'It's Time To Be More Realistic': Data Centers' Nuclear Ambitions Suffer Setback

A future in which data centers are powered by small nuclear reactors, easing the substantial energy use that plagues the property type, is likely further away than industry leaders had hoped.

Recent strides by companies hoping to deploy small modular reactors were a source of optimism earlier this year, with influential voices in the data center world pointing to nuclear solutions to a worsening power crunch. But setbacks for these projects in recent weeks brought with them a reality check.

“With emerging technologies there’s a danger of believing your own hype instead of taking a realistic view, and I kind of bought into it,” said Compass Datacenters executive Tony Grayson, a retired Navy submarine commander and one of the data center industry’s leading voices on nuclear power. “We need to take a reset and be realistic on the timelines. We’re looking at the mid-2030s right now.” 

Proponents say SMR plants can be built more cheaply, faster and in more places than traditional nuclear plants, with none of the risk of large-scale environmental catastrophe often associated with nuclear energy. Those elements make this kind of plant attractive for data centers as the industry seeks to decarbonize and solve its power problems.

For much of 2023, it looked like data centers might be the use case that helped bring SMRs to commercial viability, with the announcement of the first SMR-powered data center project and a series of favorable decisions from federal regulators. This success fueled widespread predictions that the first commercial SMRs would be operational by the end of the decade and would likely become a key part of the solution to the data center industry’s energy woes. 

But that may have been a mirage.

This month, a project by a Utah utility group that was supposed to be the first use of a commercial SMR was abruptly canceled. The announcement came after growing concerns over the viability of NuScale Power Corp., maker of the only federally approved modular reactor and widely regarded as being closest to launching a commercial SMR product.

The loss of NuScale’s most credible customer amid investor accusations of escalating costs and misleading claims has even the most vocal advocates for nuclear power within the data center space tempering their expectations that SMR-powered data centers will be a reality anytime soon.

SMRs like NuScale’s are closer in size and scale to the reactors powering the Navy’s submarines and aircraft carriers than to the massive nuclear generating stations typically used by utilities. Generally producing between 20 and 300 megawatts, they require a fraction of the footprint needed for older, gigawatt-scale nuclear plants.

Small reactors aren’t a new technology — the first nuclear-powered warship launched in 1955. But previous small reactors were custom-built and far too expensive to be commercially viable. It is only within the past two decades that companies like NuScale, Rolls Royce and Toshiba, supported by significant government subsidies, began making serious strides toward modular designs that could potentially be manufactured cheaply enough to provide electricity at competitive rates. 

SMRs have had a growing allure within the data center sector. Developers and operators in major markets have been struggling to find development sites with access to the massive amounts of electricity these facilities need, all while facing pressure to phase out fossil fuels. SMRs, at least in theory, can be built quickly and can be located close to data center clusters to skirt transmission constraints. 

In the first 10 months of 2023, SMRs started to look like a real near-term solution to the data center industry’s power problems, with NuScale leading the way. In late January, for the first time, U.S. nuclear regulators fully approved a NuScale-designed small modular reactor for commercial use. That followed a decision to allow SMRs to be built closer to densely populated areas than traditional nuclear plants, significantly increasing the number of potential development sites. 

Meanwhile, NuScale claimed to be on track for what would be the first commercial SMR deployment in the U.S.: a power plant for a conglomerate of Utah utilities called the Carbon Free Power Project that was scheduled to be completed by 2029.

Momentum built in early October when NuScale announced it was partnering with developer Standard Power on the first SMR-powered data centers, including a pair of gigawatt-scale crypto mining facilities in Pennsylvania and Ohio. By all appearances, NuScale was ahead of the pack in making SMR-powered data centers a reality in the short-term future.  

Yet over the following 30 days, NuScale suffered a series of setbacks and blows to its credibility that forced leaders across both the nuclear and data center sectors to significantly temper their expectations about not only NuScale, but the viability of commercial SMRs in general.

On Oct. 19, NuScale’s share price plummeted after short-seller Iceberg Research announced it was taking a short position on the company. Iceberg published a scathing analysis of NuScale, portraying its progress toward deploying a commercial SMR as largely smoke and mirrors.

The report claimed NuScale’s Utah project was unlikely to proceed and accused Standard Power of being “a fake customer” with a “pipe dream” project that NuScale had only partnered with to distract from the impending collapse of the Utah reactor. 

“This contract has zero chance of being executed as Standard Power clearly does not have the means to support contracts of this size,” the report read. “Crypto mixed with nuclear energy — What could possibly go wrong?”

While NuScale pushed back on some of these assertions, Iceberg’s skepticism regarding the CFPP proved prescient when Utah Associated Municipal Power Systems and NuScale announced in early November that the project had been canceled. 

The decision killed what would have been the first commercial SMR in the US. And with no other projects anywhere near as far in the development and regulatory approval process, data center industry insiders say the news was a reality check that commercial SMRs are a lot further away than many believed. 

“One of the realities that’s setting in for people is that we’re not going to be powering data centers or any industrial applications this way anytime soon ... we're talking about 2030 as a timeline just for proof of concept, test reactors and things like that,” said Alan Howard, principal analyst at Omdia’s cloud and data center research practice and the co-author of a 2022 study on SMRs. “I don't see anybody on the nuclear side of the industry going ‘oh, my God, this is a nail in the coffin,’ but these are definitely growing pains.”

NuScale has been regarded as a bellwether for the commercial potential of SMRs in general, but opinions within the data center industry are split on the degree to which the past month’s turmoil reflects issues unique to NuScale or whether it casts doubt on the technology’s potential for widespread adoption. 

On one hand, the cancelation of the CFPP was made significantly more likely by the structure of the contract between NuScale and UAMPS, a deal Howard called “an ill-conceived agreement to begin with.” The contract could be terminated if the dozens of municipalities who buy power from UAMPS didn’t commit to buying 80% of the project’s power by the end of this year. Less than 30% was spoken for at the time of the project’s cancelation. 

The contract effectively created an unwieldy number of co-developers for NuScale to navigate, according to Marc Bianchi, a managing director at TD Cowen specializing in energy transition and nuclear power. None of those municipalities carried any financial risk if the project failed, due to NuScale’s promise to refund development costs. So, while it may have been a bad contract and a poor choice for a first customer, Bianchi said it isn’t necessarily reflective of broader issues with SMRs.

“The project that got terminated was sort of a unique situation,” Bianchi said. “There’s reason to believe that there were some unique hurdles that may not apply to projects done by a for-profit entity.”

But experts point to one of the primary reasons some of those utilities weren’t willing to commit to buying power from the CFPP: the escalating costs associated with the project. 

The original target power price for the CFPP in 2020 was $55 per megawatt-hour, but by 2022 that target price had grown closer to $100 per MWh, even with significant federal subsidies. Over the same period, there was also a significant jump in the project’s overall development costs, which include both the reactor and the facilities needed to support it. These costs ballooned from $3.6B for 720 MW in 2020 to $9.3B last year, despite the project being reduced to 462 MW. 

Compass’ Grayson said these cost overruns and expanding development timelines reflect challenges with commercial SMRs more broadly that are just now coming into focus. While the appeal of SMRs is that they can be deployed at a fraction of the time and cost as a traditional nuclear project, Grayson said the data center and nuclear industries are being confronted by the reality that early SMR projects are going to see the same runaway costs and delays that have plagued the industry for decades. 

But even if modular reactors do become cheaper and faster to build over time, Grayson said the hype around the technology has ignored the fact that the reactor itself is only half the battle when bringing an SMR plant online. He pointed to the rising development costs at the CFPP as evidence that if the sector continues to struggle with the build-out of the plant supporting the reactor, it can render the cost and speed benefits of the reactor itself meaningless. 

“Ultimately an SMR is a construction project, and I think we lied to ourselves a little bit when we said SMRs are going to make things easier,” Grayson said. “Yes, it's going to make manufacturing your reactor easier, but you still have to build the darn plant, and nuclear has never done well at large construction.”

“In the past it took 35 years to build a reactor, now you’re going to get it down to 10 years when all you’re changing is manufacturing the reactor vessel,” Grayson added. “Maybe it’s time to be more realistic on the timelines of that happening.”