Why We Should Scale Water‑Cooled SMRs

01Introduction

Every measure of human flourishing — lifespan, income, health, the time we spend on work we choose rather than work we must do — rises with energy consumption. And roughly four fifths of that energy still comes from burning fossil fuels.

As long as that remains true, humanity is forced into a bad trade: limit energy consumption, or accept ever more air pollution and environmental destruction. Neither arm of the trade is acceptable. Asking billions of people to stay poor is not a climate strategy, and burning the atmosphere to lift them out of poverty is not an energy strategy. The way out is not to use less energy. It is to make clean energy so cheap and so abundant that the trade disappears.

Nuclear fission is the one proven energy source with the physical headroom to actually replace fossil fuels rather than merely supplement them. A single fuel pellet the size of a fingertip carries the energy of a ton of coal. A fission plant runs day and night, in any weather, at any latitude, on a pad measured in acres rather than square miles. The fuel is abundant, the waste is small and contained, and the operating record — some twenty thousand reactor-years worldwide^[4] — is the deepest evidence base in the history of energy technology.

The moment for scaling it has arrived, because demand has arrived. Artificial intelligence is the largest new load the grid has seen in a generation: the International Energy Agency projects data-center electricity consumption will more than double to roughly 945 TWh by 2030,^[1] and US data centers alone are on track to consume 6.7–12% of the nation's electricity within a few years.^[2] This load is a worst case for intermittent generation: it cannot be sited only where the sun is strongest, it must run around the clock to pay for the silicon inside it, and it is growing faster than transmission can be permitted. As Sam Altman put it, "we need fusion or we need radically cheaper solar plus storage… at scale."^[3] Fission is the third option his sentence leaves out — and unlike the first, it works today.

Demand shocks of this size mint new energy companies. Many will be created in the next few years; a handful will define how the next terawatt gets built. Mission-driven companies can seize this opportunity to build toward a future in which human well-being and its ecological impact are finally decoupled — where the energy that powers a better life does not have to be paid for in a degraded world.

This paper makes the case for a specific way to seize it: small, water-cooled pressurized water reactors, built in factories, shipped complete, and deployed by the thousands. We explain why fission is the only energy source that can carry firm clean power to civilizational scale (§2), how manufacturing — not physics — sets today's nuclear costs and how volume resets them (§3), why light water is the coolant that lets a manufacturing company move fastest (§4), and how a factory-sealed, truck-shipped, submerged containment turns a nuclear construction project into a delivery (§5–6). We then examine the alternative honestly: a quantitative model, built on a decade of satellite weather data, of what it actually costs to serve an always-on load with solar, batteries, and diesel backup (§9) — and what cheap firm energy would mean for the climate problem as a whole (§10).

02Why nuclear fission?

An energy source fit to power the next century has to clear three bars at once: it must be clean, it must be scalable, and it must be baseload. One technology sits in the middle of that Venn diagram.

By clean we mean no combustion products — no carbon dioxide, no particulates, no smog. By scalable we mean deployable nearly anywhere on Earth, at terawatt scale, without exhausting a geographic niche. By baseload we mean firm: power that is there every hour of every day, independent of weather and season.

Solar and wind are clean and magnificently scalable — and intermittent by physics, not by engineering immaturity. Hydro and geothermal are clean and firm — and locked to the rivers and geology that make them work; the good sites are largely taken. Fusion will join the center of the diagram if and when it works commercially; it does not work today. Fossil fuels are scalable and firm — they built the modern world on exactly those two virtues — and they are the pollution problem we are trying to solve. Only fission clears all three bars, today, with sixty years of operating history behind it.

Venn diagrams are cheap; numbers are better. The honest way to compare fission with its strongest competitor — solar photovoltaics plus batteries, with diesel generators backing up the gaps — is to ask what it costs to serve a constant, always-on load as you raise the fraction of that load the clean system must carry. We built that model from hourly satellite weather data (the full methodology, assumptions, and additional locations are in §9). Figure 2 shows the result for West Texas — arguably the best place on Earth to try this, with world-class sun and unlimited flat land.

Read the curve left to right. At 90% clean — diesel engines covering one kilowatt-hour in ten — solar plus storage delivers energy at about $156/MWh. Respectable. But each additional "nine" of clean reliability roughly compounds the price: $274/MWh at 99%, $404/MWh at 99.9%, and $506/MWh at 99.99%, more than three times where the curve started — in the best solar geography in America. In cloudier places the multiplier is worse (§9). The cause is structural, not incremental: the last few percent of hours are long winter cloud runs that only enormous batteries can bridge, and battery capacity purchased for a two-week event earns revenue on no other day of the year.

The reference lines make the comparison uncomfortable in both directions. Even Vogtle-class new-build nuclear — the most expensive nuclear power ever constructed, at $142–222/MWh^[5] — undercuts 99%-clean solar in the best solar state in the country. Meanwhile the operating US nuclear fleet produces power for about $32/MWh all-in.^[6] That gap — between what nuclear costs to run and what it costs us to build — is the entire opportunity, and it is the subject of the rest of this paper.

To be clear about what this comparison does and does not say: at low firmness, solar is genuinely cheap, which is exactly why it dominates today's marginal grid additions, and why we expect it to keep growing. The claim is narrower and more important. For the loads that will define the next decades of demand — always-on compute, industry, electrified everything — the market is asking for firm clean power, and on that request the solar cost curve bends the wrong way precisely where the demand curve is steepest.

03How to scale fission quickly

Nuclear power's problem has never been the physics. Fission's cost problem is a manufacturing problem — and manufacturing problems are the kind humanity knows how to solve.

3.1 The gap is volume, not physics

A useful diagnostic for any manufactured product is the ratio of its finished cost to the cost of its raw materials — the "idiot index," in the phrase popularized by SpaceX.^[9] An index near one means the manufacturing is approaching the physics limit. A large index means the cost lives in process, overhead, and organization — all of which yield to scale.

Table 1. Ratio of deployed cost to raw-material cost across manufactured products. The pattern is volume: products made by the millions sit at 2–4×; products made by the handful sit at 50× and above.

A 1 GW pressurized water reactor contains roughly $90–160M of raw material — steel, concrete, copper, zirconium, uranium. Vogtle Units 3 and 4 were deployed for roughly $35 billion, about $15,000 per kilowatt^[7] — an idiot index above 100. The same reactor design, built by Korean industry at Barakah, came in near $5,500/kW; Chinese domestic builds land at $2,800–3,500/kW. Same physics, same steel — a threefold-to-fivefold difference in industrial organization.

Zoom into a single component and the anatomy of the gap becomes visible. The AP1000 reactor pressure vessel is forged from SA-508 Grade 3 low-alloy steel — a specialty steel, but not an exotic one, at roughly $1,200–1,500 per tonne. As raw forgings and cladding, the vessel is $1–2M of material. As nuclear-qualified material — the same steel with witnessed melts, full traceability, and certified test reports — it becomes $16–29M, a premium of 11–22× on metallurgically identical metal. As a finished, delivered vessel: $100–150M. Yet the petrochemical industry routinely buys hydrocracker reactors — comparably sized, comparably thick-walled, fully ASME-documented forged vessels — for $26–30M, because it buys them at volume. Applied to the AP1000 vessel, that benchmark implies a well-scaled cost of $22–42M.

The gap between $22–42M and $100–150M is not physics. It is volume.

3.2 Vertically integrate the supply chain and the deployment

Why hasn't volume happened? Because virtually every piece of the historical nuclear value stream is fragmented and bespoke. A utility hires an architect-engineer, who hires an EPC contractor, who buys from component vendors, who subcontract forgings, who bill cost-plus through four layers of stacked margin and paperwork. No single party owns the end-to-end cost, so no single party can compress it. Financing costs during a decade of construction add 30–50% on top of everything beneath them; engineering, procurement, and project management overheads add 30–40% more. None of this buys a single additional watt.

The remedy is the same one that transformed launch: one company that designs the product, builds the product, and deploys the product — and therefore feels every dollar of cost and every week of delay as its own. Forge's insourcing sequence follows leverage, easiest first: final assembly (integrate vendor forgings, tubes, and valves); the tube shop (everything plumbing-related is cheap and fast to internalize precisely because the working fluid is water); the forge (pressure-vessel forgings are the industry's deepest bottleneck and its highest-margin gap — a $94M internal forging line replaces $12–20M-per-reactor external shipsets at roughly $2M each, repaying itself in under a year of production); then balance-of-plant, then fuel. The forge step matters beyond our own bill of materials: only four companies on Earth can produce ultra-heavy nuclear forgings, none of them American. China identified this bottleneck early and scaled heavy forging. The West has not.

3.3 Build in factories and ride down the cost curve

Every manufactured product that got cheap got cheap the same way: cumulative volume. Wright's law — costs fall a fixed percentage with every doubling of units produced — took solar modules down 98% since 2000^[8] and lithium batteries down nearly as far. Nuclear is the one energy technology that has never been allowed on this curve. Western reactors are built as one-of-a-kind civil-works projects: every project restarts from zero, re-learns its lessons, re-qualifies its processes, and disbands its workforce at the end. Meanwhile the regulatory documentation burden grew roughly sevenfold during the 1970s alone^[10] and never contracted — overhead that a factory amortizes across a production run but a construction site pays fresh every time.

Moving the reactor into a factory inverts all of it. Tooling, process qualification, and inspection get paid for once and amortized over hundreds of units. Statistical process control replaces artisanal paperwork — quality that is measured into the process rather than documented onto each part. Work moves indoors, to fixed stations, with a permanent workforce that gets better every unit. This is not a hypothesis about nuclear; it is the observed behavior of every serially manufactured product in Table 1, including the safety-critical, heavily regulated ones.

3.4 Co-design the factory with the product — and build the first unit in it

A factory bolted onto a finished design inherits every one of that design's manufacturing mistakes. The order matters: the reactor and the production system that builds it must be designed together, by the same people, at the same time — and the first unit must come out of the factory that is capable of scaling. A prototype hand-built in a lab teaches you how to hand-build prototypes. If the first article doesn't come down the line, the line's lessons arrive after the design is frozen — and the team designs the wrong hardware.

"Manufacturing-first" is therefore not a company ethos; it is a set of concrete changes to how the company operates. Designers work alongside manufacturing from the first sketch, designing both the product and the system that builds it. Designers are physically present for the build and deployment of their hardware until it is produced at rate. And the metric that governs the development lifecycle is not time to first prototype but time to at-rate production — because in factory-built nuclear, thoughtful design-for-manufacture creates order-of-magnitude cost differences, and its absence balloons both cost and schedule.

3.5 Make sure the design is deployable

A factory-built reactor that needs a bespoke site is a contradiction: the savings leak out through the civil works. Deployability has to be designed in. That means no dependence on rivers, lakes, or cooling towers — Ember rejects its heat to dry air, so siting is not hostage to hydrology. It means every subsystem arrives as a factory-built, factory-tested module in a standard shipping envelope, set on a pad, bolted down, and connected. On-site work is final installation, not construction. Megawatts should take months, not years.

3.6 Build the largest reactor that ships

Small is a means, not an end. Many costs of a nuclear plant are nearly flat with power level — licensing, siting, security, operations staff, grid connection — so dollars-per-megawatt-hour improves as those fixed costs spread across more megawatts. If reactors were free to be any size, bigger would win; that logic built the gigawatt fleet. But the moment a component outgrows the road network, deployment falls off a logistics cliff (§6): factory-build stops being possible, heavy site construction returns, and with it the cost structure of §3.1. The optimum is therefore sharp: the largest reactor whose complete power block still ships as routine freight. Every design decision at Forge pushes power upward until it hits that constraint, and no further.

04Why water-cooled?

Choosing a reactor coolant is choosing a supply chain, a regulatory history, and a set of problems to spend the next decade solving. Judge every candidate on three axes — fuel, coolant, moderator — and light water wins all three at once.

A company whose thesis is cost reduction through serial manufacturing should pick the reactor that lets it spend its engineering budget on manufacturing, not on re-deriving nuclear technology. That principle, applied honestly, eliminates the exotic coolants fast.

4.1 Gas-cooled reactors fail on logistics

For a factory-and-shipping thesis, high-temperature gas reactors disqualify themselves almost immediately. Helium is a poor heat-transfer medium — its volumetric heat capacity is a small fraction of water's — so a gas core needs several times the volume for the same thermal power, and its graphite moderator multiplies that volume again. The result cannot ship as a sealed, fueled power block at any interesting power level; it must be assembled on site, which surrenders the factory advantage that motivated the exercise. The fuel compounds the problem: TRISO particle fuel is technically impressive, but no commercial TRISO fabrication capacity exists in the United States at fleet scale — building it is a parallel, decade-scale industrial program that must run alongside reactor licensing. We prefer to fight one decade-scale industrial battle at a time.

4.2 Sodium trades one hard problem for five

Sodium fast reactors are the serious alternative, and their advantages are real: low pressure, high power density, excellent thermal margins. But walk the three axes.

Coolant. Sodium melts at 98 °C, so every meter of pipework must be heat-traced and insulated so the coolant never freezes — even in shutdown; thawing a frozen sodium line without cracking it is a delicate, slow operation. It reacts violently with water and burns spontaneously in air, so the entire circuit lives under an argon blanket, and opening it for maintenance means draining, purging, and chemically passivating first. Conventional centrifugal pumps give way to electromagnetic pumps; valves come from a thin specialty supply base; dissolved corrosion products must be continuously managed. Every sodium-wetted component is an exotic, low-volume specialty item. And because sodium and water must never meet, the design demands an entire intermediate coolant loop — pumps, heat exchangers, pipework, instrumentation — that contributes nothing to power generation. It exists to manage a hazard created by the coolant choice itself.

Fuel. Most sodium designs need high-assay low-enriched uranium (HALEU), enriched to 5–20%. Commercial HALEU production outside Russia barely exists; TerraPower's flagship slipped years when its Russian fuel supply evaporated in 2022.^[11] A supply chain with one geopolitically compromised thread is not a foundation for a production line.

Moderator. Fast reactors dispense with the moderator — a real simplification — but fast-spectrum operation is itself a regulatory novelty that the NRC framework was not written to evaluate. Every analysis method, every precedent, every reviewer instinct at the regulator was formed on thermal-spectrum water reactors.

4.3 Water wins all three axes simultaneously

Fuel: a global commodity. Standard UO₂ fuel enriched below 5% is produced by multiple qualified fabricators across the US, Europe, and Asia — tens of thousands of assemblies per year, to tight tolerances, from an enrichment industry with spare capacity. The 17×17 pressurized-water fuel assembly is one of the most thoroughly characterized industrial products ever made. A developer building a PWR-based reactor inherits this infrastructure at no cost. There is no fuel qualification program on our critical path, because the fuel was qualified decades ago.

Coolant: water requires no supply chain. It is universally available and chemically boring. Every pump, valve, fitting, and heat exchanger comes from a large, competitive market — or can be produced internally on ordinary capital equipment, which is precisely what makes the vertical-integration ladder of §3.2 climbable. There is a quieter advantage that matters enormously to a factory: leak testing, quality verification, and factory acceptance testing are all done with water. Every completed module can be wrung out on the factory floor with the actual working fluid, cheaply and safely, before it ships.

Moderator: there isn't one. In a pressurized water reactor the coolant is the moderator. Light water's most elegant simplification is that an entire category of fabrication, maintenance, and operational complexity — graphite blocks that swell and crack, moderator handling, dust management — simply does not exist.

Beneath all three axes sits the evidence base: roughly twenty thousand reactor-years of civil operating experience, the large majority of it in water-cooled plants,^[4] and a regulatory framework written by and for light-water technology. Sixty years of operating experience means solved problems stay solved.

None of this is to say the exotic coolants were foolish choices — they were deliberate trades, made for real advantages, by serious engineers. But departing from the light-water baseline means confronting an immature industrial supply base and a regulatory framework that was not written for your technology — both of which add capital and time to scale. For a developer whose thesis is cost reduction through serial manufacturing, those are the wrong problems to be solving first.

The light-water PWR is not convenient on one dimension — it wins on all three simultaneously.

05Submerged, shippable containment

The containment building is where nuclear construction projects historically went to die: years of nuclear-grade concrete, poured on site, inspected rebar by rebar. Forge's answer is to make containment a factory product — a sealed steel capsule that ships with the reactor inside it, and spends its working life underwater.

Ember's containment is a compact steel pressure capsule, fabricated, outfitted, and leak-tested in the factory, enclosing the reactor vessel, the steam generators, and the primary loop. It is not a building; it is a component. That distinction carries the whole economic argument of §3 across the last hurdle: the one structure that historically had to be site-built nuclear construction becomes another item on the production line, welded indoors, tested with water, and driven to the site.

At the site, the capsule is installed below grade, submerged in a water-filled pool. Submersion is not a flourish — it does three jobs at once:

• An always-on heat sink. The pool surrounds the containment with a passive thermal reservoir. In any event that takes the normal heat-removal path offline, decay heat conducts and convects through the capsule wall into the pool — no pumps, no valves, no operator action, no external power. The water inventory buys days of passive grace, and topping up a pool is a fire hose problem, not an engineering problem.
• Shielding and protection. Meters of water and earth over the capsule provide radiation shielding and physical protection from external events as a side effect of the installation geometry, rather than as purpose-built hardened structure.
• A conventional site. With containment and its heat sink packaged this way, the building above is ordinary industrial construction — a metal enclosure over a lined concrete pool. The nuclear-grade scope stays in the factory, where §3 says it belongs.

Operating water-cooled reactor modules underwater is not a novelty the regulator has never seen: NuScale's NRC-approved design places its modules in a shared below-grade pool, establishing precedent for the configuration.^[18] What is new here is the role the capsule plays in the business: the containment is the shipping envelope. The unit that leaves the factory — sealed, tested, fueled at the appropriate stage of the logistics chain — is the unit that gets lowered into the pool. Nothing is field-erected around it.

06Logistics constraints

The American road network is the largest machine ever built for moving heavy objects to arbitrary points. Design a reactor to ride it, and every industrial site on the continent becomes a candidate power plant.

The rules of the road are concrete and public.^[17] A standard tractor-trailer moves up to 80,000 lb gross on interstates without any permit, inside an envelope 13'6" tall and 8'6" wide. Above that, a well-understood regime of routine oversize/overweight permits reaches to roughly 16 feet of width and into the low hundreds of thousands of pounds on multi-axle heavy-haul trailers — the loads that move transformers, excavators, and bridge beams every day of the week. Beyond that lies superload territory: months of per-state permitting, route surveys, police escorts, utility crews lifting lines — and finally the loads that cannot move by road at all. A gigawatt-class pressure vessel at ~500 tonnes is in the last category: it travels only by barge and heavy rail to the handful of sites that have them, then crawls the final miles on self-propelled transporters. The map of places such a plant can exist is thin lines along rivers and rail spurs.

Ember is designed backward from these limits. The heaviest single shipment — the lower vessel assembly with its steam generators and pumps — is capped at 120,000 lb, squarely inside the routine heavy-haul class. Every other module of the plant, from the dry-cooling units to the chemistry skids, packs into standard 40-foot ISO high-cube envelopes at container freight weights. No barges, no route studies, no crawler transporters. If a truck can get there, your power can get there too.

The consequence is a different shape of company. A reactor that ships as routine freight is manufactured at the tempo of a factory and deployed at the tempo of a logistics operation. Site selection stops being a multi-year study of rivers, rail spurs, and crane paths, and becomes a real-estate question. And the addressable market stops being "the places heavy industry can reach" and becomes, to first order, the road network — which is to say, everywhere.

07Sizing the reactor power level

08Cost analysis

09The challenge of scaling solar to meet AI demand

We are not solar skeptics: photovoltaics are the cheapest marginal electron humanity has ever produced, and Table 1 shows why — solar rode the factory cost curve all the way down. The problem is not the panel. The problem is the night, the winter, and the two cloudy weeks in a row.

9.1 What the sun actually delivers

A solar array's output is set by geometry and weather. Geometry is deterministic: the sun rides lower in the winter sky, days shorten, and output falls — by a factor that grows sharply with latitude. Weather is statistical: cloud systems arrive in runs, and the runs matter far more than the averages. Any honest model of firm solar power therefore needs real, hourly, multi-year weather data — a single "typical year" underestimates exactly the rare events that size the batteries.

We built the model on the National Solar Radiation Database (NSRDB), NREL's satellite-derived hourly irradiance record,^[12] accessed through the European Commission's PVGIS system,^[13] which converts irradiance, air temperature, and panel physics into hourly AC output for a utility-standard single-axis-tracking array (14% system losses). We use ten years of hourly data per site — 2005–2015, excluding 2012, which contains a documented 12-day satellite data gap that would masquerade as a two-week blackout. Four sites span the realistic siting conversation: West Texas (the reference case: elite sun, cheap land), Phoenix (the desert best case), Northern Virginia (where the data centers actually are), and Minneapolis (the cloudy north).

Figure 5 shows the seasonal spine of the problem. West Texas delivers a 23% annual capacity factor and holds 15.5% even in December — a June-to-December ratio of 1.9. Minneapolis delivers 24.5% in June and 6.7% in December — a ratio of 3.6. No plausible battery bridges a season; in the north, winter must be covered by overbuilding the array by that ratio, and the overbuilt capacity is surplus the other nine months of the year. Layered on the seasonal cycle are the weather runs: in our ten-year record every site shows multi-day stretches of near-zero output — 38 consecutive hours below one-fifth of average output in West Texas, and 94 hours (four days) in Minneapolis.

9.2 The model

The setup mirrors the real procurement problem an AI campus faces. A constant 1 kW load must be served every hour. We install S kilowatts of tracking solar and E kilowatt-hours of battery (88% round-trip efficiency); surplus solar charges the battery, deficits drain it, and any residual gap is covered by on-site diesel generation, which also carries the full load rating as backup capacity (with N+1 margin). For every candidate system we simulate all ten years hour by hour, then find the cheapest (S, E) pair that limits diesel to a target share of energy. Sweeping that target from 10% down to 0.01% traces the cost-versus-reliability frontiers of Figures 2 and 6. Costs are deliberately middle-of-the-road 2024–25 US utility-scale numbers (Table 2); the model and data are open in Forge's repository.

Table 2. Model cost assumptions — standard, middle-of-the-road industry numbers, in real 2025 dollars. None of the conclusions depend on pessimism: halve the battery price and the frontier keeps its shape, shifted down ~25%.

9.3 Results: the last nines are unaffordable

Every location shows the same signature: a gentle slope through 90–99%, then a wall. The wall moves with climate — Phoenix hits $322/MWh at 99.99% clean, West Texas $506, Northern Virginia $622, Minneapolis $789 — but it is always there, because it is made of the same physics everywhere: rare, long, dim stretches that only stored energy can cover, served by batteries that spend 350 days a year idle.

Figure 7 decomposes the West Texas frontier. At 90% clean the system is mostly solar and its cost is honest. By 99.9% the batteries dominate; by 99.99% the system is a battery farm with a solar accessory: 15 kW of panels and 101 kWh of storage per kilowatt of load — four days of batteries — plus the full diesel fleet still standing behind it. Scale that to a single gigawatt campus and it reads: 15 GW of panels covering on the order of a hundred square miles, and 101 GWh of batteries — more than all US grid storage installed through 2023 — for one campus, in the best solar geography, still burning diesel one hour in ten thousand.

And if a developer settles for 99% instead, the residual 1% is not a rounding error: for a 5 GW AI campus it is ~440 GWh of diesel generation a year — over thirty million gallons of fuel — through machines whose entire reason for existence was to avoid combustion.

9.4 What would change this picture

Real systems have levers our model omits: flexible training workloads that shed load in the worst weeks, continental transmission that averages weather across regions, wind blending against the winter trough. Each helps; none is free; and all of them are ways of paying different money for the same missing property — firmness. Transmission is a decades-long permitting project. Load flexibility trades away the utilization that justifies the silicon. Our model's honest conclusion is not that solar fails — it is that solar's economics are those of a fuel-saver, not a foundation. The foundation has to be firm, and firm is what fission is.

10How to solve climate change

Cheap firm energy does not just stop the problem from growing. Cheap enough, it runs the problem backward.

Today, removing carbon dioxide from the atmosphere is dismissed as fantastically expensive — and at today's prices, it is. Direct air capture consumes roughly 2 MWh per tonne of CO₂; at $50/MWh electricity and ~40 gigatonnes of annual emissions, offsetting the world's output would cost about $4 trillion a year in energy alone. Nobody will pay that. But every term in that product is movable:

Table 3. The arithmetic of running climate change backward. Halve emissions through electrification, quarter the capture energy through engineering maturity, and cut the energy price tenfold through factory-built fission — and offsetting every remaining tonne becomes a $50B/yr line item, smaller than the world's annual fossil fuel subsidies by an order of magnitude.

The $5/MWh figure is the radical one, and it is the one this paper is about. Forge's product roadmap walks the factory cost curve toward it deliberately: Ember, the first 25 MWe unit, targets energy at roughly $80/MWh — competitive firm clean power at first-of-a-fleet volumes. Its successors, produced at the rates of §3, target $30 and ultimately $10/MWh and below. These are targets, not promises — but they are targets with a mechanism: every halving of cost in Table 1's mature industries was bought with cumulative doublings of volume, and reactors built by the thousand get to buy the same curve.

This reframes the climate problem in a way sacrifice-based framings never could. If clean energy is expensive, climate policy is a permanent global argument about who absorbs the loss. If clean energy is the cheapest energy, the transition is simply what the market does next — and the residue of two centuries of emissions becomes a utility bill that a handful of wealthy nations could pay without noticing. Energy this cheap doesn't just decarbonize the grid. It desalinates water, synthesizes fuel, un-burns the atmosphere — and decouples human progress from environmental destruction, permanently.

11Conclusion

Factory-built water-cooled reactors sit at the intersection of four trends: an AI demand shock that has made firm clean power the scarcest commodity in the energy market; a proven light-water technology base with twenty thousand reactor-years behind it and a regulator built to license it; a manufacturing playbook — vertical integration, serial production, design-for-deployment — proven out in commercial launch by the generation of engineers now building nuclear; and a logistics network that can put a sealed reactor module anywhere a transformer can go.

Nothing in this paper requires new physics. The fuel is qualified, the coolant is water, the reactor is the most studied machine of the twentieth century. What is new is the industrial thesis: that the gap between $32/MWh operating nuclear and $15,000/kW construction projects is not a law of nature but an artifact of building cathedrals where we should have been building trucks. Solar showed what the factory cost curve does to a clean energy technology. Rockets showed the curve applies to safety-critical, heavily regulated hardware. Fission is next.

We aren't trying to reinvent nuclear power. Only how nuclear power gets built.