Tech leaders could pull advanced nuclear over the finish line, NERC CEO says
Data centers are driving up electricity needs, and leaders in the power sector are wondering how to meet the demand as well as ambitious federal decarbonization goals.
The head of the North American Electric Reliability Corporation said the answer to this complex reliability problem might come from the very data centers whose AI needs require new power capacity on the grid.
“If the data center issues that we’ve talked about are the catalyst for really advancing our understanding of nuclear, I think that’s a great contribution that they’re making beyond all the lifestyle and societal benefits” associated with AI, Jim Robb, NERC’s president and CEO, said at a virtual press meeting hosted by the United States Energy Association on Wednesday. He points to recent investments in nuclear and small modular reactors.
Last month, Microsoft signed the first nuclear fusion power purchase agreement with Helion Energy, and tech leaders are continuing to get involved with advanced nuclear technology: Bill Gates is currently the chairman of the board of Terrapower, which is developing small modular reactor technology, and Jeff Bezos is backing a fusion plant to be built in the UK by General Fusion. Industry experts have noted an interest in advanced nuclear solutions from high energy consumers.
“Getting, you know, guys like Gates and Bezos … that have gobs of money to really push that technology and get that into a commercializable form, I think will serve this country incredibly well,” Robb said.
However, whether or not those investments push advanced nuclear to market, NERC’s long-term reliability assessments show material growth. According to NERC, data center loads could be as much as the load of a city like San Francisco, and “exacerbates all the reliability challenges around resource adequacy,” Robb said.
The alternative will be for the power sector to adjust to the high demand for flexible baseload power with natural gas or coal, experts said. Other estimates suggest that even more power capacity will be needed on the grid.
In a recent survey of utilities, the Electric Power Research Institute (EPRI) found that data center customers are making service requests “on the order of 4 to 5 GW” for co-located facilities, said Daniel Brooks, Corporate VP of Integrated Grid and Energy Systems at EPRI.
“That’s a significant requirement in terms of additional supply and delivery capacity that has to be planned, sited, permitted and constructed,” Brooks said at the same virtual press conference.
The heavy demand from the tech sector for more power has led to requests for extending the life of existing coal plants and to add more natural gas. Last month, Goldman Sachs predicted global gas markets to increase 50% by 2029.
“There’s no way to stabilize the grid without the use of some firm baseload power,” Dan Brouillette, president and CEO of the Edison Electric Institute (EEI), told reporters.
The Biden administration aims to decarbonize the grid in the next decade, but Brouillette says that natural gas continues to have a role in the power mix given the high demand from data centers.
“As I see it here, fully decarbonizing the grid, electricity production in the United States is probably not going to happen by 2035,” he said. “That does not suggest for a second that we won’t make every effort to reduce emissions and to continue the path toward decarbonization.”
But the power sector will face difficulties meeting the demand even if it embraces new gas resources, as data centers can come online much faster than new flexible power capacity, Brooks said.
“Coordination between the data center community and stakeholders and the utility community is going to be key to make sure the grid can expand to reliably and affordably meet that [demand] over the timelines that are needed,” Brooks said.
Despite the present concern for AI-driven load growth, also highlighted in NERC’s summer reliability assessment, Robb shared his personal expectation that the demand will lessen in the sector.
“It’s probably not as dramatic as we think right now just because we’ll learn as time moves on,” he said.
AI chips being used in data centers can become more energy efficient. Earlier this week, Nvidia Corp. CEO Jensen Huang announced the next generation of chips, slated for 2026, that will feature lower energy consumption.
“Algorithms will also get better over time,” Robb said, noting the power sector faced similar concerns with the growing presence of the internet decades ago.
