I work near Northern Virginia’s “Data Center Alley,” where massive windowless buildings hum behind security fences, quietly powering our digital lives. These facilities handle everything from streaming services to banking systems. Most are designed using a dangerous assumption: that tomorrow’s weather will look like yesterday’s.

It won’t.

In 2022, London hit 40°C (104°F) for the first time in recorded history. Google Cloud and Oracle data centers simultaneously went dark. Not from a cyberattack or equipment failure, but from heat. A California heat wave took down another major facility months earlier. We called these events “unprecedented,” but they weren’t. The ten hottest years on record all occurred between 2014 and 2024.

And yet, we’re still building 25-year infrastructure using historic weather data from a climate that no longer exists.

The heat problem

Server racks generate intense heat. Cooling systems consume roughly 40 percent of a typical data center’s electricity budget. That share climbs as AI and high-performance computing push rack densities higher.

Most capacity models add new server loads to historical cooling requirements, assuming the hottest day from the past three decades represents the worst we’ll face in the next three.

That assumption creates costly blind spots. Heat waves last longer now, and nights stay warmer. Equipment works harder for the same result. Heat hammers electrical grids precisely when data centers need them most. During extreme temperatures, households max out air conditioning. Transmission lines lose capacity as they overheat. Data centers require maximum electricity exactly when grids deliver minimum reliability. Goldman Sachs projects data center power demand will jump 165 percent by 2030.

Wildfires now burn year-round in some regions. Smoke infiltrates air systems and corrodes equipment long before flames arrive. Precautionary power shutoffs can last for days. Water scarcity creates another trap. Evaporative cooling can consume five million gallons daily. You can’t cool servers with water that doesn’t exist.

Hurricanes are growing stronger and wetter. These cause the majority of long-duration power outages. When grid failures combine with physical damage, facilities stay dark for extended periods.

Outdated maps

Most planning meetings rely on flood maps based on historical rainfall patterns. These maps are dangerously outdated. Warmer air holds seven percent more water vapor per degree Celsius of warming. Storms are intensifying.

We recently worked with a global technology firm to evaluate potential data center sites in a hurricane-prone region. The engagement combined historical records with high-resolution climate projections and synthetic storm modeling. The analysis showed that rainfall events currently classified as 100-year storms could occur every 35 to 45 years within a facility’s operational lifespan.

By screening more than 200 candidate counties and selecting a handful of inland sites with lower projected hazard exposure, the client could reduce estimated downtime and storm-related capital expenditure. The design enhancements were projected to avoid more than $10 million in outage costs over 25 years.

Low carbon pressure turns weather into a deal term

Physical risk tells only part of the story.

Transition risks relate to policies, technology, reputation, and, importantly, market shifts. Pressures from customers to transition to a low-carbon economy now shape contracts, siting, and design.

The largest customers of data center capacity sit in big tech, and big tech champions strict emissions targets. For many data center customers, purchased cloud services and colocation sit inside Scope 3 emissions accounting. That makes data center operations a key factor in corporate climate plans.

A high-emissions facility raises a customer’s reported carbon footprint and raises the cost of meeting targets. A low-emissions facility does the opposite. Procurement teams already ask for hourly carbon data, renewable sourcing details, and credible paths to lower emissions intensity over the contract term. They ask for proof that the building can run on clean power at scale, not only certificates on paper.

Operators who treat emissions intensity as a design constraint win trust, secure contracts, and protect long-term value. Operators who ignore it face tougher terms, shorter contracts, and costly retrofits.

The path forward

We have the climate data and tools to meaningfully anticipate future conditions and design for them. The best time to manage these risks is during site selection, before any foundation is poured. Teams already weigh land costs, fiber routes, and tax policy. Climate hazards belong on that list with equal weight: heat exposure, flood pathways, wildfire risk, water stress, and storm vulnerability.

Boards and investors should demand climate-adjusted design criteria for every major project. Utilities and local governments should integrate data center growth into grid planning. Operators should treat water and energy as operating constraints with full public visibility.

The data centers we build today will operate until 2050 or beyond. They’ll face conditions we’ve never seen. Planning for that reality is basic risk management and fiduciary responsibility. A weather budget belongs in every financial model. The cost is measurable. The payoff is steady service, fewer catastrophic failures, and infrastructure built for the climate we’ll actually have.