Billions to Trillions: Can the grid keep up with AI without derailing climate progress?

A surge in AI-driven electricity demand is colliding with the United States’ push to decarbonize in a moment that requires both urgency and discipline, according to speakers at Duke’s third annual “From Billions to Trillions” summit. In two sessions focused on AI and power systems, Duke faculty members helped frame a central question: How can we scale infrastructure fast enough for AI while protecting affordability, reliability, and climate gains?

IRA momentum meets an unexpected surge

In a fireside chat about identifying opportunities for digital infrastructure investment to catalyze clean energy solutions, Aaron “Ronnie” Chatterji, Chief Economist at OpenAI and the Mark Burgess & Lisa Benson-Burgess Distinguished Professor of Business and Public Policy at Duke’s Fuqua School of Business and Sanford School of Public Policy, spoke with Kristina Costa, former deputy assistant to the president and director of the White House Office of Clean Energy Innovation and Implementation.

Image

Costa, now managing director for policy and strategic engagement at the University of Chicago’s Institute for Climate and Sustainable Growth, reflected on the passage of the Inflation Reduction Act in August 2022, noting how much has shifted since then. At the time, she said, debates centered on equity, climate resilience, and competitiveness in a rapidly changing global economy—themes that remain today. But the energy implications of generative AI were not yet clear.

Looking back, she described clean energy investment driven by the law as “a massive shift in a relatively short period of time in response to policy signals from the federal government.” She pointed to roughly $400 billion in announced investments and record-setting clean energy deployment in 2023 and 2024. “We saw that there are policy tools that can work extremely effectively,” she said.

Image

Yet she also acknowledged that deployment outpaced integration. Utilities and grid operators, she said, are still “upgrading the lines, the superstations, the grid management” and working to integrate new power onto the grid. At the same time, projections show data center demand rising sharply. “We’re in a moment right now,” Costa said, describing a system under pressure from load growth that was not fully anticipated when the IRA was drafted.

She cautioned policymakers to avoid overreacting to speculative projections and to weigh promises carefully. Government officials, she said, must balance companies “promising you the sun, moon, and stars” with impartial data and long-term system needs. She also stressed that while AI tools can accelerate analysis, they must remain “expertise-enhancing,” with engineers and regulators validating outputs to protect reliability.

“Massive blow-through” and system-wide strain

Image

The focus shifted from federal policy to grid execution during a panel on scaling energy for AI demand, moderated by Jackson Ewing, director of Energy and Climate Policy at Duke’s Nicholas Institute for Energy, Environment & Sustainability.

Ewing opened by describing “massive blow-through with some ambiguity” in projected demand, while emphasizing “no doubt that this is going to be a significant increase” for the power system. The challenge, he said, will require changes in “the way that we generally transmit and distribute power.”

Rebecca Kujawa, founder and CEO of Zerra Partners and former president and CEO of NextEra Energy Resources, agreed that capital is available but pointed to structural constraints. In the past, she noted, 20 megawatts was the largest load utilities expected from an industrial customer. Today, proposals involve “a gigawatt, two gigawatt, five gigawatt loads,” which can double the size of a utility system and force new conversations about risk and cost allocation.

Image

Don Moul, president and CEO of the Tennessee Valley Authority, described steady growth across TVA’s seven-state region—from more than just data centers—and emphasized the utility’s statutory obligation to least-cost planning. TVA relies on an integrated resource plan and a more granular power supply plan, he said, evaluating cost, regulation, and portfolio balance. At the same time, he acknowledged risks beyond fuel volatility, including the risk of “overbuilding infrastructure” and leaving “everyday ratepayers holding the bag.”

Allison Clements, president of 804 Advisory and former Federal Energy Regulatory Commission commissioner, urged the panel to consider how much unused capacity already exists. “I like to say I’m a carpool mom,” she said. “Four times a year I have four kids. I don’t go out and buy a new car.” The grid, she argued, has “so much latent opportunity” if regulators and utilities create incentives that reward performance, flexibility, and smarter use of existing assets.

Speakers repeatedly returned to flexibility as a near-term bridge. Curtailable load and demand response can help manage peaks, but they require system investment and coordination. As Ewing and others made clear, AI’s energy demands will test more than generation capacity. They will test whether market rules, regulatory processes, and planning frameworks can move quickly enough to align economic ambition with climate responsibility.