Welcome to the new Energy Central — same great community, now with a smoother experience. To login, use your Energy Central email and reset your password.

Dan Yurman
Dan Yurman
Expert Member
Top Contributor

Major IT Platforms Want Power from Nuclear Plants, but They Don’t Want to Build Them

Anyone who thinks the big IT platforms like Microsoft, Google, Facebook, and Amazon, among others, are going to do anything more than offer power purchase agreements for 24X7 365 reliable power from nuclear reactors is engaging in wishful thinking. This became clear during a panel discussion this past week (6/16) at the meeting of the American Nuclear Society held in Las Vegas, NV.

Two recent wire service reports indicate this is not an option for “big data.” Despite some  exploration of “risk sharing” by DOE Secretary Jennifer Granholm, and staff from the DOE Loan Program Office, the answer to the question about these firms sharing the risk of investing in SMRs and full size nuclear reactors is definitely a “no.  However there is some innovation underway related to power purchase agreements involving several nuclear utilities and their big data customers.

On June 4th DOE Secretary Jennifer Granholm basically said there is no such thing as a free lunch when it comes to power for AI related data centers.  She said that if big IT firms and their data centers want 9% of US electricity power, they should consider building SMRs to provide it. Granholm pointed out that the Biden administration is asking big technology companies to invest in new climate-friendly power generation to cover their surging demand.

“We’ve been talking with data companies. The large ones have commitments to net-zero and would like to see clean baseload power,” Granholm said in an interview with Reuters.

“If the tech companies are coming in and are going to pull clean power from the grid, they should bring the power with them.”

She said the administration had discussed the possibility that companies could band together to make use of small modular reactors for nuclear energy, and could simultaneously place orders to reduce costs.

This is the “fleet” approach that uses economies of scale relative to supply chains to get costs down for multiple reactors. In the UK Rolls-Royce has planned a fleet of 16 470 MW PWRs with this principle at its heart.

The concerted opposition to this position by big IT firms planning to build build nuclear plants to power hyperscale data centers means that despite the current and future estimated rapid increases in demand for electric power to run large data centers that host power gobbling artificial intelligence semiconductors, all the risk to build new full size reactors and/or small modular reactors (SMRs) is going to be carried by the utilities that want to sell power to these data centers.

For evidence of this state of affairs, consider two recent wire service reports by Bloomberg and S&P Global about the panel discussion and related industry developments.

According to Bloomberg, the aforementioned four big data firms have no reservations about getting their power from nuclear reactors. This isn’t an issue of “green” renewables v. industrial nukes. They just don’t want to have to put up the cash to build reactors.

Big Data Won’t Rolls the Dice to Invest in Nuclear Power Plants

In keeping with the Las Vegas setting for the nuclear energy industry gathering, it became clear from the panel’s discussions that the big four big data firms, and probably others, do not want to roll the dice, or gamble putting cash on the table, for small nuclear reactors or nuclear plants of any size.

S&P Global reported that multiple big data industry experts joined in a chorus of “no way” statements about risk taking related to investing in nuclear power plants.

During a panel discussion at ANS meeting, Briana Kobor, an executive at Google, said that the company is willing to pay a premium for the reliable power for its data centers, but it draws the line on investing in the reactors that would provide the power.

Tech companies “are not going to write checks,” said Peter Freed, former clean Energy Buyers Alliance board member and former director of energy strategy at social media company Meta. He added that big data centers will do business with nuclear power utilities through power purchase agreements

“We do need to have nuclear technology to meet the load growth” of cloud data centers and AI centers” Freed added.

At the same ANS panel, Adrian Anderson, general manager for energy and sustainability at Microsoft, said technology companies should focus on their strength, building data centers, and rely on utilities to deliver the power.

In pointed remarks, he added, if a data center needs clean wind power, the operator should reach out to a utility, not a wind-turbine provider. He applied the same principle to nuclear power.

“The notion that individual buyers should be assessing nuclear technology for risk is a ridiculous concept. We are not going to own a nuclear project. Company capital is better deployed on data centers because we make money.” Anderson said.

Repurposing Power Purchase Agreements

Goggle’s Kobor also told the ANS panel that power purchase agreements (PPAs) have a single purpose, and that is to buy electricity. She said big data firms do not see PPAs as being vehicles to be used to leverage investments in new grid capacity or new reactors large or small.

Instead, Kobor mentioned a recent agreement her company and others signed with Duke Energy. Earlier this month responding to growing demand, Duke Energy, Amazon, Google, Microsoft and Nucor Steel executed agreements to accelerate clean energy options. It is expected that the new pricing plan will be more efficient, less time consuming, and less costly to administer than traditional RFPs.

In memorandums of understanding (MOUs) signed this month, the companies proposed developing new rate structures, known as “tariffs” in the utility industry, designed specifically to lower the long-term costs of investing in clean energy technologies like new nuclear reactors, large and small, and long-duration storage, through early commitments.

The proposed Accelerating Clean Energy (ACE) tariffs would enable large customers like Amazon, Google, Microsoft and Nucor to directly support carbon-free energy generation investments through innovative financing structures and contributions that address project risk to lower costs of emerging technologies.

ACE tariffs would facilitate beneficial on-site generation at customer facilities, participation in load flexibility programs and investments in clean energy assets – features attractive to customers with large-scale energy needs like hyperscale data centers.

The ACE tariffs would represent new, voluntary pricing structures for Duke Energy’s large commercial and industrial customers. Duke Energy’s five-year capital plan will continue as planned and these tariffs would be subject to regulatory approvals in North Carolina and South Carolina. It seems plausible that Duke has built into the new tariffs revenue streams to feed new capital investments which might include SMRs, new full size reactors, or uprates at existing plants.

Duke’s Reserve of NRC Licenses

Duke Power also has something else available that could plausibly feed a longer term investment horizon for nuclear power and hyperscale data centers. It has licensing work complete for six Westinghouse 1,150 MW AP1000s. An NRC COL is valid for 40 years.

  • Levy Nuclear Plant, Units 1 and 2 (NRC issued COL July 2008)
  • William States Lee III, Units 1 and 2 (NRC COL December 2007)
  • Shearon Harris, Units 2 and 3 (NRC suspended February 2008)

If Duke ever decides it wants to get back in the business of building full size reactors, it has NRC licenses for the equivalent of 6,900 MW of nuclear power to offer to rate payers including big data centers. The states of Florida, North Carolina, and South Carolina would very likely be delighted to see Duke build these reactors to attract big data firms to locate there.

Southern Not Ready for More New Reactors

While Duke is innovating with rate structures to engage big data customers, at the ANS conference Chris Womack, CEO of Southern Company, clearly has some reservations about building any more AP1000s now that the new two Vogtle AP1000s units are in revenue service.

An Associated Press headline from May 2023 has had the effect of chilling the ambitions of nuclear utility CEOs to build new full size reactors. This includes Womack. AP reported, “Georgia nuclear rebirth arrives 7 years late, $17B over cost.”

Womack’s experience qualifies as meeting the Everett Dirksen test of big cash outlays for big projects. Urban legends report the famous congressman and House Speaker saying, more or less, “a billion dollars here, a billion there, pretty soon you are talking about real money.”

There was some encouragement for new large reactors at the ANS meeting from DOE Secretary Granholm, who sees the need to keep supply chains running and reactor construction expertise current. This means starting new builds of AP1000s in the US in the next few years.

Womack declined to come up to room temperature in support for these concepts. Yet, like any good CEO, he is apparently keeping his options open saying that there is a potential for Southern to make a decision on new reactors by the end of this decade. Even so, it will take more than one or two new full size reactor projects to move the nuclear industry beyond first of a kind construction challenges.

What About Insurance for Cost Overruns?

One of the reasons big data firms react with pearl clutching horror over the idea of making investments in new nuclear reactors to power hyperscale AI centers is the risk of massive schedule delays and cost overruns. The experience with the two Vogtle reactors is clearly on their minds.

At the ANS panel there was also a discussion about the DOE Loan Program Office offering insurance, or other financial incentives, to buffer the risks of cost overruns for new reactor construction.

In a “background” email exchange about the overrun insurance concept with a veteran  expert on nuclear energy investment strategies, the following observations were shared with this blog about the prospects of such an arrangement.

“Any loan facility (and any insurance approach would need provisions to guard against “moral hazard” (i.e., having people just plan to use the loan) and misuse. Those would include, among other things: 

  • commitment to a set of best practices;
  • independent diligence on project engineering and
  • budget; a “deductible” of some type, where the sponsor would have to cover some first portion of the overrun; and
  • a copayment, so the loan would have to be matched by further sponsor contributions.
  • in addition, since it would be a loan, it would have to be repaid.

“There are a lot of other issues here so these items just scratch the surface”

 “The challenge with an insurance approach is that there is very little data on the likelihood of a cost overrun, which would make insurance very difficult to price, and there is a very small population of projects. Insurance works best when there is a large population to share the risk and there is a lot of data to establish the likelihood of events. 

Asked if the “RAB investment method” used in the UK  could work in the US, the expert said, “it would be difficult in the US because the rate regulations are by state, and many states have abandoned them. I think the components from RAB that could be implemented here, independent of the rate-regulatory structure, are price supports for early nuclear energy projects and the kind of cost-overrun protection some have been discussing.”

Separately, he added that on the other hand, the expert said that although the DOE loan program office has “plenty of loan authority,” revolving loan authority is generally a good approach. Perhaps that will change in the future and a revolving approach could be helpful”

Data Center Demand for Power Continues to Surge

Several financial models of supplying power to data centers from existing nuclear reactors are already in use according to a data center trade press report. According to a June 10th research note from S&P Global Ratings, data centers could drive 35 GW of additional load growth by the end of the decade. This number presents opportunities for existing nuclear plants to secure supply deals with nearby data centers. Potential nuclear operators poised for these deals include Constellation Energy Group, PSEG Power, and Vistra.

According to the S&P report cited in the trade press report, Talen Energy’s deal with Amazon Web Services (AWS) may serve as a model for future power purchase agreements with nuclear utilities.. Talen built a data center near its Susquehanna nuclear energy plant in Pennsylvania and later sold the facility to AWS, along with a 10-year power supply contract.

S&P’s report indicates that this model could be replicated by other nuclear operators like Constellation Energy, PSEG Power, and Vistra, which have dual-unit sites. Public Service Enterprise Group (PSEG) in New Jersey is currently in talks to supply data centers with capacity from its Hope Creek and Salem nuclear power plants in southern New Jersey, according to a statement last April from CEO Ralph LaRossa.

The S&P report noted, “Nuclear complexes such as the Salem and Hope Creek generation stations are potential candidates. They have adequate interconnection, significant generation redundancies, and are ideally located to provide edge computing near load centers.”

Constellation, the largest nuclear operator in the U.S., is exploring the possibility of building advanced reactors at existing plants to power data centers. Vistra is reportedly in discussions with data center operators in Texas and Ohio.

# # #

4 replies