Workshop tackles AI data center power, security challenges

Industry experts from tech companies, equipment makers, utilities and data center developers gathered recently at the Department of Energy's Oak Ridge National Laboratory to brainstorm research and technology pathways to securely integrate AI data centers with the electric grid.

"Data centers, especially when they are training AI, are reshaping the nation's electric load profile faster than any sector in history," said Michael Pesin, deputy assistant secretary for the Grid Systems and Components Division of DOE's Office of Electricity.

His remarks kicked off the Next-Generation Data Centers Power and Security Workshop at ORNL, which focused on supporting American AI leadership while keeping both data centers and electricity affordable, resilient and secure. Topics included data center integration with the grid, power architectures, supply chain challenges, new grid technologies and cybersecurity.

To support that goal, Pesin said existing power plants must remain online and approval processes for new transmission lines must be streamlined. At the same time, AI data centers must limit their impact during extreme weather and manage rapid swings in power demand, which can fluctuate by hundreds of megawatts during AI training.

Pesin introduced a "do no harm" principle, emphasizing that data centers should integrate with the grid without causing instability and be capable of providing support during emergencies. He cited a recent severe cold spell on the East Coast as an example, when a DOE order, issued under Section 202(c) of the Federal Power Act, facilitated access to backup generation from industrial sites and data centers, successfully preventing grid instability and rolling blackouts.

National laboratories play a vital role in translating science into practical solutions, said Robert Wagner, ORNL associate laboratory director for Energy Science and Technology. ORNL recently announced a new Next-Generation Data Center Institute consolidating its expertise in energy technologies, high-performance computing, cybersecurity and grid science.

The workshop identified priority research areas including direct-current power architectures to reduce energy loss, flexible and networked microgrids, real-time energy demand modeling and feedback, U.S. manufacturing workforce development, secure data center design and security testing for grid equipment.

Grid interconnection and data center flexibility

Utilities and tech companies discussed the difficulty of forecasting future data center energy needs. "Due to their rapid development, a challenge is that we don't fully understand the long-term power requirements for each facility," said Ray Knotts, a senior director at Tennessee Valley Authority. "We do not want to hinder data centers, risk reliability or increase our system costs based upon volatile engineering estimates."

Grid and data center experts delved further into how data centers could support the grid by reducing demand during periods of peak stress on the system.

"Faster grid connection is the incentive for the hyperscaler to consider flexibility," said Anish Gaikwad, deputy director of the Electric Power Research Institute, which facilitates a Data Center Flex program. "We don't need flexibility every hour of the year - only a few hours during certain portions of the year may suffice." If data centers reduce demand at those times, interconnection waits can be shortened, since utilities might not need to add as much capacity.

During her keynote address, Sparrow Mahoney, CEO of STAK Energy, advocated for data centers that don't rely on the grid at all. Her company plans a large data center platform in Alaska powered by natural gas from the adjacent Prudhoe Bay. "The AI data center load shape is changing and data architectures have advanced, so places like Alaska's North Slope become not only viable, but we argue that they have become inevitable," Mahoney said.

Supply chain and workforce challenges

Wherever they are built, data centers rely on tight supply chains for gas turbines, switchgears, circuit boards and transformers. Many shortages stem from foreign-sourced materials such as electrical steel for transformers, copper for conductors, semiconductors for chips, graphite for grid batteries and gallium for fiber optics.

Some manufacturers said more recycling of rare materials from retired equipment could help. Workshop panelist Ross Berntson, president and CEO of Indium Corporation, advocated pursuing new domestic and North American mining, refining and capturing trace metals. Additionally, new technologies are emerging that reduce the metal consumption and improve system performance and efficiency, such as solid-state transformers that use less copper.

However, new technologies may create new supply chain bottlenecks. Workshop panelist Evila Melgoza, data center product and technology leader for Schneider Electric, welcomed innovations like solid state transformers and direct current architectures but expressed concern about their supply chains. Schneider is proactively scaling and expanding its manufacturing capacity as surging demand meets unprecedented data center growth.

Melgoza emphasized the importance of continued investment in workforce development for manufacturing readiness. She said that building a strong pipeline of technicians, power distribution electricians, and power electronics specialists will be essential to supporting the industry's growth.

Workshop participants said national labs can help smooth AI data center integration by developing validated load models for data center energy use and by creating a "resilience hub" to develop and test next-generation data center technologies.

UT-Battelle manages ORNL for the Department of Energy's Office of Science, the single largest supporter of basic research in the physical sciences in the United States. The Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit energy.gov/science. - S. Heather Duncan