Energy Department's advisers call for agency's own AI data center testbed
Add Axios as your preferred source to
see more of our stories on Google.
/2024/08/25/1724607390467.gif?w=3840)
Illustration: Brendan Lynch/Axios
The Department of Energy's advisory board proposed creating the agency's own AI data center to study ways to lessen energy usage, according to a recently released report.
Why it matters: The department is wrestling with the rapid growth of energy requirements to power AI data centers, and the corresponding threat to climate goals if the new electricity production is met by fossil fuels.
The intrigue: Among many other recommendations, the advisory board calls on DOE to create a "data-center-scale AI testbed" within the agency.
- It would allow scientists from DOE, academia and industry to help make such data centers more energy-efficient and flexible in how they tax the grid.
- This testbed would be distinct from the agency's existing computing facilities.
Zoom in: Presented to Energy Secretary Jennifer Granholm on July 30, the report focused on so-called hyperscale facilities requiring 300 to 1,000 megawatts of electricity or greater, many of which are being built with lead times of just 1 to 3 years.
Zoom out: In preparing the report, the board consulted with Big Tech, power suppliers and many others and came away with the conclusion that there is an urgent need for deploying more "flexible, firm electricity supply."
Between the lines: There have been many numbers thrown around to measure the near- and long-term increase in energy demand associated with data centers, but the board notes there are plenty of reasons to be skeptical.
- "Individual electricity providers are sometimes getting multiple requests for the same data center and developers are also exploring multiple sites around the country (and world) for the same facility," Tom Wilson, a principal technical executive for the nonprofit Electric Power Research Institute (EPRI) and an advisory board report coauthor, told Axios via email.
- Future breakthroughs in energy efficiency and the efficiency of AI functions themselves are also being overlooked.
- In addition, proprietary concerns limit the government's visibility into some of the energy demand-related factors.
