neighbors. Neso stressed that its margin notice is a routine tool and that customers face no supply interruptions. Even so, tight margins mean the operator may again pay far above average prices to gas plants to keep the system balanced.

Why AI data centers raise the stakes for power networks

Data centers form the backbone of cloud services, payment systems and communication. They are also among the fastest growing loads on power networks. In Germany alone, installed capacity has more than doubled since 2010 and passed 2,730 megawatts in 2024. The federal economics ministry expects that figure to climb beyond 4,800 megawatts by 2030, driven largely by cloud and AI applications.

AI changes what happens inside these facilities. A server rack that once drew five to ten kilowatts now regularly pulls 50 kilowatts or more for AI workloads. Nearly all of that electricity leaves the building as heat. The hotter it gets outside, the harder the cooling systems must work, so a data center’s power demand climbs precisely when the grid is already under maximum stress. A June 2026 analysis by the climate risk firm First Street found that 54 percent of global data center capacity sits in markets exposed to chronic heat or drought stress.

The relationship runs in both directions. A March 2026 study involving the University of Cambridge measured a data heat island effect around AI data centers of roughly two degrees Celsius on average, with extreme cases reaching 9.1 degrees. The researchers estimate that more than 340 million people worldwide live close enough to feel that extra warmth. Data centers do not just suffer from heat waves. They can add to them locally.

When cooling fails during a heat wave

The London heat wave of July 2022 offered a preview of what failure looks like. As temperatures in the city passed 40 degrees Celsius for the first time in recorded history, a Google Cloud data center lost several cooling systems at once, forcing engineers to power down parts of its servers to prevent damage. An Oracle facility in south London ran its cooling units at their limit until they failed, causing an outage of about 19 hours that disrupted cloud services around the world.

The risk is growing. The Prometheus study, produced by researchers at Google and the University of Pennsylvania, concludes that data centers must expand their cooling capacity by 11 percent on average by 2044 just to keep today’s outage risk steady. At the most exposed sites, the required increase reaches 48 percent. Backup power belongs to the same problem. An emergency generator also needs cooling to run reliably in extreme heat, and if that cooling fails, the entire site loses its last line of defense.

The water problem behind the power problem

Cooling during a heat wave often means evaporating water, which opens a second front of stress. An analysis of US development plans found that 517 of 809 planned data centers will be built in areas that spent the past year in drought. Large facilities can consume up to five million gallons of water a day, comparable to a town of 50,000 people. Total US data center water demand is projected to reach 73 billion gallons a year by 2028, up from about 17 billion in 2023.

The obvious fix carries a catch. Closed loop cooling systems recirculate their coolant instead of evaporating water, which sharply cuts water use. Yet they require more electricity to run, and that power frequently comes from plants that themselves need large volumes of cooling water. Meta’s planned Hyperion data center in Louisiana illustrates the tradeoff. The site will use closed loop cooling, but its energy supply will require the output of ten power plants that run on gas. During a heat wave, saving water can mean shifting the strain straight back onto the power grid.

How operators can prepare for extreme heat

Heat related outages are not inevitable. Specialists at the consulting firm Drees and Sommer, along with the research cited above, point to practical steps that reduce risk considerably.

Design for tomorrow’s extremes

Cooling systems should be tested against the extremes of the coming 20 years rather than the weather of the past. Forecast tools such as the Extreme Annual Design Conditions from ASHRAE Meteo provide a solid basis, and experts recommend adding a safety margin of four kelvin on top of projected peak temperatures.

Shift flexible workloads to cooler hours

Not every computing task needs to run at the hottest moment of the day. If you schedule backups, data analysis and large data migrations during the cooler night hours, you ease the cooling load exactly when reserves are thinnest.

Protect the backup plan

Emergency power must cover servers and cooling together. A generator that only keeps the servers alive simply delays an outage by a few minutes. Microgrids with their own generation and storage can carry critical systems through disturbances in the public grid.

Match the cooling method to the load

Air cooling remains economical at moderate densities, but racks drawing 50 kilowatts and more often exceed what air can handle. Liquid based systems move far more heat, and hybrid designs let operators convert only the hottest zones. Closed loops keep water consumption low, which matters in regions where water itself becomes scarce during a heat wave.

Regulation is pushing in the same direction. Under Germany’s energy efficiency law, data centers that entered operation after July 2026 must reach a power usage effectiveness of 1.2, and existing sites must hit 1.3 by 2030. Efficient cooling is becoming a legal requirement rather than a design preference.

Heat resilience is now grid resilience

Heat waves expose a feedback loop at the heart of the AI buildout. Hotter air forces more cooling. More cooling demands more electricity. That demand lands on a grid whose own supply is weakened by the same heat. The Neso warnings and the London outages show that this loop is already spinning. The encouraging part is that the cheapest and most reliable unit of cooling is the one a facility never needs. Where data centers are built, how they are cooled and when their workloads run will decide whether AI growth keeps colliding with every heat wave or learns to ride through them.