America’s largest Data Center Uses More Electricity Than Houston

A complex network of cables in a data center with a monitor in the foreground.
Photo by panumas nikhomkhai on Pexels

One of the most hotly debated issues in AI is the amount of electricity its data centers consume, and the extent to which this drives up rates for residents. It could also accelerate the use of fossil fuels. Both concerns need to be weighed against how fast AI revenue is actually growing and forecast to grow, as well as against the availability of capital to fund construction.

There are more data centers in the US than most Americans realize. Today, according to Pew, the figure exceeds 3,000, with another 1,500 under construction. One complication is that various research organizations use different methods to measure this, so figures vary by source. The Information reports that more than 500 data center projects have been delayed, primarily because local residents don’t want them nearby.

One thing is certain: barring a national law blocking construction, the number of US data centers will increase by the thousands before the end of the decade. The World Economic Forum puts international data center investment at $7 trillion. This is expected to happen because the largest AI companies have nearly endless access to capital — largely because they include some of the world’s most profitable companies, led by Microsoft, Alphabet, Amazon, and Meta. Private companies with access to capital, along with direct investments from Wall Street, add to the total.

So what actually gates the pace of data center construction? First and foremost, the pace of AI revenue growth. The Economist puts current annual AI revenue at about $200 billion. Anthropic, a leading AI company, has an annual revenue run rate reported at $100 billion, though the company has not confirmed that figure. Revenue will need to grow many times over to justify the scale of data center investment underway.

A hand giving thumbs up next to profit chart on a whiteboard, indicating success.
Photo by Nataliya Vaitkevich on Pexels

Revenue will be the single biggest factor shaping the industry’s future — and AI use has already become extremely expensive for some companies. According to Yahoo, an early warning sign came from Uber: its CTO revealed in April that the company had burned through its entire annual AI budget in just four months. Impressive headlines about AI capacity growth raise the question of whether that growth might eventually drive away customers facing costs like these.

Many companies have sidestepped much of this expense by using open-source models, some of which come from China. These models may not be as advanced as those from companies like Anthropic, but cutting-edge technology often isn’t necessary for routine tasks. TechRadar recently reported that Chinese AI models can cost up to 90% less to run than leading US alternatives — a gap already shaping where developers choose to run their AI workloads. Even so, these cheaper models still require data center capacity.

Electricity availability is as much a concern as its cost. Upgrading the US grid and building new power sources — whether fossil fuel or renewable — could take years. And because the grid is owned and operated by many different companies, how the system functions end-to-end is itself a challenge. As the Federal Reserve Bank of Dallas wrote, coordination problems, not technological limits, are the biggest bottleneck linking AI and energy.

Even before these issues are resolved, the AI data center build-out in the US is already massive. How massive? ClimateCrisis247 analyzed data from several sources to compare data center electricity use with the power needs of entire cities — and the figures are staggering. The SpaceX Colossus 2 center near Memphis draws 946 megawatts, enough to power roughly 750,000 homes simultaneously, making it the largest known AI data center.

The table below lists current data centers, their power output in megawatts, and cities with comparable household electricity use, based on data from the U.S. Census Bureau’s American Community Survey (ACS) and Epoch AI. Because the equivalents aren’t a perfect match, editors selected the closest pairing in each case, so some cities appear more than once.

Data CenterCity EquivalentIT Power (MW)
Colossus 2 (Tennessee)Houston946
New Carlisle (Indiana)Houston910
Fairwater (Atlanta)Philadelphia636
PrometheusPhoenix631
Stargate (Abilene)Austin421

More from ClimateCrisis 247

Similar Posts