Age of Nuclear-Powered AI
Last week, Google announced that it plans to buy power from a fleet of small modular nuclear reactors (SMRs) from a start-up, Kairos Power, to feed its exploding electricity demand for artificial intelligence. The first reactor is slated to come online by 2030, followed by additional deployments through 2035. While the tech giant called this the “world’s first” such corporate deal to buy nuclear power, it appears several other tech firms too have been turning to startups developing new nuclear technology as they scramble to meet the massive energy needs of their generative AI apps, data centers, and storage facilities. Amazon, for instance, bought a nuclear-powered datacenter in Pennsylvania in March, and last month, Microsoft struck a deal that would reopen a unit of the Three Mile Island nuclear plant, also in Pennsylvania, which in 1979 was the site of the worst nuclear accident in the US. This renewed push for nuclear power is fueled by the idea that, in addition to being emissions free, advanced nuclear technology is safer and more sustainable than traditional nuclear power plants. SMRs are especially appealing because they are faster and cheaper to build as well. But, of course, all this is too good to be totally true. As Edwin Lyman, director of the nuclear power safety program at the Union of Concerned Scientists, says, “the hype surrounding SMRs is way overblown.” Lyman, who follows the nuclear industry closely, wrote a detailed critique of next-gen nuclear reactors back in 2021 showing that, among other things, these reactors were not necessarily more economical, safer, or more reliable than the older ones. Nor do they do away with the vexing problem of radioactive waste. That critique, he writes, is still valid today. With both nuclear energy and AI, there is potential for good. But right now, the pitfalls along the way are many. We should proceed with caution.
|
