Smaller nuclear reactors (SMRs) are a costly dead end, especially for AI

Abstract

For decades, prices for new nuclear plants kept rising, and they are now the most expensive form of power. But solar, wind, and battery prices kept dropping, becoming the cheapest. New reactors grew so costly the U.S. and Europe all but stopped building any. Nuclear’s share of global power peaked at 17% in the mid-1990s but was down to 9.2% by 2022 and 9.1% in 2024. New reactors are inflationary and lead to higher energy bills for consumers even if they’re never turned on. The only U.S. plant built in decades, the $35 billion Vogtle plant in Georgia, is “the most expensive power plant ever built on earth,” with an “astoundingly high” estimated electricity cost. Georgia ratepayers’ bills are rising by over $220 a year. In 2023, state regulators made customers pay for most of the cost of the reactors—“on top of a monthly surcharge” they’ve had to pre-pay for years, totaling $1000. South Carolina consumers still pay for two never-completed reactors. Since these 1100-megawatt (MW) reactors are so costly, “small modular reactors” (SMRs) under 300 MW have been hyped, especially for AI data centers and hydrogen. But SMRs are a dead end—with high risks of cost overruns, delays, and reliability/safety problems. That’s why efforts to commercialize them have failed for decades. Worse, Trump’s policies “severely increase the risk of expensive, unexpected nuclear accidents,” Scientific American warned in March. SMRs also have tariff risks since they need foreign sales, foreign uranium, and foreign components to succeed. For decades, reactors have kept getting larger to capture economies of scale. So, SMRs face significant shrinkage diseconomies and a higher cost per MW than large reactors like Vogtle. Cost escalation is endemic to SMRs (see figure below). So, SMRs would mean even higher rates for consumers than big reactors. In 2025, solar, wind, and batteries represent 93% of planned U.S. utility-scale electric-generating capacity additions. Also, recent studies find advanced geothermal energy is on track to provide baseload and potentially dispatchable power three times cheaper to build than Vogtle by 2030.