Small Modular Reactors Finally Have Concrete in the Ground. The Economics Are Still Unproven.

The Slideware Years Are Over

For most of the last decade, small modular reactors were a category of renderings. Sub-300 MWe designs, factory-built and truck-delivered, promised to fix everything conventional nuclear got wrong: no ten-year builds, no multi-billion-dollar overruns, no bespoke engineering on every site. What they mostly produced was press releases.

That changed in 2026, and it changed in the least glamorous way possible — with permits, foundations and rebar.

What Actually Exists Now

Three projects are worth knowing by name, because they are the ones with physical work underway.

• Darlington, Ontario. Ontario Power Generation began construction on a GE Vernova Hitachi BWRX-300 in May 2025. Canada’s nuclear regulator cleared the first regulatory hold point — the reactor building foundation — on March 30, 2026, and OPG applied that same month for a 20-year operating licence. The first of four units is targeted for 2030, which would make it the first grid-scale SMR in the Western world.
• Kemmerer, Wyoming. TerraPower received an NRC construction permit in March 2026, the first the agency has ever issued for a commercial non-light-water power reactor, and broke ground the following month. Its sodium-cooled Natrium design targets up to 500 MWe and is going up at a retiring coal plant — reusing the transmission interconnection, the switchyard and the workforce.
• Regulatory queue. NuScale remains the only developer with an approved NRC design, uprated from 50 MWe to 77 MWe in May 2025, and booked its first revenue in the fourth quarter of 2025. X-energy’s Xe-100 high-temperature gas design has an NRC safety evaluation expected around November 2026.

None of this is a demonstration reactor on a national lab campus. These are utility projects with licences, financing and construction schedules.

The Number Nobody Puts on the Slide

Here is the part the enthusiasm tends to skip. Ontario’s four-unit Darlington programme is costed at C$20.9 billion in 2024 dollars for 1,200 MW. The first reactor alone accounts for C$6.1 billion, plus C$1.6 billion of shared infrastructure — cooling water tunnels, administrative buildings — that the three later units will also use.

Work that out and the first unit lands north of C$25,000 per kilowatt. The programme average improves considerably as the shared costs spread and each build learns from the last, which is precisely the modular thesis. But the opening number is not cheap power, and pretending otherwise is how this industry got its credibility problem.

We already have a case study in what happens when the arithmetic goes the wrong way. NuScale’s Carbon Free Power Project, six 77 MWe modules totalling 462 MW at Idaho National Laboratory, was cancelled in November 2023. The target power price had risen from $58/MWh in 2021 to $89/MWh, a 53% increase, driven by a construction estimate that went from $5.3 billion to $9.3 billion. Utilities were asked to subscribe 370 MW. They committed 116 MW, and the project died.

Nothing about the reactor failed. Inflation, interest rates and supply chains did the damage — the same forces that have always punished capital-intensive construction. Being small did not confer immunity.

What Changed Is the Buyer

So why is 2026 different from 2023? Not because the costs came down. Because a new class of customer showed up that weighs price differently.

Every major hyperscaler has now signed nuclear supply of some kind. Microsoft contracted with Constellation on a 20-year agreement to restart Three Mile Island Unit 1 — 835 MW of existing plant, first power targeted for 2027. Amazon anchored a $500 million round in X-energy in October 2024 and has contracted for output from Talen’s Susquehanna plant. Google signed with Kairos Power for an initial 50 MW demonstration, scaling toward roughly 500 MW by 2030. TerraPower’s $650 million Series C in June 2025 included NVIDIA’s venture arm.

Read those deals carefully and a pattern emerges: the near-term ones buy power from existing reactors, while the SMR commitments are equity, options and demonstrations. AI operators need firm, carbon-free power on a timeline that wind and solar permitting cannot meet, and they have balance sheets that tolerate a price a municipal utility could not. That is a genuinely new demand signal. It is not yet proof that a factory-built reactor is cheap.

What to Watch

The honest position is that SMRs have cleared the regulatory and engineering hurdles and have not yet cleared the economic one. Three checkpoints will settle it:

• The X-energy NRC evaluation, expected around November 2026 — whether a second reactor type reaches the buildable stage or the pipeline stays narrow.
• Darlington units two through four. The whole modular argument is that unit four costs dramatically less than unit one. Ontario is the only place that will actually test it this decade.
• Whether an SMR PPA gets priced in public. The hyperscaler deals disclosed so far are mostly for existing plants or early-stage equity. The first published price for factory-built new-build output is the number that matters.

If you are tracking this sector, ignore the announcements and watch unit costs on repeat builds. A first-of-a-kind reactor being expensive tells you nothing — every first one is. A fourth-of-a-kind still being expensive would tell you the modular premise itself was wrong, and that is the result the industry cannot survive.