BasicsSmall modular reactors stopped being a slide-deck concept somewhere in the last eighteen months. TerraPower broke ground on an actual reactor vessel foundation in Wyoming. China’s CNNC is weeks from connecting a commercial SMR to the grid. Microsoft, Amazon, Google, and Meta have signed enough nuclear power purchase agreements to add up to nearly 10 gigawatts of committed capacity. And yet the public conversation about SMRs is still running on assumptions that were outdated the moment they were written down.
Some of that is genuine confusion between regulatory milestones that sound similar but mean very different things. Some of it is leftover skepticism from the industry’s most public failure. And some of it is just people repeating what they read in 2023 without checking whether it’s still true. Here are seven claims about SMRs that get repeated constantly, and what the record actually shows.
Myth 1: an NRC stamp means a reactor is basically getting built
NuScale Power is still the only company with a full NRC design certification for an SMR, and it holds two: the original 50 MWe module and, as of the agency’s February 2026 announcement, an uprated 77 MWe version, good for a 462 MW six-module plant. That’s a real, hard-won regulatory achievement. It is also not the same thing as a reactor under construction.
Design certification tells the NRC a blueprint is safe to reference. It says nothing about whether a specific utility will build it, finance it, or connect it to a grid. NuScale’s flagship project, the six-module Carbon Free Power Project in Idaho, was cancelled in November 2023 after utilities backed out over rising costs. Design approval didn’t save it.
Compare that with TerraPower, which has no NRC design certification at all for its Natrium reactor, a 345 MWe sodium-cooled fast reactor with molten salt storage. What it has instead is a construction permit, approved by NRC commissioners in March 2026, the first ever granted for a commercial non-light-water reactor, and workers are now pouring the actual foundation in Kemmerer, Wyoming.
The three-stage reality worth remembering:
Design certification: the blueprint is cleared, no site required
Construction permit: a specific plant at a specific site gets approved to build
Operating license: required before the reactor can actually run
NuScale has stage one locked down twice over and nothing under construction. TerraPower skipped straight to stage two and has cranes on site. Neither company has cleared stage three yet.
Myth 2: small reactors are automatically cheap reactors
The pitch has always been that factory-built modules dodge the cost overruns that plague giant one-off plants. The Carbon Free Power Project tested that pitch in the real world, and the real world won. NuScale’s target price for power from the Idaho project climbed from $58 per megawatt-hour to $89 per megawatt-hour, a 53% jump, before the project collapsed entirely under a $1.4 billion DOE cost-share deal that never got spent as planned.
That doesn’t mean the underlying economic argument is dead. It means it’s unproven at scale. The industry’s actual bet isn’t that reactor one is cheap, it’s that reactor six through ten get cheap through repetition, the same learning curve that brought down costs in naval reactor manufacturing. Most developers cite a target of roughly 40% cost reduction once a design hits serial production. TerraPower’s Kemmerer plant alone runs up to $4 billion for a single unit. Whether that curve actually bends the way the slide decks promise won’t be knowable until several identical units are built and operating, not modeled.
If you’re pricing an SMR deal today, ask which number you’re actually being quoted: the first-of-a-kind price, or the promised sixth-unit price. Those are very different bets.
Myth 3: no SMR has ever produced a real watt of electricity
This one was defensible until roughly six months ago. It isn’t anymore.
China’s CNNC completed cold functional testing on Linglong One, a 125 MWe pressurized-water SMR (also called the ACP100) at the Changjiang Nuclear Power Plant on Hainan island, in October 2025, and is targeting grid connection in the first half of 2026. First concrete went down in July 2021, putting the project on a roughly 58-month construction timeline, a real data point for anyone trying to model how long these things actually take once permitting is done. CNNC expects the unit to produce about 1 billion kilowatt-hours a year, enough for roughly 526,000 households.
Russia’s floating Akademik Lomonosov plant, meanwhile, has been quietly supplying grid power to the remote town of Pevek since 2020. It’s not glamorous, and it’s not American, but it’s real electrons on a real grid.
Worth sitting with: the “it’s all still theoretical” argument has a shrinking shelf life. Every quarter that passes without an update on where Linglong One actually stands is a quarter where the myth gets a little more wrong.
Myth 4: every SMR needs the same exotic fuel
Not even close. HALEU, high-assay low-enriched uranium, enriched between 5% and 20% U-235 versus the roughly 3% to 5% used in today’s fleet, is required by most of the advanced non-light-water designs: TerraPower’s Natrium, X-energy’s Xe-100, Kairos Power’s Hermes, Oklo’s Aurora, and Radiant’s Kaleidos all need it. Conventional light-water SMRs, including NuScale’s design and GE Hitachi’s BWRX-300, run on the same standard low-enriched fuel already used across the existing US reactor fleet.
The HALEU side of that split is genuinely tight. Centrus Energy remains the only US commercial HALEU producer, and its demonstration cascade in Piketon, Ohio, turns out roughly 900 kilograms a year, a fraction of the multi-ton annual demand once Natrium and Xe-100 start pulling fuel. DOE is throwing real money at the gap: a $900 million task order to Centrus in January 2026, part of a $2.7 billion, ten-year commitment, on top of a separate NRC Part 70 license X-energy’s fuel subsidiary won in February 2026 for a facility rated at 5 metric tons of uranium annually, enough to feed up to 11 Xe-100 units. The urgency traces partly to the 2024 US ban on enriched uranium imports from Russia’s Rosatom, which had been the dominant commercial HALEU supplier.
So: some SMRs face a real fuel bottleneck. Others could be fueled tomorrow with supply chains that already exist. Lumping them together obscures which projects are actually at risk of delay and which aren’t.
Myth 5: Big Tech’s nuclear deals are just press releases
By mid-2026, tallying every publicly announced hyperscaler nuclear commitment gets you to roughly 13 deals and 9.8 gigawatts of committed capacity across Microsoft, Google, Amazon, and Meta. These aren’t handshake agreements sitting in a drawer. Microsoft’s $16 billion, 20-year power purchase agreement for the Crane Clean Energy Center, the restarted Three Mile Island Unit 1, covers 835 MW and just got accelerated to arrive in the second half of 2027 after FERC approved a transmission waiver on June 1, 2026 that moved 760 MW of grid-connection rights over from a neighboring site. DOE closed a $1 billion loan for that project in November 2025.
Amazon put roughly $700 million into X-energy for up to 12 Xe-100 units at a Pennsylvania campus, part of a broader $50 billion partnership targeting 960 MW. Google signed an order-book arrangement with Kairos Power for 500 MW. Meta’s January 2026 round of agreements, spanning Vistra, TerraPower, Oklo, and Constellation, targets up to 6.6 gigawatts, including funding for two Natrium units and a 1.2 GW campus with Oklo in Pike County, Ohio.
Keeping all of that straight, who signed what, which MW figures are contracted versus aspirational, which regulatory approvals are still pending, is genuinely a full-time job at this point. The context that makes news like this meaningful is exactly what SMRbrief Pro was built to provide.
Myth 6: SMRs are a niche play for small or remote grids
The IAEA still frames part of the case for SMRs around islands, remote communities, and grids too small for a traditional 1,000+ MW plant, and that use case is real. But the gigawatt-scale numbers above tell a different story about where the actual capital is going right now. Data center operators aren’t buying SMRs because their local grid is undersized. They’re buying them because AI training clusters need firm, continuous power at a 95%-plus capacity factor, something intermittent solar and wind can’t deliver on their own. IDTechEx currently forecasts the global SMR market growing to $53.8 billion by 2036 and toward $300 billion by 2046, driven substantially by exactly this demand curve.
If your mental model of an SMR customer is a mining camp or an Arctic town, it’s time to update it. The 2026 customer is increasingly a hyperscale data center campus with a power appetite measured in hundreds of megawatts.
Myth 7: the US has this race locked up
American headlines about TerraPower’s construction permit and NuScale’s design certifications can create the impression the US is comfortably ahead. China’s timeline argues otherwise. Linglong One reaching commercial operation in the first half of 2026 would make it the first land-based commercial SMR to actually generate power, years before NuScale, TerraPower, or the UK’s Rolls-Royce SMR reach the same milestone. China had roughly 29 reactors under construction nationwide as of early 2026, a construction pipeline far larger than anything currently underway in the US, UK, or Canada combined.
The US retains real strengths worth naming:
The only construction permit ever issued for a commercial non-light-water reactor
The deepest pool of private capital, courtesy of hyperscaler nuclear deals
A dedicated, multi-billion-dollar federal push to solve the domestic HALEU bottleneck
But “ahead” isn’t a settled fact, it’s a contested claim that depends heavily on which milestone you’re measuring. Grid connection first, or commercial-scale advanced reactor technology first? Those might not be the same race.
Which of these seven gets corrected in the coverage you read next depends less on the technology and more on whether the reporting keeps pace with a sector that’s now moving on a quarterly, not yearly, clock.