Extreme Heat and Severe Weather Emerge as Threats to AI Data Centers

A record European heatwave is highlighting a growing threat to the AI boom: extreme weather that stresses data centers and the power grids they depend on at once.

By Olivia Grant Edited by Maria Konash Published:
Insurers and operators warn that extreme heat and severe weather are becoming major risks to AI data centers and power grids. Image: Luis Graterol / Unsplash

A record-breaking heatwave across Europe is spotlighting a mounting risk to the AI build-out: the extreme weather that threatens the data centers and power grids underpinning it. High temperatures strain electrical infrastructure and spike air-conditioning demand, which can overload grids and trigger blackouts.

The problem is not confined to Europe. Insurers and operators say severe weather has shifted from a background concern to a central one as the industry pours hundreds of billions of dollars into facilities that are increasingly exposed to the climate.

The financial signal is already visible in insurance data. Over the past three years, severe weather has become the leading cause of loss in Zurich Insurance’s US data center builders’ risk portfolio, now driving about a third of losses, the company’s head of international construction Patrick McBride told CNBC.

Part of the reason is location. Many new data centers are rising in suburban and rural areas where land is cheap but historical weather records are thin, leaving operators with large, concentrated assets in poorly understood hazard zones. A study by climate analytics firm First Street found that 79% of global data center capacity faces elevated risk from acute hazards such as flooding, extreme winds and wildfires, while 54% sits in areas under chronic stress from heat and drought.

The build-out is accelerating that exposure rather than reducing it. McBride said 64% of capacity under construction this year is outside traditional hubs like Northern Virginia, moving into frontier markets such as West Texas, Tennessee, Wisconsin and Ohio, where tornadoes, hail and high winds can batter the vast roofs, cooling towers and solar installations these sites depend on.

Emerging markets carry their own risks: Brazil faces heat challenges, and in Europe data centers are migrating toward the warming Iberian Peninsula. Marsh Risk’s Joe Macejak warned that poorly managed climate risk could threaten the “capital stacks” financing the data center revolution.

The Compounding Problem

The sharpest danger is that heat hits data centers and grids simultaneously. Cooling already accounts for roughly 40% of a data center’s energy use at normal temperatures, and that share climbs during heatwaves, exactly when air conditioning is straining the grid.

As Rhizome chief executive Mishal Thadani put it, data centers need the most power precisely when the grid has the least to give. He pointed to Turin, Italy, where May highs near 38 degrees Celsius put underground cables under thermal stress and caused repeated blackouts. A single large facility can draw as much power as 100,000 homes, so heat and surging load hit the same wires at the same moment, a dynamic most planning models still underestimate.

How Operators Are Adapting

The industry is responding on two fronts: where it builds and how it cools. Microsoft said it designs facilities to run across a wide range of conditions, using site selection, redundant systems and real-time monitoring to manage heat and severe-weather risk.

On the hardware side, Nvidia said last week that its newest AI servers can run cooling liquid at 45 degrees Celsius, higher than before, which matters because raising chiller temperatures by even one degree can cut cooling energy use by about 4%.

Insurers, meanwhile, are pushing the sector to treat climate as a core underwriting input rather than an afterthought, since outdated models built on historical data increasingly fail to capture how often extreme heat now arrives. How well operators adapt will shape both the reliability and the economics of the AI era.

AI & Machine Learning, Cloud & Infrastructure, News
Exit mobile version