Energy consumption for artificial intelligence technologies is a “critical” emerging mission area for the Department of Energy as the agency looks to advance the benefits of AI and machine learning systems while minimizing the technologies’ carbon footprint.
Speaking at an Axios forum on Tuesday, Helena Fu, the director of the Office of Critical and Emerging Technologies at Energy, said that her agency sees automated systems’ burgeoning demand on the power grid as a way to usher in more clean firm power — clean energy solutions that are not dependent on weather as solar and wind are — with help from industry partners.
“This rise in energy demand is something that has been planned for, is needed for the energy goals that we need to be achieving,” Fu said. “It’s not only coming from the data centers, it’s also coming from electrification, it’s coming from the fact that manufacturing is coming back to the United States. So we think that this is really a critical golden moment and opportunity to pair the demand from data centers for training AI with many of these companies’ commitments to clean firm power to really supercharge the deployment of clean energy.”
The culprit behind AI technologies’ demand on the U.S. energy grid is data. With a larger volume of data needed to both train large AI and ML models, data centers processing these operations will require more energy, some of which is generated through fossil fuel consumption. Increasingly large data center needs will result in greater electricity demands.
“AI is different in the sense that the loads are much larger oftentimes, and the expectations of companies as well is much faster,” she said.
Fu said that through various internal initiatives — namely the Frontiers in Artificial Intelligence for Science, Security and Technology program — her agency is looking to address the energy challenges posed by AI-enabled computing. She noted that, apart from AI, other digital services coming online and more manufacturing work based in the U.S. are also two leading burdens on U.S. power sources.
Fu also said that Energy Secretary Jennifer Granholm directed an advisory board to investigate future tech-driven demands on the energy grid in a public briefing slated for next week.
“The department is laser focused on meeting this moment on energy availability, not just for AI but for all of the growing electrification needs,” she said. “What we think is actually needed in this moment is really a focused effort to bring the various parties to the table here, and make sure that people understand what the needs are.”
One potential solution comes from Energy’s longstanding Exascale Computing Project. As the official U.S. network of high-performance computers, Exascale has previously driven innovation in energy-efficient computing efforts, according to Fu. In conjunction with the successful public-private sector partnerships that were spurred by Exascale, Fu anticipates the project to develop a new generation of “highly efficient” AI computation systems.
“I think the Exascale Computing Project is a really good example of how we were able to work on [energy efficiency goals] over time, in deep partnership with a number of companies to really enable this new era of Exascale,” she said.
Despite the anticipated energy needs, AI still brings significant benefits to Energy’s research objectives. Fu echoed staff from Los Alamos National Laboratory in citing the myriad benefits large frontier models can have to “supercharge scientific discovery” across a bevy of subjects.
“If we have the data and the compute, we really want to be able to develop models that deeply understand science that deeply understand math and physics and chemistry, and we have those experts at our national labs who do understand all of these different disciplines,” she said. “We think that in partnership with the industry and academia, we’ll be able to develop extremely powerful models that will be applied to our broad missions.”
