Editorial: New nuclear should power decarbonization

It wouldn’t be that hard to cut greenhouse gas emissions if we could source abundant and affordable energy without releasing carbon dioxide. Nuclear power has the potential to deliver just that: a stable, large-scale energy source with few CO₂ emissions.

Although renewable electricity such as wind and solar can also power electrolytic hydrogen production, nuclear has the unique ability to provide constant low-CO₂ electricity 24/7, heedless of sunlight and weather­—a critical advantage for commercial-scale plants, which also run around the clock. Some readers may be heartened, then, that nuclear power has experienced a dramatic comeback in popularity over the past year or so. But there is a curious twist.

Utilities are building and restarting nuclear power plants not to decarbonize but to fuel the energy demands of artificial intelligence systems and data centers.

In September, Microsoft signed a contract with Constellation in which the software giant will pay to restart a retired reactor at Three Mile Island. Separately, Google expects to start up a small modular reactor (SMR) from Kairos Power by the end of the decade. Amazon and Meta are eyeing nuclear power too, though regulators halted specific deals at those firms over the past few weeks.

While climate change destroys lives and cities in real time, industry is pulling out all the stops to power computers that talk back to us better.

A more useful way to capitalize on the green promise of nuclear energy would be to use the carbon-free energy source to power chemical processes such as hydrogen production.

Hydrogen has immense potential to curb carbon emissions, and methods for making it without greenhouse gas emissions pair well with the energy output profile of nuclear plants. Fuel cells, engines, and boilers can use hydrogen instead of fossil fuels to generate heat, mechanical work, and electricity.

Beyond hydrogen’s use as a combustion fuel is the potential to use it as an electron source. For example, it can reduce iron oxides to metallic iron in steelmaking, it can strip oxygen atoms out of biomass to convert agricultural and food waste into useful chemical feedstocks, and it can catalytically combine with captured CO₂ to form alcohols and hydrocarbons.

The problem is that most of the hydrogen on the market today is made from fossil methane using a process that spews out more than 7 kg of CO₂ for every 1 kg of hydrogen it produces. Electrolysis can split water into hydrogen and oxygen without involving carbon, but that process demands a ton of electricity, making it only as sustainable as the power plant it’s connected to.

Besides electricity, nuclear energy can also provide direct heat for chemical plants so they don’t have to burn fossil fuels. Most plants run on steam, both to heat reactions and to provide mechanical force. That’s Dow’s plan in Seadrift, Texas, where the chemical maker intends to replace its aging natural-gas fired boiler with four SMRs rated at about 200 MW each.

But making nuclear a viable energy source for hydrogen production will require more R&D. New nuclear chemistries and plant types, such as thorium-fueled molten-salt reactors and helium-cooled SMRs, are designed with inherent safety in mind. And dozens of SMR designs have been proposed and are being tested around the world. Yet SMR technology has had a slow start because of research, capital, and regulatory costs. In fact, cost was a major factor in NuScale Power’s 2023 decision to scrap its planned SMR plant in Idaho.

Volumes have been written about why nuclear projects always run over their capital expenditure budget, but one reason is that each is a bespoke, first-of-its-kind infrastructure project. So maybe it’s not so bad for computer and software companies to foot the bill as the early adopters. And maybe if we get good at building nuclear plants, even if it’s to power ChatGPT, the rest of society can use that technology to decarbonize things that matter more for the climate.

This editorial is the result of collective deliberation in C&EN. For this week’s editorial, the lead contributor is Craig Bettenhausen.

Views expressed on this page are not necessarily those of ACS.