The conversation around the race to overcome the energy bottleneck of artificial intelligence generally centers around top tech companies, major utilities, and the ecosystem of startups that has sprung into existence in the last few years to work around the challenge.
Not to be left behind, the Department of Energy today released a brief on data center electricity demand, which outlined its own role in the rapidly evolving and somewhat chaotic rush to track AI’s electricity use — and make sure the U.S. has enough electricity.
The report outlines a portfolio of “readily scalable and cost-competitive” technologies (including those outlined in the agency’s Pathways to Commercial Liftoff series) that should be deployed to meet rising energy demand. Solar PV, land-based wind, battery storage, and energy efficiency solutions like virtual power plants are ripe for immediate investment, the report said.
Tracking energy use from data centers as well as energy efficiency is one area where DOE expects to play a unique role, said Carla Frisch, acting executive director for DOE’s Office of Policy.
“We want to strike the balance of making sure we’re collecting the data that’s really necessary and then also limiting the reporting burden,” Frisch said.
Striking that balance involves several approaches. The agency is currently in the midst of updating a report on data center energy use based on research from Lawrence Berkeley National Lab that takes into account chip shipments. That’s key, Frisch said, “to make sure we’re very accurate and understanding the coming energy demands.”
But the agency is also looking beyond estimates and forecasts, and toward potentially shifting disclosure standards and requirements.
DOE is collaborating with the Energy Information Administration “to work through, for data centers and crypto currency more broadly, what are all the ways that we can appropriately collect information, voluntarily and with mandatory approaches over time.”
That could eventually include some type of efficiency metric, which is also something the national labs have been working on for years, Frisch said.
But at the moment, given the lack of requirements for companies to disclose that type of information, the focus is primarily on sharing best practices.
“What’s interesting about this industry is, they’re all in high competition right now, so they’re not necessarily sharing information with each other,” Frisch said. “That’s one of the roles that we can play here at the federal government….To the extent that they are willing to share information with us, we can share anonymized information back out.”
DOE has already seen interest in this approach from data center developers, Frisch said, and the agency has already shared some data through the Better Buildings Better Plants program, which is focused on sharing industrial energy efficiency strategies.
“There’s not a lack of focus on energy efficiency, because it’s a really big economic opportunity for these facilities,” she added. “Each unit of efficiency is really helping them to not think about power from some other source.”
This comes several months after DOE released a report outlining the potential, and pitfalls, of AI for the grid, as well as four key areas to prioritize for development: grid planning, permitting, operations, and resilience.
