A decade ago, small modular reactors were a niche engineering curiosity. Now data-center operators, utilities, and national governments are betting billions that they can deliver carbon-free power faster than a conventional plant ever could.
Small modular reactors, or SMRs, are nuclear plants designed to generate under 300 megawatts of electricity apiece — a fraction of the output of a conventional reactor — using components small enough to be built in a factory and shipped to site rather than poured and welded in place over a decade. That single design choice is the whole pitch: less custom engineering, shorter construction schedules, and, proponents hope, costs that fall as more units come off the line.
The renewed interest is not coming from environmental ministries. It is coming from server farms. Cloud providers training ever-larger AI models need firm, round-the-clock electricity, and they need it in a hurry — something wind and solar alone struggle to guarantee without heavy battery backup. Several hyperscale operators have already signed agreements with reactor developers to power new data-center campuses directly, bypassing years of queueing for grid interconnection.
The economics are still unresolved
Independent cost modeling has been far less enthusiastic than the marketing. Several academic analyses of the leading designs put projected costs well above today’s wholesale power prices, and the industry’s own history offers a caution: the flagship U.S. SMR project of the last decade was cancelled in 2023 after its target price crept steadily upward. Backers argue those numbers reflect a first-of-a-kind plant, not a mature factory line, and that costs will fall the way they did for solar panels once real order volume arrives.
“We are not choosing SMRs because they are the cheapest electron on the grid. We are choosing them because they are the only electron we can contract for a decade in advance.”
Only China and Russia currently operate a true SMR commercially, and licensing timelines in the West still run to years, not months. Whether the technology becomes the workhorse of the next grid or another expensive false start likely won’t be settled by engineering alone — it will be settled by whichever buyer is willing to sign the first dozen contracts at scale.
