Climate Change, Future Heat and Data Center Site Risk

Most engineering inputs for a data center are statistics of historical weather. ASHRAE’s climatic design tables give dry-bulb and wet-bulb values that are exceeded a set share of hours in an average year, plus the mean and standard deviation of annual extremes and return-period maximums.⁷ Rainfall design uses NOAA Atlas 14, which NOAA has confirmed does not include the future effects of climate change.⁴ Flood elevations come from FEMA maps of current conditions.

That works when the climate is stable. It is less reliable when the record itself is moving. The Fifth National Climate Assessment (NCA5) found that observed warming over the contiguous United States and Alaska is higher than the global average, and rated that finding virtually certain.¹ EPA’s temperature indicator, built on NOAA data, shows U.S. average temperatures rising more quickly since the late 1970s, at 0.32 to 0.51°F per decade since 1979.² A Congressional Research Service brief updated in July 2026 notes studies finding regional increases in U.S. heat waves from 1981 to 2015 and increases for the contiguous U.S. from 1981 to 2018.³

A campus permitted in 2027 and expanded in phases through the 2030s will operate in a hotter climate than the one its design data describe. This guide covers how to account for that at the site stage. For how climate affects cooling choice and free cooling hours in general, see climate and cooling in site selection.

Mechanical engineers size heat rejection to a design condition. ASHRAE’s 0.4% annual dry-bulb value for Atlanta, for example, is 94.0°F, while the 0.4% values for the hottest months are higher, about 94.8°F for June and 98°F for July.⁸ A data center that must hold inlet temperatures through every hour cannot treat the annual 0.4% value as a ceiling, because by definition it is exceeded about 35 hours in an average year.

ASHRAE also publishes extreme annual statistics and 5-, 10-, 20- and 50-year return-period maximum dry-bulb temperatures.⁷ Many mission-critical designs check equipment selections against one of these rarer values, then add margin. The question climate change raises is whether a historical 20-year value will still be a 20-year event in 2045. The data that set it are a multi-decade average of past weather, and design tables are revised only periodically.⁷⁸

The same NOAA and North Carolina Institute for Climate Studies data project a marked increase in days above 100°F across the Southern Great Plains by 2070–2099 compared with 1976–2005, under both lower and higher scenarios.¹³ For a site in Texas or Oklahoma, that changes how often dry coolers and chillers run near their limits, how much water evaporative cooling draws on peak days, and how much electrical capacity the cooling plant needs at peak.

• Ask the engineer which design value the cooling plant is sized to and how many hours per year it is exceeded today.⁷
• Run the selection at a higher return-period value plus a forward-looking increment, and price the difference.
• Check whether liquid cooling at warmer supply temperatures reduces sensitivity to peak ambient temperatures for the planned rack density.

Heat failures are not hypothetical. On July 19, 2022, as the U.K. reached a record of about 40°C, Google reported a simultaneous failure of multiple redundant cooling systems in a London-area facility; the building could not keep a safe operating temperature given the cooling failure and the outside heat, and Google shut it down.⁹ The same day, Oracle said “unseasonal temperatures” caused a cooling problem in its London region and powered down part of its infrastructure.¹⁰

Heat also stresses the grid that serves the site. NERC’s 2026 Summer Reliability Assessment flagged New England, the Pacific Northwest and Saskatchewan as at elevated risk of shortfalls in above-normal heat, and noted that several areas had lowered load forecasts because large loads such as data centers were connecting more slowly than requested.¹⁵ Hot days push regional demand to its peak at the same time a data center’s cooling load peaks, which is why utilities look hard at large loads in summer studies and why on-site ride-through and backup fuel matter more in hotter places.

The same Uptime survey found that more than a third of respondents’ managements had not formally assessed their data centers’ vulnerability to climate change.¹⁴ Many existing sites were designed to the hazards understood when they were built. For lenders and investors, that is a diligence item, not a forecast debate; see our site risk guide for investors and lenders.

Two kinds of flood risk change with climate. The first is heavy rainfall. NOAA is building Atlas 15 in two volumes: Volume 1 updates current precipitation frequency estimates to account for changes in historical observations, and Volume 2 adds model-based estimates projected into the future.⁴ NOAA has said contiguous U.S. data are expected in 2026, and Volume 1 will supersede Atlas 14 when published.⁴ The project was paused for about a month in 2025 before NOAA received permission to proceed through fiscal year 2026.¹⁶

The second is sea level. NOAA’s 2022 Sea Level Rise Technical Report projects 10 to 12 inches of rise along U.S. coasts from 2020 to 2050, about as much as was measured over the previous century, and moderate (typically damaging) flooding more than ten times as often as today.¹¹ Rise varies regionally because land and ocean heights both change.¹¹

Federal policy moved the other way in 2025. The Federal Flood Risk Management Standard had applied a climate-informed approach to federally funded projects; it was rescinded by executive order on January 20, 2025, and FEMA stopped applying it to FEMA-funded projects effective March 25, 2025.⁵ FEMA noted that the standard never applied to flood insurance rate mapping.⁵ As of July 2026, a proposed rule to remove HUD’s version had been published, while FEMA had not yet proposed rulemaking to formally reverse its own.⁶ For mapped risk today, start with our floodplain guide and the base flood elevation.

Climate change seldom rules a site out on its own, and there is no single correct projection to design to. The goal is to know which assumptions are historical, how sensitive the design is to them, and what it costs to add margin.

1. 01Pull the ASHRAE design and extreme values for the nearest station, and ask which one the cooling plant uses.⁷
2. 02Look at projected hot-day counts for the region and test cooling and electrical capacity against a hotter design point.¹²¹³
3. 03For rainfall, check whether the jurisdiction still uses Atlas 14 and whether Atlas 15 values are available for the site.⁴
4. 04For coastal or tidal sites, overlay NOAA sea level scenarios on site grades, access roads and the utility substation, not just the building pad.¹¹
5. 05Check water: hotter peaks raise evaporative demand at the same time drought can tighten supply; see water stress and drought risk.
6. 06Write the assumptions into the basis of design so later phases are not built to a stale climate.

For the broader hazard picture, including wind, wildfire and seismic risk, see natural hazard risk for data centers. If you want a site tested against these questions before you commit, you can get a site reviewed.