Comment: Microsoft deal doesn’t yet signal nuclear power revival

By Liam Denning / Bloomberg Opinion

When the Three Mile Island nuclear plant became (in)famous in 1979 for a reactor accident, Microsoft Corp. was a baby, still a year away from a landmark deal with International Business Machines Corp. that would pave the way for the Windows revolution.

So Microsoft’s latest landmark deal, to revive the plant’s other reactor to power its artificial-intelligence ambitions, positively drowns in symbolism. The longed-for nuclear renaissance has finally met its transcendent moment; or has it?

The debated U.S. nuclear revival conflates two things. One is keeping open the existing fleet, much of which was built in the 1970s and 1980s and began suffering mostly cost-related shutdowns over the past decade. The other concerns building genuinely new plants, an event rarer than total eclipses these days. Microsoft’s deal hews more to the former. It has agreed to buy 20 years’ worth of output at a fixed price, providing Constellation Energy Corp., the owner, enough certainty to spend $1.6 billion to put the reactor in Pennsylvania, which closed for economic reasons in 2019, back into service by 2028.

Constellation didn’t disclose the price but made certain to give analysts enough clues to work it out. The consensus has Microsoft paying about $100 to $120 per megawatt-hour, roughly double prevailing wholesale prices on the PJM grid, which the plant will serve. Unlike with an earlier deal between Amazon.com Inc. and Talen Energy Corp. concerning another nuclear plant in Pennsylvania, Microsoft isn’t taking the power directly to a data center. Rather, it will continue buying power from the grid and pay Constellation the difference to the prevailing market price, receiving credit for the zero-emissions electricity being generated.

This virtual power-purchase agreement is unambiguously bullish for Constellation, nuclear power and power markets in general. While the Amazon deal sparked excitement in March, its direct link between a reactor and a data center inherently limited applicability elsewhere.

There are also limited opportunities to restart other dormant reactors like the one at Three Mile Island, with NextEra Energy Corp.’s Duane Arnold plant in Iowa being the most obvious candidate. But a virtual supply contract means existing nuclear plants owned by independent generators can sign similar deals, which is why shares in Constellation, which owns more than half the country’s merchant nuclear capacity, have soared 24 percent since the Microsoft deal was announced. In a neat twist, Microsoft helped Constellation develop the software to enable this 24/7 matching of zero-carbon power with a customer’s needs. Plus, Microsoft’s willingness to pay $50 or more per megawatt-hour over the prevailing wholesale prices, a much bigger premium than Amazon’s, sets a new price floor and reinforces the sense that power markets overall are tightening. From a climate-change perspective, tapping the cavernous pockets of Big Tech to keep the U.S. nuclear sector going — and revive a few dormant plants — is a victory.

Yet there are still formidable obstacles before Big Tech takes what might be assumed to be the next logical step: contracting for new nuclear capacity.

This starts with costs. Constellation says it can restart the reactor at a cost of about $1,900 per kilowatt of capacity; a bargain compared with the almost $8,000 estimated for a generic new plant and the more than $15,000 that it cost for the last reactors built in the U.S., at Southern Co.’s Vogtle plant in Georgia. Even adding in distribution charges to arrive at an all-in cost to Microsoft of about $130 to $140 per megawatt-hour, that remains roughly a third of Bloomberg NEF’s estimated levelized cost of new nuclear capacity in the U.S. It is notable that Microsoft isn’t liable for any problems with Three Mile Island’s restart. “If it costs more, it’s on our dime,” Constellation’s chief executive officer explained to analysts.

Big Tech’s priority for data centers is getting a connection to the grid quickly, and preferably with zero-carbon power. Existing nuclear plants, or the few that can be revived relatively soon, are ideal; indeed, Pennsylvania Gov. Josh Shapiro is nudging PJM to speed Three Mile Island’s restart along.

New nuclear plants, meanwhile, are inherently long-term bets. Even the touted small modular reactors that seem ideal for data centers have suffered setbacks around costs and time lines similar to their larger forebears and seem unlikely to be commercialized at scale this side of 2030.

Big Tech will keep funding research efforts on small modular reactors, and even nuclear fusion, and remains the obvious buyer of any output. Not many companies can credibly sign a power contract worth a nominal $15 billion or so. (This assumes Microsoft paying $100 per megawatt-hour for 865 megawatts of capacity at 95 percent utilization over 20 years.) But in their rush for AI dominance, hyperscalers will also invest in, and contract with, other options in the intervening years.

These range from efforts to make processors and their cooling systems more efficient — a longstanding trend in tech — to contracting with natural gas plants. The latter, while carbon emitters, are plentiful on the PJM grid and can be marketed as being at least cleaner than coal. It should be noted that PJM sits atop the prolific Appalachian shale formation, where gas habitually trades relatively cheap. With that in mind, modular reactors will likely compete with another nascent technology, carbon capture, for Big Tech’s attention. In the background, of course, cheaper, faster, zero-emission supply from renewables and batteries will continue to take market share in electricity provision, even if data centers back that up with gas-fired power.

The biggest hurdle for new nuclear plants in the U.S. remains their long, and often variable, time line, presenting risks to investors and reasons to try alternatives. This is not to say no new reactors will be built at all. The combination of national security concerns and Big Tech’s balance sheet provides the best shot in a generation. In addition, federal support has amped up; indeed, roughly half the estimated profits for Constellation on the Three Mile Island deal will consist of tax credits. (This assumes revenue of $100 per megawatt-hour, fuel and operations costs of $40, depreciation and interest costs of $15, and corporate and state taxes of about $11. This results in an estimated net profit of about $34 per megawatt-hour. Clean energy tax credits add about $30 to this, equating to 47 percent of an adjusted net profit of about $64 per megawatt-hour.)

That same support, however, reflects the challenge in persuading private money to take on the pricing and timing risks of new reactors. Nuclear enthusiasts can pocket the victory and bask in the symbolism. Much else must go right for a renaissance to take hold.

Liam Denning is a Bloomberg Opinion columnist covering energy. A former banker, he edited the Wall Street Journal’s Heard on the Street column and wrote the Financial Times’s Lex column. More stories like this are available on bloomberg.com/opinion. ©2024 Bloomberg L.P.