AI boom is great news for the nuclear power dreamers
Uranium is so hot right now, mining CEO is glowing with enthusiasm
Growing demand for power hungry AI datacenters has executives at Canadian uranium mine Cameco glowing with anticipation.
Speaking on the resources firm's Q1 2024 earnings call Tuesday, CEO Timothy Gitzel indicated the computationally intensive nature of AI workloads, and the challenges faced obtaining adequate power for the datacenters where they run, will drive up sales of uranium for use as a nuclear fuel.
Datacenter power demands are expected to swell considerably over the next few years. An International Energy Agency report [PDF] from earlier this year found that global datacenter power consumption was on track to nearly double from 460TWh in 2022 to just over 800TWh by 2026.
This trend has spurred new interest in nuclear power plants and reactor tech, as the cloud and hyperscalers leading the AI boom struggle to secure enough power and other resources to support their growing infrastructure while holding true to their sustainability and net-zero commitments.
"Keeping both clean energy and energy security in mind, the outlook is more positive in the context of the current debate around AI, datacenters, and the considerable energy needs that are coming with the rapid evolution of technology," Gitzel explained, citing claims that generative AI queries consume 10x the power of a Google search. "Gone are the days of rolling out new technology without worrying about future potential runaway environmental impacts … nuclear is the clear winner."
As we've previously reported, some of this is already playing out as the land around existing nuclear power plants becomes prime real estate for bit barns looking to secure access to large quantities of power.
One of the largest examples we've seen so far was Amazon Web Services' $650 million acquisition of Cumulus Data's nuclear-powered datacenter campus in early March.
The datacenter site is located alongside the 2.5 gigawatt Susquehanna nuclear power plant in northeast Pennsylvania. Under the terms of an agreement, the station will guarantee Amazon access to up to 960 megawatts of power for the cloud giant’s expanding datacenter footprint at the site over the next several years.
Startup NE Edge is also looking to get in on the action with plans to build a pair of AI datacenters adjacent to the two gigawatt Millstone nuclear plant in Waterford, Connecticut – but has faced headwinds winning support for the project among residents concerned such a project could drive up energy prices.
These kinds of datacenter-generator colocation projects offer a couple of advantages, chief among them being direct access – sometimes referred to as "behind the meter" – to buy power at reduced prices.
- Think tank funded by Big Tech argues AI's climate impact is nothing to worry about
- As AI booms, land near nuclear power plants becomes hot real estate
- AI energy draw from Chicago datacenters to rise ninefold
- AI is changing search, for better or for worse
If setting up shop next to an existing nuclear power plant isn't on the cards, datacenter operators are increasingly becoming utilities in their own right – building large solar, wind, geothermal, and natural gas plants to reduce reliance on the main grid.
For instance, last northern summer, Microsoft received approval for a 170 megawatt gas power plant to keep its €900 million ($960 million) datacenter development outside of Dublin from running out of juice. Amazon is also employing natural gas fuel cells at some of its Oregon datacenters for similar reasons.
Natural gas power plants and fuel cells don’t help to reduce the companies' carbon footprints – quite the opposite in fact – so Microsoft recently began exploring the use of small modular reactors (SMRs) to power future datacenter builds.
SMRs are essentially miniaturized nuclear power plants – not unlike the ones found in submarines and aircraft carriers. The idea is that a modular design will enable mass production, lowering the price of nuclear energy considerably.
It’s important to note that SMRs remain an idea – none have been built. Analysts predict it'll likely be a decade or more before we see the first one deployed. ®