AI Data Centers Turn to Flexibility to Ease Energy Strains

The rapid growth of AI data centers is causing concerns about their energy consumption. A recent experiment in Texas showed that it's possible to reduce power usage by as much as 75% for a short period without affecting performance.

AI data centers require enormous amounts of electricity, often drawing more power than a small city. To mitigate this issue, researchers are exploring ways to control and curtail the power use of AI chips.

A joint experiment between Luxor Energy, a company with roots in Bitcoin mining, and Bentaus, which makes software that controls how much power computer chips use, successfully reduced the power draw of an Nvidia B200 chip by 75% within half a second without affecting performance. The test was a success but involved only a single chip.

Large data centers contain tens of thousands of chips, and scaling this solution to meet their needs is a significant challenge. However, the experiment demonstrates the potential for AI data centers to sort work by urgency and occasionally ask the grid for less power, reducing strain on the electricity supply.