Lifetime Citizen Portal Access — AI Briefings, Alerts & Unlimited Follows
Argonne researchers say AI can make the electric grid smarter — but warn data centers will strain power, water and costs
Loading...
Summary
At Argonne’s Out Loud lecture, laboratory researchers and utility representatives said artificial intelligence can improve grid operations—speeding simulations, forecasting demand and spotting equipment problems—but cautioned that rising data‑center demand will require major generation, storage and water planning and coordinated cost allocation.
Argonne National Laboratory hosted an Out Loud lecture and panel where researchers and utility officials described how artificial intelligence can help operate a more reliable electric grid while flagging infrastructure, water and cost challenges from large data centers.
Paul Kearns, Argonne’s laboratory director, opened the event by linking the two technologies: "The grid energizes AI, and AI empowers the grid," he said, noting Argonne’s computing resources, including the Aurora supercomputer, that support research into AI‑driven grid tools.
Dr. Henry Huang, director of Argonne’s energy systems and infrastructure assessment division, said faster, higher‑frequency measurements and far larger datasets make AI essential for real‑time sensing, decision‑making and action on the grid. He described AI applications already in development at Argonne — equipment health monitoring, short‑term demand forecasting and edge cybersecurity — and cited research results that, in pilot work, AI reduced certain equipment failures by roughly 60% and lowered related costs by about 40%.
"AI holds the promise to understand grid capacity, process bigger grid data, and eventually help the grid sense, decide and act in real time," Huang said, while also warning that training models and running large data centers consume substantial electricity and cooling resources.
Computational mathematician Ki Baek Kim outlined a common workflow used in the research — combine, predict, recommend, explain — where AI ingests weather, sensor and mapping data, runs fast simulations and delivers ranked, explainable recommendations to operators. Kim showed examples in which AI accelerated simulations by orders of magnitude, allowing operators to assess thousands of "what‑if" scenarios in minutes rather than hours.
An Exelon/ComEd transmission planning official, identified in the program as Hai Bin (transcript spelling varies), said the industry is preparing for much larger data‑center loads. He said a typical modern data center can be on the order of 50 megawatts and described proposals and discussions for much larger facilities — in some cases hundreds of megawatts — that would require substantial new generation and network upgrades. He cited research by federal and industry bodies calling for large investments in generation and grid infrastructure to preserve reliability.
Panelists and audience members raised three recurring concerns: who pays for the upgrades, how to protect local water and land resources, and how to govern AI safely.
On costs, panelists said impacts on consumer bills are complex and depend on market design and local planning. The speakers described existing utility programs to help customers and said utilities, regulators and AI developers need to work out how to allocate the incremental cost of generation and grid upgrades. Panelists cited regional projections in which AI‑related loads could account for more than 5% of electricity consumption in some states and a much larger share in areas with heavy data‑center development.
On water, Argonne researchers said cooling needs and site selection are critical. Huang described a geospatial energy mapper (GEM) that overlays power networks, water resources, land use and geologic constraints to identify suitable locations and avoid undue impacts on agriculture or local water supplies.
On governance and trust, the panel emphasized domain‑specific models, careful data governance and operator oversight. "We want to trust AI, but it matters how the AI model is developed and what inputs it uses," Huang said; panelists urged guardrails, privacy protections and human‑in‑the‑loop designs.
The event closed with a reminder that the talk was recorded and that Argonne and its partners continue to study practical solutions — including battery storage, demand‑management techniques and reuse of waste heat — to reduce the operational strain of AI infrastructure on the grid.
The panel did not propose any regulatory changes at the meeting; speakers described ongoing engineering studies, partnerships and utility programs as the next steps.
