UNIVERSITY PARK, Pa. — Penn State’s School of Electrical Engineering and Computer Science and Institute of Energy and the Environment recently hosted the AI Data Center Integration to Power Grids Workshop at the Engineering Collaborative Research and Education (ECoRE) Building on the University Park campus.
The all-day workshop on May 22 brought together researchers, regulators and utility companies to discuss challenges presented by the energy consumption associated with artificial intelligence (AI) data centers, specialized facilities that house essential infrastructures for developing and deploying AI and machine learning models. The event featured 11 speakers, two industry keynotes, three technical sessions, a panel session and three parallel breakout sessions.
Among the event coordinators were Nilanjan Ray Chaudhuri, professor of electrical engineering; James Freihaut, professor of architectural engineering; and Seth Blumsack, professor of energy policy and economics and of international affairs in the John and Willie Leone Family Department of Energy and Mineral Engineering. Chaudhuri, Freihaut and Blumsack each co-led various sessions during the workshop.
Chaudhuri said the event emphasized challenges in key areas such as grid integration, architecture design and legislation and policy. Topics discussed included the sharp fluctuations in power consumption between resting and computing periods, the trend toward higher direct current voltages to distribute large amounts of electrical currents and the need to co-site data centers with certain types of facilities — such as those that rely on industrial heat — to reduce operating costs and maximize co-benefits.
“We asked the panelists how Penn State could help,” Chaudhuri said. “The consensus was that there’s a great need for power engineers, which is something of a bottleneck when it comes to managing interactions with the bulk power grid. This need could be addressed through strategic investment into academic research.”
According to Freihaut, the workshop also covered issues affecting the areas around AI data centers, including increases in energy costs as well as the potential of data centers to use large amounts of community water and land space. Also important to consider, he said, are the built-in sensing devices that disconnect a data center from the central grid and connect it to local power systems as needed to maintain power quality. Multiple data centers undergoing this switch at the same time could cause a regional blackout.
“The power demand associated with AI changes on sub-second, second and hourly bases, unlike the more gradual shifts associated with most industrial operations,” Freihaut said. “The central grid is having difficulty providing enough power to these data centers, which is putting serious strain on local and regional transmission organizations. These types of load demand-supply interactions are completely new engineering challenges to the electric grid system.”
According to Chaudhuri, the workshop forged and strengthened connections between Penn State experts and external partners, which are expected to support interdisciplinary collaborations across the University and industry.
“Our ongoing research in AI was lauded as highly relevant,” Chaudhuri said. “It was mentioned at the workshop that Penn State's principles-based approach and fundamental research are invaluable to the stakeholders involved. If we continue that work, we could significantly de-risk the challenges associated with the rapid expansion of data centers.”
For the full agenda and list of speakers, visit the workshop’s event page.