Ex-Google CEO details massive AI energy needs at House hearing, advocates for fusion and SMR R&D

The AI race hinges not just on algorithms and silicon, but increasingly on raw power. The International Energy Agency (IEA) projects that global electricity consumption by data centers will more than double by 2030, with AI-specific data centers alone expected to quadruple in consumption. By then, the energy used for AI processing in the U.S. could surpass that of traditional energy-intensive industries like steel and cement manufacturing, as The Guardian has reported.

David Turk, former Deputy Secretary of Energy, cited Lawrence Berkeley National Lab figures during a House Committee on Energy and Commerce hearing, projecting data centers could consume up to 12% of all US electricity by just 2028. That is a significant jump from 4.4% in 2023.

Former Google CEO Eric Schmidt shared his thoughts on these themes at the hearing. “People are planning 10 gigawatt data centers now,” he said. “An average nuclear power plant in the United States is one gigawatt,” he told lawmakers. He implicitly questioned the nation’s capacity to meet such demand rapidly given current build rates.

‘We need the energy in all forms’

Schmidt’s testimony underscored a looming bottleneck for U.S. technological leadership. Schmidt’s subsequent call for an “all energy sources” approach highlights not only a significant infrastructure challenge but is a sort of call to action — namely, to develop new approaches for power generation, grid management, and energy efficiency. “What we need from you… is we need the energy in all forms, renewable, non-renewable, whatever. It needs to be there, and it needs to be there quickly,” he stressed.

He specifically mentioned fusion and small modular reactors (SMRs), a type of compact nuclear fission reactors. While endorsing SMRs as “the right answer” technologically, Schmidt expressed exasperation with the roadblocks preventing their contribution: “One of my personal frustrations is the regulatory structure around nuclear and SMRs… How many SMRs are in use in America today? Zero,” he stated. He continued: even a “fast approval time is considered to be 12 years. That defies logic.”

This regulatory friction persists even as the industry seeks closer ties to power sources; multiple representatives noted a trend of new data center projects co-locating directly with nuclear or natural gas generation facilities in states like Pennsylvania.

Schmidt calls to tackle delays to modernize grid

Schmidt also highlighted systemic delays in connecting power to the grid: “It takes on average 18 years to get the power transmissions and so forth… We need to find federal ways to preempt that and make it happen faster.”

“The interconnection queues are a very good example… the delays are crazy,” Schmidt noted. Indeed, national lab data confirms hundreds of gigawatts of proposed generation projects, primarily renewables and storage, are currently stalled for years awaiting study and approval in these queues. “People they have the money, they have the ability to get the power built and they can’t interconnect it. That’s a good example of grid modernization [needed].”

Schmidt warned of the potential competition with China in the AI space, and its ramifications from a national security perspective. “If [China] comes to super intelligence… first, it changes the balance of power globally in ways that we have no way of understanding, predicting, or dealing with,” he warned, linking the success of the U.S. AI effort directly to resolving these fundamental energy infrastructure challenges.

Committee Chair Brett Guthrie referred to Schmidt’s presentation/assessment as “sober.”