The Hotter AI Gets, the More Power Data Centers Consume

Data centers are the engines of the internet and the physical backbone of today's AI race. Their expansion is outpacing what power grids can handle, and resistance to data centers is already appearing worldwide.

According to the BBC, the number of data centers worldwide has grown from about 8,000 in 2021 to 12,250. The U.S. ranks first with 4,767, and Virginia has the highest density, home to the famous data center corridor. In terms of installed capacity, the U.S. has 53.73 GW, China 31.90 GW, the EU 11.92 GW, Japan and South Korea 6.58 GW, India 3.55 GW, and the UK 2.59 GW. For reference, the Three Gorges Dam has a total installed capacity of about 22.5 GW—U.S. data center capacity is more than twice that.

Some mega-projects offer a more intuitive sense of scale. The world's largest operational data center is China Telecom's facility in Hohhot, covering about 1 million square meters—one square kilometer—roughly the size of a small city. SpaceX's Colossus project covers 70,000 square meters; the Stargate project, involving OpenAI, covers 3.5 square kilometers; and Meta's planned Hyperion project spans 11 square kilometers, equivalent to 11 times the size of the Hohhot facility.

As for electricity consumption, industry estimates suggest that data centers worldwide now use more electricity than all of France. AI workloads are extremely compute-intensive and place stringent demands on the power supply. Today, behind every AI conversation are rows upon rows of server racks consuming electricity.

AI's bottleneck may not be just chip shortages; electricity is becoming an even harder constraint. Data centers are not going away, but whether the power grid can keep pace with this growth is one of the most important questions to watch in the coming years.

Data from Data Center Map, as cited in a BBC report: Original link.