Data Centers Gobbling Up Existing Nukes Threatens Grid Decarb Goals

As AI-hungry data centers drive significant spikes in electricity demand, companies, regulators, grid operators, and consumers face difficult questions. How can we meet new data center demand in a way that doesn’t drive up costs for all consumers, or result in increasing emissions from the power sector from a rush to gas? And to get a sense of scale, a recent EPRI report estimated data centers could grow to consume 4.6% to 9.1% of U.S. electricity generation annually by 2030 (but worth noting the report also included recommendations to manage this growth).

More recently, a new wrinkle has emerged: the nuclear option.

The biggest players in the data center space—namely Amazon, Google, Microsoft, and Meta—also have very aggressive near-term climate targets. As a result, they’re keenly interested in securing not only reliable, affordable power that can deliver nearly 24/7, but also clean power. Unfortunately, that is driving some of those companies to explore tapping in to existing nuclear supply. And in a recent transaction between merchant generator Talen Energy and Amazon—the first of its kind—plans are being put in place to divert nuclear power (long subsidized heavily for the clean power it delivers to the grid) to data centers.

This would be a bait and switch on decades of public funding of nuclear power. While that deal gives the appearance that Amazon is powering its data center with zero emissions electricity (technically it is), the net result would actually be an increase in emissions of carbon on neighboring grids, and pollution like NOx in the communities near gas plants that would come online to meet demand that was previously met by nuclear. (Notably, Exelon and AEP have also recently challenged that transaction on consumer impact grounds; outcomes of that dispute will set an important precedent nationally on the issue).   

What's the Scale of the Exposure Here?

As a one-off, the Talen/AWS transaction is a bad outcome from an emissions perspective, but it could be recovered from. However, that’s not the only one. 

We’re seeing increasing interest from a number of other companies that own existing nukes to replicate this sort of agreement. Fed up with volatile energy and capacity markets in ISO-NE, PJM and MISO, many nuke plant owners would be more than happy to sign a long-term power purchase agreement with a data center company, thereby providing that coveted revenue certainty, and the comparable bump in stock price that would ensue.  

We’ve heard noise around Dominion’s Millstone plant in Connecticut, PSEG is practically hanging out a shingle from their New Jersey plants, and the previously retired Palisades plant in Michigan also appears to be attracting some attention. Notably, both the NJ and Palisades plants are also lining up to divert power to hydrogen production as part of the DOE’s Hydrogen Hubs, where federal grants and tax credits offer the potential for similarly stable and lucrative revenue streams for nuclear operators as data centers. And as the Washington Post reports, there are even (very early and vague) rumblings surrounding an interest from Constellation in restarting Unit 1 of the infamous Three Mile Island nuclear plant that was retired 2019 because it was no longer economic.

All told, based on 2022 generation data from these facilities, if even half of the existing nukes in the wholesale electricity markets of PJM, MISO, NYISO, and ISO-NE were to enter into behind the meter agreements with data centers (or electrolyzers) in the coming years, it could result in roughly 155 million short tons of carbon being emitted annually until that zero emissions capacity could be backfilled.

We’re already in a massive “carbon hole.” Digging deeper by cannibalizing existing zero emissions resources is precisely not how to dig our way out.

Wait, what? Taxpayers Potentially Subsidizing a Nuke Plant that No Longer Supplies Energy to the Grid?

There’s also a major fairness issue at play here. As a result of concern from many decisionmakers that the existing nuclear fleet was at risk of retiring, the Inflation Reduction Act (IRA) included a generous production tax credit (PTC) for existing nukes through 2032. However, at the time, no one could have foreseen the risk posed by those same plants taking the money and running by abandoning their existing obligations and customers and then going behind the meter to contract with a data center.

One solution to correct for this exposure would be to deny PTC eligibility to plants that choose to go behind the meter. Under that construct, if a data center company wants to provide an above market premium for that nuclear power and also make-up for the PTC that same plant is losing, then so be it. But the economics of such a contract would look very different with that added cost.

Without Additionality, We’re Backsliding on Power Sector Decarb

NRDC is not new to reckoning with the impact of massive new load commitments to the present-day grid. 

We have long advocated for the “three pillars” in the context of ensuring hydrogen production and use in fact reduces emissions overall. Without these guardrails, hydrogen will  mutate into an expensive and dirty boondoggle that ends up setting us back on emissions. One of these pillars is additionality (also known as incrementality), which simply means that if we’re going to power something new with zero emissions electricity—whether it be a hydrogen electrolyzer or data center—it must be powered by a resource that is additional (i.e. not “existing”). Otherwise, that new demand is in fact directly responsible for increasing emissions from the broader grid, since removing an existing nuclear or renewable energy plant from the grid supply immediately results in those same MWh being backfilled by gas plants for at least the near- and medium-term. Due to lengthy interconnection queues, new clean energy to fill this gap could take years to come online.

What's Next?

NRDC is fundamentally not “anti-data center.” Hard stop.

But without comprehensive plans for how to mitigate potential negative consumer and emissions impacts from that new large electricity demand, we will be setting ourselves back significantly in the climate fight. (This post is focused squarely on grid impacts from data centers; water impacts are another area that developers and regulators should keep a close eye on to build in guardrails, such as this initiative from Google to recycle wastewater).

The good news? We have the technological, regulatory and financing tools to avoid that negative outcome. Microsoft’s recent agreement with Brookfield for 10.5 GW of new renewables is a great start. We stand ready to work with data center developers, state and federal regulators, and other stakeholders to lean into this challenge and thereby turn it into an opportunity to dramatically accelerate the clean energy solutions we have at the ready—we just need a lot more of them. And fast.