Editor’s note: This piece was updated on October 16, to include the news of Amazon’s investments in SMRs. It was updated further on October 17 to include comment from Google.
In late September, Microsoft announced it would finance an effort to restart a shuttered nuclear reactor at Three Mile Island in Pennsylvania, to the tune of $16 billion. The deal made the OpenAI investor the first tech company to tap large-scale nuclear specifically for its data center energy needs.
But while that deal may have signaled that hyperscalers are eyeing nuclear, Microsoft’s peers aren’t following in its exact footsteps. This was made clear by announcements from both Google and Amazon this week that they will be investing in small modular reactors.
Google is not only targeting a different technology, though — it’s also taking an “orderbook” approach to procuring power, purchasing from multiple reactors that will be developed by Kairos Power.
Amazon, meanwhile, is splitting the difference. After signing a deal earlier this year to colocate a data center next to an existing reactor in Pennsylvania, the company said today it has made three new agreements, targeting SMRs. These latest deals include funding the feasibility phase of an SMR project in Washington state with Energy Northwest, and a partnership with Dominion Energy to “explore the development” of an SMR project near the existing North Anna nuclear plant.
On its face, the SMR approach may appear inherently riskier than Microsoft’s Three Mile Island revival. After all, SMRs haven’t been commercialized at the scale of large reactors, and the SMRs central to the deal haven’t yet been built; the first won’t be online until 2030.
However, Google is betting that the SMR approach may lead to faster construction and more predictable timelines, as well as more flexible deployment compared to traditional reactors. Amazon, in its own announcement, also pointed to the small physical footprint of SMRs, which means they can be built closer to the grid.
According to the Department of Energy, SMRs may have an easier time reaching “Nth-of-a-kind” stage — meaning they become de-risked enough to be considered “bankable” by infrastructure investors and can access low-cost capital. Large reactors, the agency said, are “difficult to construct in a manner that reaches NOAK cost given megaproject issues and a proliferation of different designs.” And, according to the agency’s recent report, it takes more capacity to bring large reactors down the learning curve.
Distinct approaches
While the approaches by all three tech companies to procuring nuclear power send good signals to the nuclear industry, Nuclear Innovation Alliance executive director Judi Greenwald said each approach targets very different end goals — and will have different implications.
Microsoft’s decision to fund the reopening of a single, existing, large-scale reactor, is likely to be relatively low cost, and will come online sooner than an SMR will, Greenwald said: “The Microsoft deal is going to get more near-term clean power, probably with less cost uncertainty.”
The disadvantage with that approach, she added, is that there is that there aren’t that many reactors available to be restarted. While we know what those existing reactors cost and a lot of the requisite machinery already exists, that means that Microsoft’s strategy simply isn’t targeting NOAK.
No matter how cost effective and streamlined the project ends up being, “there just aren’t that many places to replicate it,” Greenwald said.
That’s compared to Google’s agreement with Kairos, which Greenwald describes as more of a “collaboration” between the two to share the risk of the first few reactors. “Google is sharing some of the risks of that development, but will also reap the benefits, because they’ll get the later set of these reactors, once they’ve brought their costs down,” she said.
Google, for its part, declined to disclose any financial details of the deal, including whether the company has signed an official power purchase agreement or something more along the lines of a memorandum of understanding. However, Google is certainly thinking about how to mitigate the risks associated with unproven tech.
“While nothing is guaranteed, the Master Plant Development Agreement defines performance milestones and a pricing model to promote transparency and collaboration and align incentives between both companies,” a Google spokesperson told Latitude Media.
The range of nuclear technologies that companies are tapping for data center power is related to how quickly they predict they’ll need that power, explained Steve Swilley, vice president of nuclear research and development at the Electric Power Research Institute.
Tech companies and utilities that need to move quickly — as in, they need power in the next five years — are more likely to tap existing nuclear resources, in the vein of Microsoft, Swilley said.
Those with a timeline more along the lines of 2035 to 2040 “have a lot more options on the table,” he added. “You could probably sit back and wait a little bit, to let some of these things play out, and then triangulate where you need to be.”
Utilities, particularly smaller utilities, are unlikely to take the risk of developing SMRs without the opportunity to share that risk, Swilley added. That’s a reality reflected in Amazon’s deal with Energy Northwest; the public power agency said it had been working for years to develop the SMR project before Amazon stepped in.
“Taking this first, bold step is difficult for utilities,” said Greg Cullen, Amazon’s VP of energy services and development, in a statement. Amazon’s “financial strength, need for power and know-how” are pushing the project forward, he added.
