A DC Atlas Frontier
Module 2 of 6
What Is an SMR, Actually?
What "small" and "modular" mean in engineering terms, why the passive-safety case is a different argument from the one made for the current fleet, and what first-of-a-kind units actually cost.
Built on DC Atlas's asset-level data across ~8,000 tracked facilities.
Say the word "nuclear" to a room of infrastructure investors and watch the room divide. Half will picture a cooling tower silhouette against an orange sky, Chernobyl, Three Mile Island, spent fuel rods glowing in pool water. The other half will picture the $18 billion Amazon just committed to the Susquehanna plant. Both pictures are real. Only one of them is current.
This module is the bridge between those two mental models. We are going to explain what a small modular reactor actually is, how the engineering has changed since the reactors most people picture, and why the safety case for modern designs is not an incremental improvement over the past but a fundamentally different approach to preventing the failure modes that defined nuclear's worst moments. The audience for this is not nuclear engineers. It is the PE fund partner, the data centre operator, and the sovereign wealth advisor who needs to understand the technology well enough to evaluate the investment case that the rest of this series builds.
Why Did Nuclear Stop?
Nuclear energy did not fail as a technology. It failed as a business. The US fleet of 93 operating reactors generates roughly 20 percent of the country's electricity and is the single largest source of carbon free power in the nation. These plants run at capacity factors above 90 percent, outperforming every other generation technology. The fuel is cheap. The operating costs, once a plant is built and licensed, are among the lowest in the industry.
What went wrong was construction. The large light water reactors that defined the industry from the 1960s onward were massive bespoke engineering projects. Every plant was effectively a prototype. Costs escalated unpredictably. Schedules slipped by years. The regulatory environment, particularly after the Three Mile Island accident in 1979 and the Chernobyl disaster in 1986, added layers of review and compliance that extended timelines further. The Fukushima Daiichi accident in 2011 reinforced public fear and triggered additional regulatory requirements globally. The Vogtle Units 3 and 4 in Georgia were completed in 2023 and 2024. The project cost $34 billion. It delivered 2,234 megawatts of capacity. That works out to roughly $15,200 per kilowatt of installed capacity. And it was years behind schedule.
The fundamental problem was not that nuclear reactors are dangerous or unreliable. It was that building a one off, multi billion dollar, decade long construction project in a regulatory environment designed for maximum caution produced cost structures that no private capital could consistently underwrite. The small modular reactor concept exists to break this pattern.
What "Small" and "Modular" Actually Mean
The terms are precise. "Small" means less than 300 megawatts of electrical output, compared to the 1,000 to 1,600 megawatt behemoths that define the current fleet. This is not a toy reactor. A single 300 megawatt unit generates enough electricity to power roughly 250,000 homes, or a 250 megawatt data centre campus with margin to spare. "Small" is relative to the existing fleet, not to the load it serves.
| Characteristic | NuScale VOYGR | BWRX-300 | AP1000 (Conventional) |
|---|---|---|---|
| Electrical Output | 77 MW per module | 300 MW | 1,117 MW |
| Reactor Type | Pressurised Water (PWR) | Boiling Water (BWR) | Pressurised Water (PWR) |
| Thermal Efficiency | 30.8% | 34.5% | ~33% |
| Passive Cooling Duration | 30+ days (unlimited after) | 7+ days (extendable) | 72 hours |
| Operator Action Required | None for 72 hours | None for 7 days | Within hours |
| Black Start Capable | Yes (first nuclear plant) | No (external source needed) | No |
| Fuel Enrichment | 4.95% LEU (standard) | Standard LEU | Standard LEU |
| Construction Target | 3 to 4 years per module | ~4 years (first unit) | ~7 to 10 years |
| NRC Status | Design certified (2023) | Pre-application review | Certified and built |
"Modular" is the more consequential term. It means the reactor is designed to be manufactured in a factory, shipped to site in components, and assembled rather than built from scratch on location. This is the difference between constructing a building on site and assembling prefabricated sections delivered by truck. Factory fabrication enables quality control in controlled environments, parallel manufacturing of identical components, and a learning curve that improves with every unit produced. A conventional reactor is a construction project. A modular reactor is a manufacturing product. That distinction changes the economics fundamentally.
The NuScale VOYGR module weighs approximately 700 tons and can be transported by rail or heavy haul to its installation site. A plant can deploy one module, four modules, or up to twelve modules depending on the customer's power needs. Each module is identical to every other module. The BWRX-300, while larger, achieves its own version of modularity by reducing the volume of safety related concrete by 90 percent compared to its predecessor design, the ESBWR. GE Hitachi reports that the BWRX-300 requires less than one tenth of the concrete per megawatt compared to larger BWR designs, which translates directly into shorter construction schedules and lower site preparation costs.
The Fukushima Test
To understand why the safety case has changed, you need to understand what actually failed at Fukushima Daiichi on March 11, 2011. The popular narrative is that an earthquake and tsunami destroyed a nuclear plant. The engineering reality is more specific, and the specificity matters because it reveals exactly which design features modern SMRs have eliminated.
The sequence, reconstructed from the NRC's official post accident analysis, was this. The magnitude 9.0 earthquake struck at 2:46 PM Japan Standard Time. The earthquake itself did not cause the accident. All three operating reactors shut down automatically and correctly when their control rods inserted. The seismic ground motion exceeded the plant's design basis, but the reactor protection systems worked exactly as intended.
The earthquake destroyed the overhead power transmission lines connecting the plant to the grid. This caused a station blackout, the loss of all external AC power. This too was manageable. The plant had emergency diesel generators designed for exactly this scenario. The tsunami arrived approximately 40 to 50 minutes later.
The chain of failure was specific: loss of external power, followed by loss of backup power, followed by loss of cooling, followed by fuel damage, followed by hydrogen generation, followed by explosion, followed by release. Every link in that chain depended on one thing: the need for active, electrically powered pumps to circulate water through the core.
This is exactly what modern SMR designs eliminate.
Two Reactors, Two Philosophies
The two SMR designs most relevant to data centre deployment take fundamentally different engineering approaches to the same problem. Understanding the distinction matters because it determines everything from cooling water requirements to site layout to operational staffing.
NuScale VOYGR: The Underwater Approach
The NuScale Power Module is a pressurised water reactor, the same family of technology that powers most of the world's existing nuclear fleet and every nuclear submarine in the US Navy. But the implementation is radically different. Each module sits submerged in a massive below grade pool of water, approximately 20 feet wide and 53 feet deep. The reactor, its steam generators, and its pressuriser are all integrated into a single compact vessel enclosed in a steel containment, which is itself submerged in the pool. There are no external coolant loop pipes. There are no large bore connections between the reactor vessel and external steam generators. The entire nuclear steam supply system is a single sealed unit sitting in water.
“No operator action required for 72 hours following a postulated accident. After 72 hours, reactor building pool water boil-off and air cooling of containment provide long-term decay heat removal.”NRC Safety Evaluation Report, NuScale Design Certification
When the reactor shuts down, decay heat, the residual thermal energy produced by radioactive fission products even after the chain reaction stops, must be removed. In a conventional reactor, this requires electrically powered pumps. In a NuScale module, heat transfers from the reactor through the containment wall into the surrounding pool water by natural convection and conduction. No pumps. No electricity. No operator action. The pool water absorbs heat for more than 30 days before the water level drops below the top of the containment vessel. After that, natural convection to air and thermal radiation from the containment surface become the dominant cooling mechanisms. The NRC confirmed that this air cooling mode is "adequate for decay heat removal for an unlimited period without external inputs."
Unlimited. Without external inputs. That is the NRC's own language, not a marketing claim.
The NRC certified NuScale's design in January 2023 for the original 50 megawatt configuration and approved the uprated 77 megawatt configuration in May 2025. NuScale is the first and so far only SMR to receive full NRC design certification in the United States.
BWRX-300: The Simplified Direct Cycle
The BWRX-300 is a boiling water reactor, the tenth generation evolution of a technology family with over 50 years of operating history. The fundamental difference from a pressurised water reactor is elegantly simple. In a PWR, the reactor heats water under high pressure without letting it boil. That hot pressurised water then transfers heat through a steam generator to a separate secondary circuit, where water boils to produce steam that drives the turbine. Two circuits, two sets of piping, one heat exchanger in between.
A BWR eliminates the secondary circuit entirely. Water boils directly inside the reactor core. The steam passes through internal separators and dryers within the reactor vessel and goes straight to the turbine. One circuit. No intermediate heat exchanger. Lower operating pressure (approximately 1,000 psi compared to 2,000 psi in a PWR), which means thinner, lighter, less expensive pressure vessels.
The BWRX-300 achieves 34.5 percent thermal efficiency, slightly higher than NuScale's 30.8 percent. Both are significantly below the 60 to 64 percent efficiency of a modern gas combined cycle plant. This efficiency gap is real and matters for cooling water requirements, since lower efficiency means more waste heat that must be rejected. But for data centre applications, the trade off is straightforward: nuclear provides 24/7 carbon free baseload power with 20 year price certainty. Gas provides higher efficiency at the cost of fossil fuel dependence, volatile fuel pricing, and carbon emissions that make corporate climate commitments impossible to meet.
The BWRX-300's key innovation is the integrated isolation valve, attached directly to the reactor pressure vessel rather than welded to downstream piping. This eliminates the classic pipe rupture failure mode that drove much of the emergency core cooling system complexity in earlier BWR designs. GE Hitachi reports that the BWRX-300 achieves a 90 percent reduction in safety related concrete compared to its predecessor, the ESBWR, and approximately 50 percent less building volume per megawatt.
Do SMRs Need HALEU Fuel?
Nuclear fuel supply is one of the most misunderstood aspects of the SMR landscape, and getting it wrong can derail a project entirely. The critical distinction is between two types of uranium fuel, and which designs require which.
Standard low enriched uranium, or LEU, is enriched to less than 5 percent uranium 235. This is the fuel used in every operating commercial reactor in the United States and the fuel specified for both the NuScale VOYGR (at 4.95 percent enrichment) and the BWRX-300. The supply chain for LEU is mature, commercial, and global. Fabricators include Westinghouse, Framatome, and GE Vernova, all with established US production facilities. Lead times are measured in months, not years. There is no supply chain risk for LEU fuelled SMRs that does not also apply to the existing fleet of 93 operating reactors.
High assay low enriched uranium, or HALEU, is enriched to between 5 and 20 percent uranium 235. This is required by several advanced reactor designs including Oklo's Aurora, TerraPower's Natrium, and X Energy's Xe 100. Until 2024, Russia's TENEX was effectively the only commercial supplier. The US banned Russian enriched uranium imports in 2024. Congress appropriated $2.72 billion for domestic LEU and HALEU production. Centrus Energy's Piketon, Ohio facility has produced initial HALEU quantities, targeting 900 kilograms per year. TerraPower's Natrium project was delayed by approximately two years specifically due to HALEU unavailability.
| Fuel Type | Enrichment | Designs Using It | Supply Chain Status |
|---|---|---|---|
| Standard LEU | < 5% U-235 | NuScale, BWRX-300, all existing fleet | Mature, commercial, multiple suppliers |
| HALEU | 5 to 20% U-235 | Oklo, TerraPower, X-energy | Constrained. Russia banned. US scaling up |
| HEU (weapons grade) | > 20% U-235 | None (commercial) | Not applicable |
The enrichment market itself is undergoing structural change. Russia controlled approximately 44 percent of global uranium enrichment capacity through Rosatom's subsidiary TENEX before the import ban. Western enrichment providers, primarily Urenco (a Dutch, German, and British consortium with a US facility in New Mexico), Orano (France), and the emerging Centrus domestic programme, are scaling capacity but face a multi year ramp. The World Nuclear Association projects that Western enrichment capacity will need to nearly double by 2035 to replace Russian supply and meet growing demand from both the existing fleet and new reactor deployments.
For a data centre operator evaluating a 20 year nuclear commitment, the fuel supply question is material but manageable. LEU fuelled designs carry lower supply chain risk. HALEU fuelled designs offer potentially superior performance characteristics (longer fuel cycles, higher burnup) but face near term supply uncertainty that adds timeline risk to any project dependent on them.
How Much Does an SMR Actually Cost?
We will build the complete economic model in Module 4, but the cost landscape deserves an honest framing here because it is the single biggest objection raised against SMR deployment, and the data does not allow simple answers.
Mixed basis: first-of-a-kind actuals, a first-unit budget, and two projections. The \
First of a kind SMR costs are high. Ontario Power Generation's Darlington BWRX-300 project, the most advanced SMR construction programme in North America, carries a first unit budget of approximately CAD 6.1 billion for 300 megawatts, translating to roughly $14,670 per kilowatt. This is comparable to the Vogtle AP1000 experience at $15,200 per kilowatt. The full four unit Darlington complex is estimated at CAD 20.9 billion, with OPG projecting a 33 percent cost reduction between the first and fourth units. NuScale's cancelled Carbon Free Power Project saw costs escalate from $5,100 per kilowatt to over $20,000 per kilowatt before the project collapsed.
The economic case for SMRs rests on the learning curve. Idaho National Laboratory models a 15 percent learning rate per cumulative doubling of units. On that assumption, deployment of 32 identical units would yield a 55 percent cost reduction from first of a kind levels. The Tennessee Valley Authority takes a more conservative view. TVA estimates nth of a kind SMR capital costs at $12,471 per kilowatt. That is a 30 percent reduction from TVA's first of a kind estimate of $17,949. South Korea, the only country with a demonstrated track record of sustained nuclear cost reduction, achieved a 23 percent reduction over more than two decades of continuous construction.
“The economic case for SMRs depends on whether factory fabrication and modular manufacturing generate learning rates comparable to aerospace, or whether nuclear-specific licensing constraints hold costs to historical patterns.”Idaho National Laboratory, Advanced Reactor Cost Analysis
The honest assessment is this: first of a kind SMRs are expensive. The learning curve is real but unproven at the scale required. What makes the investment case credible is not that SMR power will be cheap in absolute terms, but that the alternative, grid dependency at current PJM prices plus capacity charges plus backup generation costs plus carbon exposure, is becoming more expensive faster than most models anticipated. Module 4 constructs that comparison in full. The numbers may surprise you.
Questions this module answers
- What does "small modular reactor" actually mean?
- "Small" means less than 300 megawatts of electrical output, against the 1,000 to 1,600 megawatt reactors that define the current fleet. "Modular" means the reactor is manufactured in a factory, shipped to site in components, and assembled rather than built from scratch on location. A single 300 megawatt unit still powers roughly 250,000 homes, or a 250 megawatt data centre campus with margin to spare.
- How long can an SMR cool itself with no power and no operator?
- A NuScale module requires no operator action for 72 hours after a postulated accident, and its pool absorbs decay heat for more than 30 days with no pumps and no electricity. The NRC confirmed that the air cooling that follows is "adequate for decay heat removal for an unlimited period without external inputs." The BWRX-300's passive system runs for 7+ days and is extendable.
- Do SMRs need HALEU fuel?
- The two SMRs most relevant to data centre deployment do not. The NuScale VOYGR uses standard low enriched uranium at 4.95 percent enrichment and the BWRX-300 uses standard LEU, both drawing on a mature commercial supply chain with lead times measured in months. HALEU, enriched to between 5 and 20 percent, is required by Oklo, TerraPower and X-energy, and its supply remains constrained.
- What is the difference between the NuScale VOYGR and the BWRX-300?
- The NuScale VOYGR is a pressurised water reactor producing 77 MW per module, submerged in a below grade pool, with 30.8% thermal efficiency and NRC design certification since January 2023. The BWRX-300 is a boiling water reactor producing 300 MW with 34.5% thermal efficiency, running a direct cycle at approximately 1,000 psi rather than the 2,000 psi of a PWR. NuScale can black start without external grid power; the BWRX-300 has not claimed that capability.
- How much does an SMR cost per kilowatt?
- First of a kind SMRs are expensive. Ontario Power Generation's Darlington BWRX-300 carries a first unit budget of approximately CAD 6.1 billion for 300 megawatts, roughly $14,670 per kilowatt, comparable to Vogtle at $15,200 per kilowatt. The Tennessee Valley Authority estimates nth of a kind costs at $12,471 per kilowatt, a 30 percent reduction from its first of a kind estimate of $17,949.