📄 earth_science

Spatially concentrated heat emissions from data centers lead to localized urban warming and heat exposure

This study integrates facility-level data with urban climate modeling to reveal that concentrated heat emissions from data centers in California create localized warming hotspots, significantly increasing land surface and air temperatures, particularly at night and in densely populated areas like Los Angeles, thereby highlighting the need for informed urban planning and facility siting.

Original authors: Jiachen Zhang, Hao Hu, Arman Shehabi, Cenlin He, Kelly Sanders

Published 2026-07-08

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jiachen Zhang, Hao Hu, Arman Shehabi, Cenlin He, Kelly Sanders

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Data Centers as "Urban Heaters"

Imagine a city as a giant kitchen. Usually, the heat in the kitchen comes from many small stoves (homes), ovens (restaurants), and people cooking (cars and traffic). This study looks at a new, very specific type of stove: data centers.

Data centers are massive buildings filled with computers that never sleep. They are the engines behind the internet, AI, and cloud storage. The researchers found that these buildings are so powerful that they act like giant, concentrated heaters sitting right in the middle of our neighborhoods, warming up the air and the ground around them more than we previously realized.

The Two Ways Data Centers Heat Things Up

The paper explains that data centers heat the environment in two distinct ways, similar to how a car heats up:

The Direct Heat (Scope 1): This is the heat the data center generates itself. Think of it like the engine of a car getting hot while it's running. The computers inside work hard, and almost all the electricity they use turns into heat that is blown out into the street. This happens right where the building is located.
The Indirect Heat (Scope 2): This is the heat generated elsewhere to make the electricity the data center uses. Imagine the data center is a house, and the electricity comes from a power plant 50 miles away. That power plant burns fuel to make power, and in doing so, it releases a lot of waste heat. The study tracks this "upstream" heat back to the data center, even if the power plant is in a different city or state.

The Key Finding: While the indirect heat is spread out over a wide area (like smoke from a factory drifting over a region), the direct heat is super concentrated. It's like having a blowtorch sitting right in the middle of a living room versus a warm oven in the kitchen. The blowtorch (the data center) creates a very intense, localized hot spot right where it stands.

What Happens to the Temperature?

The researchers used a sophisticated computer model (like a high-tech weather simulator) to see what happens when they add this extra heat to California's cities.

The "Hot Spots": They found that the heat doesn't spread evenly across the whole state. Instead, it creates "hot spots" in specific cities where data centers are clustered, like Santa Clara (Silicon Valley) and downtown Los Angeles.
Ground vs. Air: The study found that the ground gets much hotter than the air.
- Analogy: Imagine standing on a sidewalk in the sun. The concrete (the ground) might feel scorching hot to your feet, while the air around you is only slightly warm. The data centers heat the ground (skin temperature) by up to 3.1°C (5.6°F), but the air temperature (what you feel on your skin) goes up by about 0.8°C (1.4°F).
Nighttime is Worse: The warming is strongest at night.
- Analogy: During the day, the wind and sun mix the air up high, so the heat gets diluted. At night, the air is calm and stays close to the ground, trapping the heat from the data centers right where people live and sleep.

Who Gets the Heat?

The study looked at who is actually feeling this extra warmth.

Santa Clara: This area has a high concentration of data centers. About 50,000 people there are exposed to an extra 0.8°C of heat every day in July. It's a small crowd, but they are standing right in the "blowtorch" zone.
Los Angeles: The heat here is slightly less intense (about 0.35°C), but because the city is so crowded, about 100,000 people are exposed to it.
The "Double Trouble": The researchers noted that this extra heat often hits when it's already hot. If the day is already 26°C (79°F), the data center adds another half-degree, pushing people closer to dangerous heat levels. It's like adding a hot blanket on top of a warm summer day.

What If We Build More?

The paper ran a "what if" scenario. If data centers grow to be 3 to 5 times bigger by 2040 (which experts predict they will), the heat won't just get a little warmer; the "hot spots" will get bigger and more intense.

In Santa Clara, the air could get nearly 2°C (3.6°F) hotter in the worst spots.
The area where people feel this heat would spread further downwind, like smoke from a fire spreading to more houses.

The Bottom Line

This study is the first to map out exactly how data centers act as localized heaters in our cities. It shows that while we often worry about the carbon footprint of the internet, we also need to worry about the physical heat it dumps into our neighborhoods.

The researchers suggest that when city planners decide where to build the next giant data center, they shouldn't just look at the electricity bill or water usage. They need to ask: "Is this going to turn this neighborhood into a sauna?" Placing these facilities away from dense crowds or in cooler areas could help keep our cities from getting too hot.

Technical Summary: Spatially Concentrated Heat Emissions from Data Centers Lead to Localized Urban Warming and Heat Exposure

Problem Statement
The rapid expansion of data centers, driven by the growth of artificial intelligence, has created an urgent need to quantify their environmental impacts. While existing assessments have focused heavily on carbon emissions, water use, and electricity consumption, the local atmospheric consequences of waste heat emissions remain poorly understood. Unlike conventional commercial buildings, data centers exhibit energy use intensities 10–50 times higher and operate continuously, acting as concentrated point sources of anthropogenic heat (AH). Current literature lacks a spatially resolved framework to quantify both direct operational heat and indirect upstream heat from electricity generation, as well as a process-based modeling assessment of how these emissions alter urban thermal environments and population heat exposure.

Methodology
The study develops a spatially resolved framework to quantify heat emissions from data centers and their upstream electricity supply in California, coupling these estimates with a physics-based urban climate model.

Heat Accounting Framework: The authors distinguish between two scopes of heat emissions:
- Scope 1 (Direct): On-site heat emissions from data center operation. The study assumes 100% of utility power is dissipated as sensible heat, representing an upper-bound scenario. Data center locations and capacities were sourced from the Lawrence Berkeley National Laboratory (LBNL) dataset (272 facilities, ~3 GW).
- Scope 2 (Indirect): Heat emissions from off-site power plants generating electricity for data centers. Using a balancing-authority (BA) attribution approach, the study traces electricity demand to supplying power plants, accounting for cross-boundary electricity exchanges. Scope 2 emissions were calculated based on waste heat from thermal power plants (excluding renewables) using EIA-923 and EIA-930 data.
Climate Modeling: The authors employed the Weather Research and Forecasting (WRF) model version 4.6.1 coupled with the Single-Layer Urban Canopy Model (SLUCM) at a 4-km resolution over California.
- Simulation Design: Ensemble simulations were conducted for July 2023. A baseline simulation excluded data center heat emissions, while a perturbation simulation incorporated gridded heat emissions (both Scope 1 and Scope 2) into the urban canopy air energy budget.
- Statistical Rigor: Nine paired simulations with perturbed initialization times were used to generate ensemble means. Statistical significance of warming signals was assessed using a two-sided one-sample Student's t-test ( $p \le 0.01$ ).
Exposure Assessment: High-resolution (100-m) population data were overlaid with simulated temperature changes to quantify population exposure. The analysis included a sensitivity filter for days where baseline air temperature exceeded 23°C, a threshold associated with increased mortality risk.
Future Scenarios: To evaluate potential future impacts, the study simulated threefold and fivefold capacity increases (representing growth by 2030 and 2040) while holding the spatial distribution of facilities constant.

Key Results

Spatial Concentration: Data center heat emissions are highly concentrated in urban clusters, particularly Santa Clara and Los Angeles. Scope 1 emissions account for approximately 59% of total data center-related heat, with 98% occurring within urban areas. In contrast, Scope 2 emissions are more spatially dispersed, with 19% occurring outside California due to cross-boundary electricity flows.
Localized Warming: Heat emissions create distinct warming hotspots. In July 2023, the maximum monthly mean warming reached 3.11°C for land surface skin temperature ( $T_s$ ) and 0.80°C for surface air temperature ( $T_a$ ) in the highest-emission grid cell in downtown Santa Clara. The warming signal is highly localized, diminishing significantly beyond 6–10 km from the source.
Diurnal and Surface Variability: Warming responses exhibit a pronounced diurnal cycle, peaking around 06:00 local time. Nighttime warming is stronger than daytime warming due to a shallower boundary layer that confines heat near the surface. The response in $T_s$ is substantially larger than in $T_a$ , with sensitivities of approximately 0.048 °C per W m⁻² for $T_s$ and 0.012 °C per W m⁻² for $T_a$ .
Population Exposure: Approximately 0.72 million people (1.85% of California's population) live in grid cells where data center heat emissions exceed 1 W m⁻². Exposure is spatially heterogeneous:
- Santa Clara: Nearly 50,000 residents are exposed to an average additional warming of 0.80°C.
- Los Angeles: Approximately 100,000 residents are exposed to an average of 0.35°C.
- When filtering for days with baseline temperatures >23°C, the exposure burden in Los Angeles remains substantial, whereas the Santa Clara exposure decreases significantly, indicating that LA warming overlaps more frequently with already hot conditions.
Future Projections: Under a fivefold capacity increase scenario (representing 2040 growth), the maximum localized warming in Santa Clara could reach 1.98°C, and in Los Angeles 1.22°C. The study notes a diminishing marginal warming effect, where the sensitivity of temperature to heat emission intensity declines as capacity increases.

Significance and Claims
The authors claim this study provides the first integration of facility-level data center information with physics-based urban climate modeling to quantify heat emissions and thermal impacts. Key contributions include:

Scope-Based Heat Accounting: Moving beyond facility-level emissions to include upstream heat from electricity generation, revealing that while Scope 1 dominates the urban heat burden, Scope 2 contributes significantly to the broader regional footprint.
Process-Based Modeling: Unlike observational studies that capture net thermal signatures (including land-use changes), this study isolates the specific meteorological impact of operational heat emissions, distinguishing between surface skin temperature and air temperature.
Urban Planning Implications: The findings identify a "localized thermal footprint" where data center clusters in dense urban areas act as significant anthropogenic heat sources. This challenges the assumption that data centers are primarily rural issues and suggests that siting decisions must account for localized urban warming and population heat exposure.
Mitigation Pathways: The study highlights that cooling technologies partitioning heat into latent heat (evaporative cooling) could attenuate near-surface warming, though this introduces trade-offs with water demand.

The paper concludes that future data center expansion following current siting patterns will exacerbate localized urban warming, necessitating integrated environmental impact assessments that consider thermal footprints alongside carbon and water metrics.