The nation's power transmission system is the main barrier to meeting the explosive electricity demand driven by artificial intelligence (AI) and data centers, said Federal Energy Regulatory Commission (FERC) commissioner Judy Chang.
"If we had a grid that's sufficiently reliable and has sufficient capacity, then we could integrate the amount of generation that the large loads are asking for," said Chang on a panel at CERAWeek by S&P Global in Houston, Texas, on Monday. "It's because we don't have adequate transmission that we're encountering these delays of having the generation meet the demand."
AI's power needs arrived with a speed and scale that have blindsided even veteran energy regulators, forcing a historically slow-moving industry to accelerate timelines in unprecedented ways. Hyperscale data centers now routinely plan 1,000MW per site and come on line far faster than utilities have ever had to accommodate, said Chang.
"Even just two years ago, we did not anticipate the pace and the size of the growth," said Chang. "The pace of change has been the surprising factor."
The rapid scramble has forced utilities, developers, and data-center operators to immediately shift into "problem-solving mode," debating who pays for new infrastructure, how to structure agreements, and how to rebalance supply and demand on a grid that was never built for gigawatt-scale loads arriving all at once.
With communities pushing back against data center development and higher electricity prices, developers and officials are increasingly settling on a framework in which the technology sector foots the bill for the necessary overhauls. Developers are increasingly pairing data-center growth with their own behind-the-meter generation, investing directly in new energy projects or situating facilities in regions with stranded or underutilized supply, said Chase Lochmiller, chief executive of data center builder Crusoe.
"This is the most well-capitalized infrastructure opportunity ever and much of that capital is going into building the energy systems needed to power AI," said Crusoe.
Bigger and smarter
Aside from infrastructure, regulators also must design market signals to incentivize large-load users to develop more flexibility that allows them to respond quickly to the grid's power needs, said Chang.
Large data centers rarely operate at their full peak load, meaning they could reduce consumption or switch to backup resources when the grid is stressed. But because they aren't paid for that flexibility and in many markets aren't allowed to offer it the way power plants can, most never invest in the equipment needed to respond.
But Chang emphasized that fixing the demand-side incentives is not enough. The grid also needs to be upgraded so operators can see and manage power flows in real time. As a result, a potential tool for managing the strain of AI-driven growth remains largely unused, even as prices rise and the system struggles to keep up, she said.
Tackling demand-side incentives will require a smarter grid that deploys the computing benefits offered by AI, said Chang.
The US is "overdue" in deploying real-time sensors and modern monitoring systems that would let operators understand line conditions minute-by-minute rather than relying on preset limits. With those upgrades in place, she said, AI could help optimize how power flows across the system and unlock capacity from both existing and new infrastructure.
