BasicsTerraPower poured its first structural concrete for Kemmerer Unit 1 on April 23, 2026, the first ground broken on a commercial non-light-water reactor in the United States in more than four decades, according to the U.S. Department of Energy. That single fact explains why the SMR-versus-traditional-nuclear question has moved from conference panels to actual capital allocation decisions. Utilities, tech companies chasing data center power, and finance ministries are all trying to work out which technology fits their timeline, their balance sheet, and their appetite for risk. ⚛️
The two categories share a fuel type and a regulator’s scrutiny. Almost nothing else lines up cleanly. Here are the five differences that actually decide whether a project moves from slide deck to poured concrete: size, build time, capital cost, safety architecture, and the regulatory path each one has to walk. 🔬
Power output sets every other number that follows
A traditional large reactor, the kind of unit Georgia Power just finished at Vogtle, generates 1,000 to 1,500 MWe per unit, per the U.S. Energy Information Administration. An SMR, by the International Atomic Energy Agency’s own definition, tops out around 300 MWe, though the label has stretched further in practice than the definition suggests. 📊
Three live projects show how much that range actually spans:
NuScale Power’s VOYGR design: 50 MWe per module, with up to 12 modules stacked into one plant
Rolls-Royce SMR: a 470 MWe three-loop pressurized water reactor, the largest design currently in serious regulatory review anywhere
TerraPower’s Natrium: 345 MWe baseline output, boosted to 500 MWe for hours at a stretch using molten salt energy storage
A conventional PWR or BWR fleet unit: 1,000+ MWe, with no ability to flex output up or down 📐
That gap is the root of everything else on this list. A smaller reactor means a smaller absolute check for any single project, even when the cost per kilowatt runs higher. It also means a smaller physical footprint and, in theory, a shorter path through licensing. Whether that theory holds up is where things get genuinely interesting.
Build time and modularity rewrite the construction playbook
A conventional nuclear plant gets built the way a cathedral gets built: on site, mostly bespoke, over roughly six to ten years, with cost overruns baked into nearly every Western project of the last two decades. SMR developers are betting on something structurally different: factory-fabricated modules, shipped to site, and assembled in a fraction of the time. 🏗️
The GLOBSEC think tank frames the advantage well. Instead of building one 1,200 MW conventional reactor, a developer could build four 300 MW SMRs, and the first one can start generating revenue while the other three are still under construction. That kind of phased commissioning is structurally impossible with a traditional single-unit plant, which has to be fully complete before it produces a single watt. ⏱️
Worth asking before backing either approach: does the organization footing the bill have the balance sheet for a single decade-long project, or does earlier revenue from a phased build matter more to the finance committee?
Factory fabrication and quality control instead of weather-exposed, stick-built site construction
Phased commissioning, where module one earns revenue while modules two through four are still going up
A smaller crane, laydown yard, and peak workforce footprint on site
A serial learning curve across repeat builds, which is the entire theory behind projected cost declines from first-of-a-kind (FOAK) to nth-of-a-kind (NOAK) units
The capital cost picture is genuinely mixed, and worth being honest about
This is the difference that gets oversimplified the most, usually in whichever direction serves the person making the pitch. The plain numbers do not cooperate with a clean story. 💰
The EIA puts the overnight capital cost of a new large light-water reactor at roughly $7,800 to $8,100 per kilowatt in 2024 dollars, per its own reporting on U.S. SMR development. SMR estimates, by contrast, are scattered: recent 2025 and 2026 analyses put first-of-a-kind units anywhere from $3,500 to $10,500 per kilowatt, depending on the design and how far along it sits on the learning curve. The International Energy Agency has even projected SMR overnight costs in the EU running around $10,000 per kW, against roughly $6,600 per kW for conventional nuclear. On a pure per-kilowatt basis, SMRs are not obviously cheaper yet, and anyone claiming otherwise is quoting a nth-of-a-kind number for a first-of-a-kind project. 📉
NuScale’s Carbon Free Power Project remains the cautionary tale everyone in the sector cites. Its target offtake price climbed from $58 to $89 per megawatt-hour before its municipal utility customers walked away in November 2023. TerraPower is trying to avoid a repeat with Kemmerer Unit 1, whose total project cost is estimated at $4.0 billion, with the Department of Energy covering $2 billion of that through its Advanced Reactor Demonstration Program. The context that makes news like this meaningful is exactly what SMRbrief Pro was built to provide.
Large reactor overnight cost: roughly $7,800 to $8,100/kW in 2024 dollars, per the EIA
SMR overnight cost estimates: $3,500 to $10,500/kW depending on design maturity and source
First-of-a-kind SMR LCOE: roughly $80 to $180/MWh, against a projected $50 to $80/MWh once designs reach nth-of-a-kind production
Kemmerer Unit 1 total project estimate: $4.0 billion, including $2 billion in DOE cost-share
Safety systems trade active machinery for passive physics
A traditional plant survives a loss-of-power event with active machinery: pumps, redundant emergency diesel generators, and layered backup power chains that utilities added after Fukushima Daiichi. Most SMR designs are trying to make that machinery unnecessary. 🛡️
The IAEA describes the shift plainly: SMR safety concepts generally rely more on passive systems and inherent characteristics, namely low power and low operating pressure, rather than active intervention. NuScale’s design leans on natural convection to cool the reactor, with no pumps, no offsite power, and no operator action required in the critical first hours of an incident. TerraPower’s Natrium takes a different route: a low-pressure sodium coolant loop, paired with a design that physically separates the nuclear island from the energy and storage island, keeping most of the plant’s systems away from anything radioactive at all. ⚡
Passive cooling: gravity- and convection-driven, no pumps or grid power required, as in NuScale’s design
Low operating pressure: reduces both the likelihood and the consequence of a breach compared with a pressurized 1,000+ MWe unit
Decoupled nuclear and energy islands: TerraPower’s Natrium keeps non-nuclear systems physically separate
Smaller core inventory: simply less total radioactive material on site to manage in a worst-case scenario
Which trade-off would readers actually take: active safety systems with sixty years of accumulated operating history, or passive systems that have barely run a handful of commercial fuel cycles?
Regulatory pathways and siting flexibility diverge by design
The U.S. Nuclear Regulatory Commission does not run one licensing track for nuclear plants; it runs several, and which SMR a developer builds decides which door it walks through. NuScale’s original design went through Part 52 design certification, the traditional light-water route, becoming the first SMR design the NRC ever certified. TerraPower’s Natrium, a sodium-cooled fast reactor, went through Part 50 instead, receiving a construction permit on March 4, 2026, the first the NRC has issued for a commercial non-light-water reactor in more than 40 years. A new, technology-inclusive Part 53 rule took effect April 29, 2026, built specifically to give non-light-water designs a risk-informed alternative path. 🏛️
Britain runs a different process entirely. The Office for Nuclear Regulation, the Environment Agency, and Natural Resources Wales jointly run a three-step Generic Design Assessment, and Rolls-Royce SMR is currently the only design to have reached Step 3, a status that followed its selection as the UK government’s preferred SMR vendor. Internationally, the IAEA continues pushing licensing harmonization so a design certified in one country does not have to restart the entire review in the next. 🌍
None of that paperwork matters if a plant cannot go where the power is actually needed. Because SMRs carry a smaller footprint and, in some cases, a smaller required emergency planning zone, they can be sited on retiring coal plants like Kemmerer, in remote communities too small for a 1,000 MWe unit, and increasingly next to the data centers driving the sharpest new demand for firm, round-the-clock electricity. 📍
United States: Part 52 design certification for light-water SMRs, Part 50 construction permits for non-light-water designs like Natrium, and a new Part 53 technology-inclusive path live since April 2026
United Kingdom: a three-step Generic Design Assessment run jointly by the ONR, Environment Agency, and Natural Resources Wales
International: the IAEA works toward licensing harmonization so a certified design travels across borders without a full re-review
Siting: retiring coal sites, remote grids, industrial clusters, and data center campuses are all now realistic SMR locations in a way they never were for a 1,000+ MWe plant
None of these five differences settles the SMR-versus-traditional argument on its own. The reactor that ultimately wins in any given market probably will not be the cheapest per kilowatt on paper. It will be the one whose owner can actually finance, build, and license it on the timeline promised to investors, regulators, and the town next door. Which of these five variables would you least want to get wrong on a first-of-a-kind project?