SMRs and Advanced Reactors: Is There a Survivor?

Executive summary — conclusion first

No listed small-modular-reactor developer yet clears the survivor bar; the name closest to a delivered, financed unit is not a pure-play at all (a pure-play being a company whose only business is small reactors) but an incumbent's product line, and what the listed developers actually own that is hard to replicate is the capital they raised, not the physics they designed.

The consensus is not wrong about demand. Firm, carbon-free power is scarce, data-centre load is growing faster than the grid can absorb it, and a reactor that could be built in a factory and financed like a gas turbine would be worth a great deal. The consensus is wrong about where in this business the money is, and about how much has actually been de-risked. Four listed developers and a fifth that reached the market in June 2026 carry a combined equity value of roughly $17.9 billion against combined trailing-quarter revenue of $44.0 million, essentially all of which is one company's cost-reimbursement work for the Department of Energy. Not one has sold a reactor, and not one has an unconditional, financed construction commitment at a disclosed price. Nuclearelectrica's shareholders adopted what the sponsor calls a final investment decision — the board decision that releases the money and starts the build — for the NuScale project at Doicești on 12 February 2026, but subject to mandatory conditions; in July the sponsor reported that those conditions had produced no concrete result sufficient to move into Pre-EPC, the stage before an engineering, procurement and construction contract is let. X-energy, for its part, says in its 23 April 2026 prospectus: "We have not yet delivered the Xe-100 or any other reactor to customers and have not achieved final investment decisions for the purchase or deployment of any of our reactors." The field has development capital, conditional decisions, power rights and options. It does not yet have priced, financed reactor backlog. Backlog is a word worth slowing down on, because the whole brief turns on it. It means work a customer is contractually bound to pay for — not a letter of intent, not a framework agreement, not a gigawatt figure in a press release. The test is whether the customer can walk away owing nothing. If they can, it is a pipeline, not a book of orders.

2026 has been a year of genuine milestones, and every one of them is smaller than it sounds. Four advanced reactors reached first criticality under the Department of Energy's Reactor Pilot Program between 4 June and 4 July. Criticality is the moment a reactor's chain reaction becomes self-sustaining — the physics switching on, and a real engineering achievement. It says nothing about how much heat the machine makes, and heat is what eventually gets sold. All four were zero-power tests. Idaho National Laboratory's director described the first of them plainly: "the chain reaction was sustained at essentially no measurable energy output. This is not electricity generation. It is not full-power operation." One unit later reached ten kilowatts thermal — about the heat of a domestic kettle bank, from a company whose commercial design is measured in megawatts. In the same window, the Nuclear Regulatory Commission granted Kairos Power a 28-month extension on the construction of a 35-megawatt-thermalA rating in heat rather than in electricity. A reactor's thermal rating (MWt) is the heat its core produces; the electrical rating (MWe) is the smaller amount of that heat a turbine converts into power. The two are not interchangeable, and a thermal figure always flatters a machine against an electrical one. test reactor it had already been permitted to build since December 2023. Those two facts belong in the same sentence, because together they describe the layer: physics arriving on schedule, construction arriving late, and economics not arriving at all.

Market capitalisations at the 31 July 2026 close. Revenue is each company's most recent reported quarter (Q1 2026 for OKLO, SMR, XE and FISN; the December–March quarter for NNE, whose fiscal year ends in September). Derivation and sources in the evidence register.

Three things follow, and they run against the way the layer is usually described.

First, the pure-plays' durable asset is their balance sheet. Oklo raised $1.18 billion in a single quarter through an at-the-market programme (a standing facility to sell new shares into the open market a slice at a time, at whatever the market is paying that day) at an average price of $95.50 a share; the stock closed at $38.83 on 31 July 2026. That is to Oklo's credit. The company converted a narrative into $2.54 billion of cash at a price the market will not offer again soon, and cash is the one input in this business that a competitor genuinely cannot copy. The reactor concepts can be copied. Dozens of companies are trying, and four of them took a core critical inside a single month. Two and a half billion dollars of equity raised at ninety-five dollars a share cannot be copied.

Second, licensing was never the binding constraint, so licensing reform does not do what the market thinks it does. The 2024 ADVANCE Act, the May 2025 executive orders, the Part 53A section of the US nuclear rulebook created as an optional alternative route to a licence. Instead of applying rules written around large light-water plants, it lets an applicant demonstrate safety in terms scaled to its own design and its own risks — the regulatory accommodation the advanced-reactor field had been asking for. framework finalised on 30 March 2026 and effective 29 April 2026, and the 550-page licensing overhaul the NRC proposed on 1 July and published on 16 July 2026 have all measurably compressed regulatory timelines — the NRC finished its Kemmerer safety review ahead of schedule, issued an environmental assessment for Dow's Seadrift project ahead of schedule, and granted TRISO-X a 40-year fuel licence in February 2026. Every one of those is real. None of them makes a reactor cheaper. The binding constraints are first-of-a-kind construction cost, fuel, and the willingness of a counterparty to sign a financed order, and the reform wave touches none of the three.

Third, the incumbents are further ahead than the pure-play narrative admits, and the reason is unglamorous. GE Vernova's BWRX-300 is a boiling-water reactorA reactor in which the water in the core boils directly and the steam it makes drives the turbine, with no separate steam-generating loop. It is a workhorse of the existing power fleet, which is precisely the point: nothing about it is new. — the least novel design in the field, deliberately so — and it is the only Western small modular reactor in the ground. Ontario Power Generation has excavated the shafts and has a budget its shareholder approved: C$6.1 billion for the first unit, C$1.6 billion of shared works, C$20.9 billion for four. That number is the most valuable disclosure in the entire sector, because it is the first honest Western price for a small modular reactor, and it is roughly US$18,700 per kilowatt for the first unit including common works — capital cost divided by the plant's power rating, the one measure that lets projects of different sizes be compared. Every marketed cost point in this industry has to be read against it.

The grades

Evidence: each company's own latest filing — OKLO 10-Q Q1 2026; SMR 10-Q Q1 2026 and 10-K FY2025; NNE 10-Q for the quarter ended 31 Mar 2026; XE 10-Q Q1 2026 and 424B4 of 23 Apr 2026; FISN 424B4 of 18 Jun 2026; GEV 10-Q Q2 2026 — plus NRC and DOE dockets cited in each company section. Method: technology readiness is graded on the licensing instrument actually held and the physical state of the lead project; economic viability on disclosed liquidity, quarterly cash use, implied runway and the quality of contracted backlog. Synthesis: a grade of "advantaged" requires both a unit being physically built and a balance sheet that does not depend on the next raise; only the incumbent clears both, and it clears them because the SMR is not what pays its bills.

GE Vernova is graded advantaged on a test its SMR programme does not, by itself, pass. The BWRX-300's own economics at Darlington are unproven, its first unit is expensive, and GE Vernova discloses no separate BWRX-300 revenue at all. The grade is earned on a different axis: GE Vernova is the only participant that can be wrong about small modular reactors and still be a profitable business, because its Nuclear Power unit turned over $1.58 billion in the first half of 2026 servicing the installed fleet. Optionality funded by an annuity is a structurally superior position to optionality funded by dilution. Three words are doing the work in that sentence. Optionality is the right to pursue something without the obligation to. An annuity is income that keeps arriving whether or not you pursue it. Dilution is paying for the attempt by issuing new shares, so every existing owner ends up with a smaller slice of whatever gets built. GE Vernova buys its option with the second. Everyone else buys it with the third. That is the whole finding, and it is uncomfortable for a sector that has been sold as a technology race.

Part I — The technology, compared — what each design is, what it needs, and how far it has got

A reader should finish Part I able to judge every Part II grade independently. The physics of fission and the global history of the industry live in the companion primer and are referenced here, not repeated.

1. Four design families, and why the differences are economic rather than aesthetic

Strip away the branding and the listed and near-listed field resolves into four families, distinguished by what carries the heat out of the core and what the fuel is made of. Those two choices determine almost everything that follows: the operating temperature, the safety case, the pressure boundary, the supply chain, the regulatory precedent, and — the point that matters here — whether anyone has built one before.

Light-water small modular reactors shrink the technology that runs the existing fleet. Water cools and moderatesSlows the neutrons thrown off by a splitting atom. Slow neutrons are far likelier to split the next uranium atom, so a moderator — water, in most reactors — is what keeps an ordinary chain reaction going. A design that leaves the neutrons fast (a fast reactor) gives up that help and must make it up with more concentrated fuel., the core sits under pressure, and the physics is the most thoroughly precedented in the world. NuScale's US460 puts six 77-megawatt-electric modules in a shared pool for 462 MWe gross. GE Vernova's BWRX-300 is a 300-MWe natural-circulation boiling-water reactor derived from a certified large design. Westinghouse's AP300 explicitly reuses the AP1000's licensed design basis and passive safetySafety that works without pumps, power, or an operator deciding to act — gravity, natural circulation and stored coolant doing the job instead. It is the safety case nearly every advanced design is built around, and the reason their marketing leans on physics rather than procedure. architecture. Holtec's SMR-300 is a two-unit pressurised-waterThe other mainstream reactor type. The core's water is held under enough pressure that it never boils; it carries heat to a separate loop where a second body of water boils and drives the turbine. The extra loop keeps the water that has been through the core away from the turbine hall, at the cost of more components to build and inspect. plant at the retired-then-restarted Palisades site. Deep Fission's Gravity Reactor is a pressurised-water core lowered a mile down a borehole, using the water column's hydrostatic pressure — about 160 atmospheres — instead of a steel containment vessel. The family's advantage is that regulators, fuel fabricators and welders all know what it is. Its disadvantage is that it inherits the cost structure that made large nuclear uncompetitive in the first place: a pressure boundary, a containment, and a fuel cycle at conventional low enrichmentRaising the share of the fissile isotope U-235 in uranium above the fraction nature supplies. Ordinary power-reactor fuel sits at the bottom of that scale. The higher the enrichment, the more energy a given volume of core can yield — and the tighter the security, handling and non-proliferation regime around the material..

High-temperature gas-cooled reactors use helium and a fuel form called TRISO — tri-structural isotropic particles, poppy-seed-sized uranium kernels wrapped in ceramic layers that act as an individual pressure vessel around every grain of fuel. X-energy's Xe-100 is the flagship: 80 MWe per reactor, optimised as a four-reactor 320-MWe plant, with helium outlet temperatures high enough to sell industrial steam rather than only electricity. Kairos Power's design substitutes molten fluoride salt for helium at similarly high temperature while keeping TRISO fuel. The safety case is the advantage, and it depends on neither operator action nor power: the fuel simply cannot melt at achievable temperatures. Process heatHeat sold as heat — a chemical plant's steam, say — rather than converted into electricity first. It is a second market for the same reactor, open only to designs that run hot enough, and it puts the seller in front of industrial buyers rather than the power grid. is a genuinely differentiated product on top of that. The problem is fuel. TRISO for these designs is fabricated from high-assay low-enriched uranium, and the West has almost none.

Sodium fast reactors abandon moderation entirely. TerraPower's Natrium is the utility-scale expression: a sodium-cooled fast core coupled to a molten-salt thermal store so the plant can swing output without changing reactor power, a genuinely clever answer to a grid full of intermittent renewables. Oklo's Aurora is the microreactor expression, 15 to 75 MWe and potentially higher, designed to be owned and operated by Oklo rather than sold. The family's advantage is fuel efficiency and the option, eventually, of burning used fuel — Oklo is explicit that its reactors are designed to run on fresh, recycled or down-blendedDiluted downward from a higher enrichment. Weapons-grade uranium mixed with natural or depleted uranium comes out at reactor assay; it is how material built for one purpose gets turned into fuel for another. A stockpile can be down-blended. It cannot be replenished that way. material, and that a recycling facility is a strategic goal for the early 2030s. Set against that, sodium is chemically violent with air and water, the operating history is thin, and the fuel is again high-assay.

Heat-pipe microreactors remove the coolant loop altogether. Sealed metal pipes carrying a working fluid wick heat from the core by capillary action, with no pumps and no primary circuit. Westinghouse's eVinci, Radiant's Kaleidos, Antares's Mark-0 and R1, and Aalo's Aalo-X all sit here, at one to five megawatts electric. This is where 2026's criticality milestones happened, and the reason is instructive: a one-megawatt heat-pipe unit with a few tens of kilograms of fuel is the cheapest possible way to demonstrate that a novel core works. It is also the furthest from a business. Westinghouse's own disclosure prices its commercial five-megawatt eVinci at a project value of $200 million to $500 million per unit.

The table makes two things visible that the prose usually hides. Every design in the field except the light-water family and Deep Fission depends on high-assay fuel that does not yet exist at commercial scale. And the designs closest to a delivered unit are, almost without exception, the least novel. That is no accident, and it is not peculiar to nuclear. It is what happens in every capital-goods industry where the first unit must be financed by someone who wants to be repaid.

2. The HALEU chokepoint — the number that decides which designs advance

Between 2019 and June 2026, the United States produced a cumulative total of just over 1.9 tonnes of high-assay low-enriched uranium. That is the whole of it — the entire output of the American Centrifuge Plant at Piketon, Ohio, under the DOE demonstration contract that Centrus Energy has held since 2019. Centrus completed the final 900 kilograms in mid-June 2026, two weeks ahead of schedule, and announced the achievement as a milestone. It is a milestone. It is also, against a Department of Energy statement that more than 40 tonnes will be needed by 2030, less than five per cent of the requirement, delivered over seven years.

The gap has been bridged, temporarily, by taking the material out of the weapons complex. Congress directed DOE under the FY2024 National Defense Authorization Act to make 21 tonnes of HALEU available on a schedule ending 30 June 2026, sourced by down-blending surplus highly enriched uranium held at the Y-12 National Security Complex, the Savannah River Site and Idaho National Laboratory. DOE has been allocating it in rounds: five developers in April 2025 (TRISO-X, TerraPower, Kairos Power, Radiant and Westinghouse), three more in August 2025 (Antares, Standard Nuclear, and Abilene Christian University with Natura Resources), and a third round on 23 July 2026 that added NASA and a second allocation for Radiant. DOE has not disclosed the quantities allocated to any individual company.

That is not a supply chain. It is a government inventory rationed by administrative decision, against selection criteria DOE has published — licensing status, fuel fabrication capability, technical maturity, and whether the requested form is available — and quantities it has not. When BWXT manufactured the TRISO compacts that took Antares's Mark-0 critical on 4 June 2026, it processed the HALEU feedstock "from scrap materials provided by NNSA." The first new reactor criticality under the Department of Energy's programme ran on National Nuclear Security Administration scrap. That is a remarkable achievement and a precarious foundation, at the same time.

The commercial answer is under construction and it is a long way off. On 1 July 2026 Centrus signed a firm fixed-price $900 million task order with DOE — $1.07 billion including options for up to $170 million of HALEU purchases — requiring it to deploy commercial-scale enrichment capacity and deliver one tonne of HALEU as uranium hexafluorideThe gaseous compound uranium is converted into so it can be enriched in a centrifuge, and the form it is shipped and traded in afterwards. Delivery as hexafluoride is delivery of enriched material, not of finished fuel — fabrication into pellets, pebbles or metal is a separate step, done by someone else. by March 2032. The initial build-out is 12 tonnes a year of HALEU capacity, plus capacity to work down a $2.4 billion low-enriched uranium backlog, with the first new capacity expected online in 2029. Centrus intends to run the existing demonstration cascade commercially in the interim.

So the fuel timeline for the whole high-assay half of the sector reads: government scrap through the late 2020s, a first tranche of new commercial capacity in 2029 at the earliest, and a contractual delivery obligation that does not bind until 2032. Set that against X-energy's stated first commercial delivery in the early 2030s and Oklo's target of a first powerhouse in 2028, and the fuel question stops being a footnote. It becomes the schedule.

The bottleneck test. If HALEU stays constrained, who is hurt and who is not? Deep Fission is the cleanest beneficiary of its own conservatism: its prospectus states plainly that the Gravity Reactor uses standard low-enriched uranium and conventional pressurised-water fuel assemblies, "more readily available than" HALEU, and it names this as a competitive strength. The light-water incumbents — BWRX-300, AP300, SMR-300, NuScale's US460 — are equally insulated. X-energy is the most interesting case: it is HALEU-dependent and it owns the fabrication step, holding NRC Special Nuclear Material Licence SNM-7007, issued 13 February 2026 for forty years, covering the TX-1 facility at Oak Ridge and, X-energy expects, a planned TX-2. That vertical position converts a dependency into a partial moatWhatever stops a competitor from copying what you do and competing the profit away — a licence that took years, a plant that cost a fortune, a contract nobody else can get. The test is not whether an advantage exists, but whether a rival with money could simply buy the same one. — TRISO-X already fabricates fuel for third parties, including space applications. But it does not solve enrichment. X-energy's own risk language is unambiguous: "In the near term, TRISO-X and its customers will depend on the U.S. government for access to HALEU given the current inability to access global markets." Owning the bakery does not help if the government owns the flour.

A regime in which the Department of Energy decides which designs receive scarce fuel is one in which the winner may be selected politically rather than economically. Every developer in the field understands this, which is why every one of them has a Washington strategy, a national-laboratory siting arrangement and a defence application. Those are rational responses to the actual competitive environment. They are not evidence of commercial viability, and the market has consistently priced them as though they were.

3. What a licence actually is — and what 2026's regulatory wave did and did not change

The vocabulary in this sector is treacherous, and the treachery is load-bearing. Four different instruments get reported as "approval," and they authorise radically different things.

1. 01
   Design approval

   The NRC finds a standard design acceptable, so it can be referenced later. Authorises no construction, names no site.
2. 02
   Construction permit

   A named licensee may build at a named site. Operating the plant needs a separate licence.
3. 03
   Operating licence

   Fuel may be loaded and the reactor run. This is the first instrument that permits selling power.
4. 04
   DOE authorisation

   A parallel federal pathway for reactors on DOE-controlled sites. Not an NRC licence and not transferable to a commercial site.

Evidence: 10 CFR Part 50 and Part 52 as applied in the Federal Register notices cited in this section; DOE Reactor Pilot Program description under Executive Order 14301. Method: instruments ordered by what they legally permit, not by the sequence any one company follows. Synthesis: the sector's reported "approvals" are overwhelmingly at steps 1 and 4; step 3 has not been reached by any advanced design in the United States.

As of 1 August 2026, NuScale holds the only standard design approval for a small modular reactor in the United States: the NRC issued it for the US460 on 29 May 2025. It authorises nothing to be built. TerraPower holds the most advanced instrument any advanced developer has — Construction Permit CPAR-1 for Kemmerer Unit 1, issued 9 March 2026, the first construction permit ever granted to a commercial advanced reactor — and construction began on 23 April 2026. Kairos holds construction permits for Hermes and Hermes 2, both test reactors. Dow's subsidiary Long Mott Energy holds a docketedFormally accepted onto the regulator's review calendar. It means the application was complete enough to be examined — the beginning of the review, not a judgement on it. Companies announce docketing; it is the cheapest good news in the sector. construction permit application for four Xe-100 units at Seadrift, Texas, for which the NRC issued an environmental assessment and finding of no significant impactThe regulator's conclusion, after a shorter environmental review, that a project does not warrant a full-length environmental impact statement. It removes a long step from the schedule. It decides nothing about safety, cost or whether the plant gets built. on 18 May 2026, along with exemptions from the requirement to prepare a full environmental impact statement. TVA's Clinch River BWRX-300 application received a favourable staff safety evaluation in June 2026 with a mandatory hearing set for 13 August 2026. Holtec filed the first part of a phased application for two SMR-300s at Palisades on 31 December 2025, docketed in February 2026, with Part 2 expected around mid-2027.

Oklo is the outlier, and it is routinely misdescribed. Oklo filed the first advanced-reactor combined licence applicationA single application seeking permission to build and to operate at once, rather than a construction permit first and an operating licence years later. Faster when it works, and expensive when it does not, because everything has to be settled up front rather than as the design matures. in the United States in March 2020. The NRC denied it without prejudice in 2022. Oklo's own Q1 2026 filing states it is "currently working toward submitting an updated custom combined license application," and that it completed a Phase I pre-application readiness assessment in July 2025 with no significant gaps identified. Its route to a first unit is therefore not the NRC at all but the DOE authorisation pathway: Oklo was selected in August 2025 for the Reactor Pilot Program, signed an Other Transaction AgreementA flexible federal contracting instrument that sits outside standard procurement rules, used to fund research and demonstration work on terms the agency and the company negotiate themselves. It is a funding and authorisation vehicle, not a licence, and it confers no commercial rights. on 25 March 2026, received DOE Idaho's approval of a Nuclear Safety Design Agreement for Aurora-INL on 17 March 2026, and broke ground in September 2025. Oklo's own framing is exact — it "plans to subsequently pursue NRC licensing to support commercial operations." Aurora-INL is a demonstration under federal authority. It is not a commercial plant and it does not become one by operating.

What the reform wave changed is real, and narrower than the headlines. The ADVANCE Act of July 2024 and Executive Orders 14300 and 14301 of May 2025 have driven three concrete outcomes: Part 53, an optional risk-informed technology-inclusive licensing framework responding to the Nuclear Energy Innovation and Modernization Act, was finalised on 30 March 2026 and took effect 29 April 2026; the NRC on 1 May 2026 proposed dedicated licensing requirements for "rapid licensing of new microreactors" and their "high-volume deployment"; and on 1 July 2026 it proposed a 550-page overhaul, "Modernizing Reactor Licensing, Safety Oversight, and Siting Practices" — published in the Federal Register on 16 July, framed explicitly as a response to Executive Order 14300 and open for comment to 31 August 2026 — which the agency projects could save the NRC and industry as much as $1.86 billion. The staff has also demonstrably accelerated: the Kemmerer safety review finished in December 2025 ahead of schedule and, by the applicant's account, eleven per cent under budget; the Seadrift environmental review finished early; and the NRC granted an exemption allowing an environmental assessment in place of a full impact statement for a commercial power reactor construction permit, which would have been unthinkable a decade ago.

None of that reduces the cost of concrete, forgings, skilled labour or interest during constructionThe financing cost that accrues while a plant is being built and earning nothing. On a long build it compounds, which is why a schedule slip and a cost overrun are the same event described two ways.. Regulatory delay was a genuine tax on the last nuclear build cycle and it is being cut. But the Vogtle overrun was not primarily a licensing overrun, and Darlington's C$20.9 billion is not a licensing number. The industry has spent a decade solving the problem it could measure. The problem that kills projects is still the one it cannot.

4. First-of-a-kind reality — announced against delivered, and the cost base rate

The historical base rate for first-of-a-kind nuclear in the West is serial overrun and, not infrequently, cancellation. First-of-a-kind is the industry's name for the first unit of a design ever built; nth-of-a-kind is what it is meant to cost once a production line has found its rhythm. Almost every economic claim in this sector is a claim about the distance between those two. NuScale's own flagship — the Carbon Free Power Project with Utah Associated Municipal Power Systems — was terminated in 2023 for want of subscribers, and NuScale still carries $5.1 million of restricted cash collateralising a letter of credit under the release agreement with that project company. The question for 2026 is whether anything in the current cycle has changed that base rate.

The dated record, as against the announcements.

Evidence: Federal Register notices 2026-09880, 2026-05067, 2025-10123, 2026-05086 and 2026-10073; Fluor 8-K of 23 April 2026; OPG on the Darlington budget; DOE on the Reactor Pilot Program. Method: only dated regulatory instruments, board decisions and disclosed transactions are included; announcements, memoranda of understanding and letters of intent are excluded by construction. Synthesis: the arc is real regulatory progress with no delivered commercial unit and one significant schedule reset.

The Hermes extension is the most informative single item, and it is barely reported. Kairos is arguably the best-executing private in the field: milestone-based DOE funding under a $629 million Technology Investment Agreement of which DOE pays $303 million and only on delivery, a Google master plant development agreement for 500 MWe by 2035, and the first US utility power purchase agreementA contract under which a buyer agrees to take a plant's electricity, usually for years and usually at an agreed price. It is what turns a power station from a hope into something a bank will lend against. A non-binding one is a letter, not a contract. for a Generation IV reactor, with TVA taking up to 50 MW from Hermes 2 to serve Google data centres in Tennessee and Alabama. And its 35-megawatt-thermal test reactor — a non-power research machine, permitted since December 2023, on a site it controls, with the money already committed — needed 28 more months. If the best-run private in the sector slips by more than two years on the simplest thing it will ever build, the schedule risk on a first commercial plant is not a tail case. It is the base case.

The cost benchmark that everything must be read against

Darlington is the live test. It is the only Western small modular reactor under construction, and unusually for this industry the sponsor has published a number its shareholder has approved rather than a target its marketing department has drafted. The arithmetic is unforgiving.

The first BWRX-300 at Darlington, including the shared works that only get built once, costs roughly US$18,700 per kilowatt. Even the unit scope alone is about US$14,800 per kilowatt. OPG projects the fourth unit at C$4.1 billion, "about 33% cheaper than the first" — a genuine and creditable learning curve, landing at roughly US$10,000 per kilowatt. A learning curve is the tendency for each successive unit to cost less than the last: a workforce that has done the job before, a supply chain already tooled for it, and mistakes that only have to be made once. It is a real and well-documented effect in manufacturing. It is also an effect, not a promise. And that is the optimistic end of the disclosed range, from the most-delivered design in the West, in a jurisdiction with an existing nuclear workforce, on a licensed site next to four operating reactors.

Westinghouse's disclosure, published through Cameco's second-quarter 2026 report, corroborates the range from an entirely independent direction. Its illustrative economics put an nth-of-a-kind two-unit AP300 project at $6–8 billion of overnight capital costWhat a plant would cost if it could be built instantaneously — all the equipment, materials and labour, with no financing charges piling up along the way. It is the standard basis for comparing projects, and it is systematically kinder than the number the owner actually writes cheques for. in 2026 dollars — roughly $9,100 to $12,100 per kilowatt on Westinghouse's own 330 MWe rating — assuming roughly 80% supply-chain commonality with the AP1000 and no schedule extension. Near-term deployments, it says, "have the potential to cost approximately 50% more." Westinghouse also estimates it will cost $0.8 billion to $1.2 billion simply to finish the design, with first commercial operations "potentially occurring in the mid- to late-2030s." This is the company that owns the only Generation III+ reactor that is fully designed, licensed and deployed, and its own guidance for a small modular derivative is a decade out and five-figure dollars per kilowatt.

The capital-recovery component alone can be computed exactly, and it is the number that decides whether any of this competes. Levelised cost is the price per unit of electricity a plant must earn, averaged over its whole life, to pay back everything it consumed: construction, fuel, staff, insurance, the eventual clean-up. The capital-recovery component is only the first of those, the slice that services the money spent building the thing before anyone turns a valve. If that slice alone exceeds what power sells for, nothing downstream can rescue the project.

At $6,000 per kilowatt — a figure no Western small reactor has come close to, and roughly Deep Fission's nth-of-a-kind target — the capital charge alone is $60 to $68 per megawatt-hourOne megawatt of power delivered for one hour — the unit electricity is bought and sold in. Prices per megawatt-hour are what a plant earns; costs per kilowatt are what it costs to build. The whole economic question in this brief is whether the second can be recovered out of the first. before a gram of fuel, an hour of operations, a dollar of insurance or a cent of decommissioning accrual. At Darlington's four-unit programme average the capital charge alone is roughly $135. At the high end of Westinghouse's own first-of-a-kind range it approaches $195. Deep Fission's business model assumes power purchase agreements at approximately $110 per megawatt-hour. Set the two side by side and the entire economic argument of the sector reduces to one question: can anybody actually get to the left-hand column?

The "factory fabrication" claim is repeated far more often than it is examined. The claim is that serial production in a controlled factory collapses cost the way it did for aircraft and gas turbines. It is a good argument in principle and it has one honest piece of evidence behind it — OPG's 33% unit-one-to-unit-four decline, which is exactly what a learning curve looks like. It also has a hard ceiling that is rarely acknowledged. Westinghouse says it could support "three to four AP300 SMR units per year" and "approximately three" eVinci units per year globally, expandable to "multiples of ten" only if the market materialises to justify the investment. Deep Fission targets a six-month installation for later units. Oklo intends to own and operate rather than sell. None of these is a Boeing production system. A learning curve requires repetition, repetition requires orders, and orders require someone to have signed.

Part II — Economics and the survivor test — who can afford to find out, and who cannot

5. Is there a survivor?

On the evidence disclosed as of 1 August 2026, no — and the honest form of the answer is that two names are financed long enough to attempt a first unit, one has bargained away the economics of the product it spent a decade licensing, and two are stories with a share price.

The survivor test has two axes and both must clear. Technology readiness asks whether the design is close to a delivered, financeable, fuel-secured unit — measured by the licensing instrument actually held, the physical state of the lead project, and whether the fuel exists. Economic viability asks whether the capital raised can earn a return before it must be raised again — measured by liquidity, cash consumption, implied runway, dilution and, decisively, the quality of the order book. These are capital-formationA business whose current activity is raising and spending money to build an asset, not selling anything. Judging one on revenue or margin is a category error; the questions that apply are how much it holds, how fast it goes out, and what physically exists at the end. businesses. A gross margin on a services contract tells you nothing. Runway, burn and the credibility of the nth-of-a-kind cost-down tell you everything. Burn is the cash a company consumes in a period net of anything it takes in; runway is how long the cash on hand lasts at that rate, counted here in quarters. Neither is a line in the accounts. Both are computed from them, and in a business with nothing yet to sell they are the closest thing to a scoreboard.

Grading the layer is hard because the two axes pull against each other. The companies with the most cash have the least proven technology, because raising a great deal of money against a novel design is precisely what a rising market rewards. The companies with the most delivered technology are inside larger firms that raised no money for it at all. That anticorrelation is the structural finding of this brief, and it is why the scorecard's upper-right quadrant is occupied by an incumbent rather than a pure-play.

"No survivor yet" does not mean the layer produces nothing, and it does not mean the designs fail. Kairos is building. TerraPower is building. GE Vernova's reactor is going into the ground at Darlington. X-energy holds a fuel licence that took years and will be difficult for a competitor to replicate. Oklo has assembled more capital than any advanced-reactor developer in history and has done so without debt. These are real accomplishments and the sector will deliver operating reactors in the 2030s. The question this brief asks is narrower and harder: which of these companies will earn a return on the capital it has consumed getting there? On that question the evidence supports one answer, and it is uncomfortable. As of today, nobody has demonstrated it, and the two names best positioned to try are best positioned because of what they raised, not what they built.

6. Three business models, and which one is actually financeable

The developer layer contains three distinct economic animals wearing the same fur, and the market prices them as though they were one.

Evidence: Cameco Q2 2026 report (AP300 design-completion estimate); Oklo 10-Q Q1 2026 (build-own-operate rationale and PPA model); X-energy 424B4 of 23 April 2026 (TRISO-X third-party fabrication). Method: classification by where the developer takes revenue and where it takes capital risk; a firm may sit in more than one. Synthesis: Model A externalises capital risk but also externalises the decision to proceed; Model B internalises both; Model C is the only one with an annuity that does not wait on a reactor order.

The instinctive ranking — licensor best, because it is capital-light — is exactly backwards in this cycle, and NuScale is the demonstration. A licensor's revenue is a call optionThe right to buy something later at a set price, with no obligation to. Its value depends entirely on a decision somebody else makes, and it can expire worth nothing while costing its holder real money to keep alive in the meantime. on a customer's final investment decision. NuScale has held the only US design approval since May 2025 and reported $565,000 of revenue in the first quarter of 2026, down 96% year on year, because the engineering work that produced the prior year's $13.4 million was completed. Being the licensing leader in a market where nobody has taken an unconditional final investment decision produces a design approval and no cash flow. That is not a temporary condition. It is the model working as designed.

Model B has the opposite property. A build-own-operate developer does not need a customer to reach a final investment decision, because it is the customer. Oklo can build Aurora-INL because DOE authorised it and Oklo is paying for it. That is why Oklo's model is worth more than its licensing position suggests, and it is also why Oklo's balance sheet is not a cushion but the product itself. Oklo's 2026 guidance is instructive: $80–100 million of cash operating expenses and $350–450 million of cash used in investing. The company is spending four to five times more on assets than on operations. That is a capital-projects business that happens to be listed as a technology stock.

Model C is the quiet one. X-energy generated $39.9 million of services revenue and $3.5 million of grant income in the first quarter of 2026, more than double the prior year, and the increase was driven almost entirely by a $21.6 million rise in ARDP activity with DOE. Its direct costs were $65.4 million against $43.4 million of total revenue and grant income — a gross deficit of $21.9 million. So the revenue is real, and it does not yet cover the cost of producing it. The structure underneath is what makes the model interesting: a forty-year fuel licence, a fabrication plant under construction since October 2024, and third-party fuel customers who are not X-energy. When the reactor business eventually earns or fails to earn, the fuel business will have a customer base of its own.

7. Company by company — the economics from each company's own filing

Every figure in this section is drawn from the named company's own most recent filing. Derivations are shown. Where a number is computed rather than disclosed it is marked.

Oklo (OKLO) — the best-capitalised developer in the history of the field

Technology readiness: mid Economic viability: strong Combined: neutral

Oklo held $2,536.9 million of cash, cash equivalents and marketable debt securities at 31 March 2026, with no debt. That is more liquidity than NuScale, Nano Nuclear and Deep Fission hold combined, and it was assembled with remarkable timing. In the first quarter of 2026 alone Oklo issued 12,376,352 shares through an at-the-market programme for $1,181.9 million net — an average of $95.50 per share against a 31 July 2026 closing price of $38.83. The shareholder who bought that paper is down badly. The company that sold it converted a narrative into a decade of runway, and did so without a covenant, a maturity, or a couponThe three things that make borrowed money dangerous. A covenant is a promise to the lender which, if broken, hands them control. A maturity is the date the principal must be repaid whatever else is happening. A coupon is the interest owed along the way. Money raised by selling shares carries none of them, which is why it cannot force a company into default..

The operating economics are disciplined. Net cash used in operating activities was $17.9 million in the quarter, against a reported operating loss of $51.2 million — the gap is mostly $15.6 million of stock-based compensation and $21.2 million of additional interest and dividend income on the enlarged cash pile. On operating cash burn alone Oklo has over a hundred quarters of runway, which is a meaningless number. The meaningful one uses the company's own guidance: $80–100 million of 2026 cash operating expenses and $350–450 million of cash used in investing. Taking the whole of the investing range as real consumption — the conservative reading, since some of it is securities purchases rather than plant — implies 4.6 to 5.9 years, roughly 18 to 24 quarters, at the 2026 run-rateToday's rate of spending or earning, projected forward as though it held steady. Useful for sizing a runway, and reliably wrong the moment a company starts doing something it has not done before — such as building a reactor.. Dilution over the year to 31 March 2026 was 24.9%, from 139.2 million shares to 173.9 million.

The technology position is where the grade is capped. Oklo has no NRC licence. Its 2020 combined licence application was denied without prejudice in 2022 and an updated application has not yet been submitted. Its route to a first unit is DOE authorisation for Aurora-INL, which is a demonstration on a federal site, not a commercial plant. It has not achieved criticality; it was selected for the Reactor Pilot Program and missed the 4 July 2026 deadline that four competitors met. Its fuel depends on the same government HALEU allocation as everyone else's, and its fabrication facility at INL is itself pending DOE authorisation.

The strongest execution evidence Oklo has produced arrived after the filing date. On 23 July 2026 Oklo received DOE startup authorisation for the Groves isotope test reactor near Lockhart, Texas — clearing fuel loading and startup testing ahead of first criticality — and the company describes the facility as "the fastest time that we are aware of to go from greenfield to substantial completion for a full-scale, privately funded and sited reactor in history," just over ten months from groundbreaking. That is a genuine and impressive result, and Groves is a low-power test reactor for isotope production, on private land, with commercially sourced fuel and equipment. It is not Aurora, it is not a power plant, and Oklo's own release mentions neither Aurora-INL nor commercial generation. What it demonstrates is build velocity under the DOE pathway, which is exactly the variable most in doubt across this layer.

The order book is the honest weak point, and Oklo's own filing describes it accurately. The company lists Meta, Equinix, Diamondback Energy, Prometheus Hyperscale and Switch as "prospective customers," and states its plan of operations includes "negotiating and executing additional letters of intent, memoranda of understanding, and master partnership agreements and converting such preliminary agreements into power purchase agreements" and "negotiating term sheets and binding power purchase agreements with customers who have previously signed nonbinding agreements such as letters of intent." That is a company telling its shareholders, in the regulated language of a quarterly report, that the pipeline is not yet a book of orders. The Meta arrangement announced in January 2026 for a 1.2 GW campus in Pike County, Ohio involves Meta prepaying for power and funding development — genuine capital and a genuine vote of confidence, and not a financed order for reactors. Oklo does have one unambiguously binding commercial contract: a procurement agreement with Siemens Energy for the power conversion system. It is buying, not selling.

Grade: neutral. Oklo is the best-financed attempt in the sector and the model that best matches the problem. It is graded neutral rather than advantaged because the survivor test requires technology readiness as well, and a company with no operating licence, no criticality and no disclosed-price binding PPA has not demonstrated it. Meta's development funding is the strongest counter-fact; it improves confidence in customer engagement but does not finance or authorise a plant. Falsifier: a first Aurora criticality followed by an executed, binding power purchase agreement at a disclosed price would move this to advantaged. A slip of Aurora-INL past 2028, or a second capital raise at a materially lower price before first power, would move it to exposed.

NuScale Power (SMR) — the licensing leader that gave away the economics

Technology readiness: mid Economic viability: weak Combined: exposed

This is the most consequential company-level finding in the brief, and it is hiding in Note 9 of a quarterly report.

On 27 August 2025 NuScale entered a Partnership Milestones Agreement with ENTRA1 Energy, its exclusive global commercialisation partner, running to at least 31 December 2045. Under the PMA, NuScale must pay ENTRA1 a "Milestone Contribution" for each NuScale Power Module anticipated to be placed in a project, in three tranches. Tranche one, 15% of the total, is payable "upon execution by ENTRA1... of a non-binding term sheet, memorandum of understanding, letter of intent or framework agreement with a third party." Tranche two, 35%, on a binding power purchase agreement. Tranche three, 50%, on a binding equipment order.

In 2025 ENTRA1 signed a non-binding agreement relating to 72 modules. NuScale recognised a one-time expense of $507.4 million, settling $247.5 million in 2025 and $259.9 million in the first quarter of 2026. Half a billion dollars, paid for a letter of intent.

At $47.0 million of partner contribution per module, on a module rated 77 MWe gross, NuScale carries roughly $610 per kilowatt of distribution cost before any reactor is manufactured. Darlington's entire fourth-unit target is about US$9,958 per kilowatt. NuScale has committed something on the order of six per cent of a competitive nth-of-a-kind capital cost to a commercial intermediary, on modules for which no binding order exists. The remaining tranches only trigger on binding agreements, which is the one genuinely mitigating feature: NuScale pays the other $2.9 billion only if real orders materialise. But the design of the agreement means the better the commercial outcome, the larger the payment, and the payment is denominated per module rather than as a share of a margin nobody has yet earned.

The balance sheet is adequate and the trajectory is not. Cash and equivalents were $341.1 million at 31 March 2026 with $549.0 million of short-term investments and $118.6 million of longer-dated investments — $1,008.8 million of total liquidity, no debt. Reported operating cash outflow for the quarter was $314.7 million, of which $259.9 million was the ENTRA1 settlement; the underlying rate is $54.8 million, implying about 18 quarters of runway. Revenue was $565,000. Full-year 2025 general and administrative expense was $609.8 million, of which $507.4 million was the single ENTRA1 payment; net loss for 2025 was $664.5 million. Total shares outstanding across both classes rose from 277.1 million at 31 December 2024 to 365.5 million at 30 April 2026, dilution of 31.9%. A $1 billion at-the-market programme opened on 26 February 2026, of which $962.1 million remained available at quarter end; the shares sold in the quarter went at a weighted average of $12.01 against $8.42 on 31 July.

One comparability trap has been widely misread. NuScale's weighted-average Class A share count rose from 127.7 million to 319.7 million year on year, which looks like 150% dilution. It is not. Most of that is Fluor's Class B units converting to Class A as Fluor sold. The correct measure is total shares across both classes, and it is 31.9%.

Fluor's exit is the most telling evidence of all. Fluor was NuScale's founding investor, its engineering partner, the contractor that performed the RoPower front-end engineering, and the source of 54% of NuScale's first-quarter 2025 revenue. It walked away entirely on 23 April 2026. It sold the final 40 million shares for $473 million gross, and its own 8-K records approximately $2.43 billion of open-market sales since September 2025 against a $570 million total investment. A four-times return is an excellent outcome for Fluor's shareholders. It also means the only large EPC contractor with deep NuScale-specific engineering knowledge has no remaining economic interest in whether a NuScale plant ever gets built. NuScale earned no revenue from Fluor in the first quarter of 2026.

NuScale's own 10-Q discloses a securities class actionA lawsuit brought on behalf of everyone who bought a company's shares over a stated period, alleging the company misled them. Filing one is cheap and proves nothing; what matters is whether the allegation survives the court's first look. filed 18 February 2026 in the District of Oregon — Truedson v. NuScale Power Corporation — naming the company, its chief executive, its chief financial officer and Fluor, and alleging false and misleading statements "relating to ENTRA1's experience, qualifications and capabilities as a developer of nuclear power plants." These are allegations, untested, and the company has not yet answered. But the existence of the claim is itself a disclosed fact from the company's own filing, and it goes directly to the durability of the commercialisation structure on which every NuScale forecast depends.

The RoPower project in Romania is NuScale's best asset. Nuclearelectrica's shareholders approved a final investment decision on 12 February 2026 for six 77-MWe modules at Doicești — 462 MWe, targeting commercial operation around 2030. That is genuine and it is the closest thing to a firm order in the pure-play field. It is also conditional on terms held in two annexes under an eight-year confidentiality agreement, which have not been published, and which the shareholder documentation states are "mandatory requirements that condition the feasibility of the project" — if any goes unmet, "the project will be deemed unfeasible." The most recent status weakens the label further: in a July 2026 exchange disclosure, Nuclearelectrica said discussions with the energy ministry had produced no concrete result satisfying the conditions needed to move into Pre-EPC. An investment decision whose mandatory conditions remain unmet and whose financing is unresolved is not the same instrument as an unconditional one, and it should not be graded as one.

Grade: exposed. NuScale holds the sector's strongest regulatory asset and has attached to it a commercial structure that transfers a very large share of the economics to a counterparty, lost its industrial sponsor, and is defending securities litigation about that same structure. Falsifier: conversion of the ENTRA1 relationship into a disclosed binding equipment order with a stated price per module, at a Milestone Contribution economics that leaves NuScale a defensible margin, would force a re-grade to neutral or better. Continued milestone payments without a binding order would confirm exposed.

X-energy (XE) — the only one with a real business underneath

Technology readiness: mid-high Economic viability: moderate Combined: neutral

X-energy listed on Nasdaq on 24 April 2026, pricing 44.25 million shares at $23.00 — four dollars above the top of the marketed range — and closing the transaction on 27 April with 50.9 million shares issued for approximately $1.1 billion net of $67.3 million of underwriting discounts. It closed the first day at $29.20. It closed 31 July 2026 at $16.31, 29.1% below the offering price and 44% below the first-day close. The capital markets have already re-rated the best-positioned pure-play in the field by nearly a third in fourteen weeks, and that repricing is itself evidence about the layer.

The pre-IPO balance sheet at 31 March 2026 showed $224.1 million of cash, $449.5 million of short-term investments and $270.4 million of long-term investments — $944.0 million — against $2,431.9 million of mezzanine preferredPreferred shares that sit between debt and ordinary equity in the queue for a company's assets, typically issued to private investors before a listing and converted into ordinary shares when it happens. Until they convert they are parked outside equity on the balance sheet, which is why a pre-IPO balance sheet can look nothing like the post-IPO one. that converted at listing. Pro formaAs if a transaction that has since happened had already happened on the date of the accounts. It is the honest way to describe a balance sheet overtaken by events. It is also, invariably, the version the company would rather you looked at. for the offering, liquidity is roughly $2.04 billion. Quarterly cash consumption was $67.3 million from operations plus $43.0 million of capital expenditure against $28.8 million of government grant reimbursement of that capex, a net $81.5 million. That implies about 25 quarters of runway at the current rate, with the explicit caveat that management states operating losses and negative cash flows "may increase from historical levels."

The technology position is the strongest in the pure-play field, and the reason is that the hardest thing X-energy has built is not a reactor. It is a fuel plant. TRISO-X received Special Nuclear Material Licence SNM-7007 on 13 February 2026 — a forty-year licence to manufacture HALEU fuel at Oak Ridge, expected to cover a second facility on the same site. TX-1 began construction in October 2024 and is expected to operate by the first half of 2028, with capacity for the first eleven Xe-100 reactors at steady state; TX-2 would support up to 44 reactors a year. That is a licensed, physical, capital-intensive asset in the one part of the chain that is genuinely scarce.

The reactor side is further behind and X-energy is candid about it. Dow's subsidiary Long Mott Energy filed the Seadrift construction permit application in March 2025; it was docketed in May 2025 for an 18-month review, and the NRC issued its environmental assessment and finding of no significant impact on 18 May 2026, ahead of schedule. X-energy expects the permit in the first quarter of 2027 and first commercial delivery in the early 2030s. The company's risk factors state that it "has not yet delivered the Xe-100 or any other reactor to customers and ha[s] not achieved final investment decisions for the purchase or deployment of any of our reactors."

X-energy describes "a more than 11 gigawatts electric, 144 reactor pipeline" across Dow, Amazon and Centrica — and qualifies it precisely: "assuming each customer exercises its contingent rights in full." Amazon's arrangement is a 2024 equity investment plus announced options for more than 5 GWe by 2039, whose first deployment is four reactors at Energy Northwest's Columbia site under a Carbon Free Development and Funding Agreement, with an upsize option to 960 MWe. Centrica's is a September 2025 joint development agreement for a possible six-gigawatt UK fleet at Hartlepool, first generation expected mid-2030s. The March 2026 Talen letter of intentA written statement that two parties would like to do business, setting out the shape of a deal neither is yet obliged to sign. It costs nothing to give and nothing to abandon. In this sector it is routinely reported as though it were an order. is explicitly non-binding; so is the IHI manufacturing memorandum. Every one of these is optionality granted to a customer, at the customer's election, funded meaningfully by the customer's equity. That is a good deal for X-energy relative to having nothing. It is not backlog.

The ARDP exposure is the sharpest single risk and the market rarely marks it.

X-energy was awarded $1.2 billion in December 2020 as one of two Advanced Reactor Demonstration Program awardees, structured as a 50/50 cost shareA federal funding arrangement in which the government reimburses an agreed fraction of approved spending and the company funds the rest. Money arrives only as eligible costs are incurred and accepted, so an award figure is a ceiling, not a cheque. of $2.4 billion of eligible costs through 2027. Its latest quarterly filing reports approximately $508 million reimbursed through 31 March 2026 — 42.3% of the award, up from $438 million at year-end. The current budget period runs from March 2025 to August 2026. Extensions beyond it "are subject to DOE discretion and approval," the total extension "may not exceed three years," and if X-energy cannot obtain extensions it "would forgo reimbursement for such costs and may face de-obligation of unobligated funds at closeout." About $692 million of the ceiling remains undrawn on a project whose first delivery is in the 2030s. The correction modestly improves the funding coverage; it does not remove the continuation risk. Meanwhile Dow will not take its own final investment decision until roughly 2028, and X-energy's filing notes that if Dow does not proceed, X-energy is under no obligation to keep funding the project but "would need to identify an alternative partner under the ARDP Agreement."

Grade: neutral. X-energy has the best combination in the pure-play field — real revenue, a licensed fuel plant, hyperscaler equity, and a customer permit under active NRC review — and it is graded neutral because it still has no order, negative gross economics, and a federal funding tail that turns on discretion. Falsifier: a granted Seadrift construction permit in the first quarter of 2027 plus an ARDP continuation into 2027–28 would push this toward advantaged. A denied continuation, or a Dow withdrawal, would push it to exposed quickly.

Nano Nuclear Energy (NNE) — a treasury in search of a programme

Technology readiness: low Economic viability: moderate Combined: exposed

Nano Nuclear is the field's most curious balance sheet. At 31 March 2026 — the second quarter of a fiscal year ending in September — it held $197.7 million of cash and equivalents (including $132.2 million of US Treasuries) plus $371.0 million of short-term Treasuries and $0.2 million of marketable securities: $568.9 million of liquidity against working capital of $565.7 million. It burned $9.3 million of operating cash in six months and spent $8.8 million on property, plant and equipment. On that combined rate the company has more than sixty quarters of runway. It has never earned revenue. Its accumulated deficit since inception is $73.2 million.

Sixty quarters of runway is not a compliment. It is the diagnostic. A developer with half a billion dollars and a $5 million quarterly burn is not executing a reactor programme; a reactor programme costs Oklo $350–450 million a year in investing alone and cost X-energy $43 million of capex in a single quarter. Nano raised approximately $378 million net in a private placement during the six months to March 2026 and diluted from 41.7 million shares at 30 September 2025 to 52.1 million at 31 March 2026, 24.8%, and to about 53.7 million by 31 July. The money arrived. The spending has not started.

The technology position is early. The lead project, KRONOS MMR, is a high-temperature gas-cooled microreactor whose intellectual property Nano acquired from the estate of Ultra Safe Nuclear Corporation in January 2025 — a company that went through bankruptcy. Its flagship deployment is a research demonstration at the University of Illinois Urbana-Champaign, for which the university is the licensing applicant to the NRC. ZEUS is a portable solid-core design and remains conceptual. Its fuel subsidiary, HALEU Energy Fuel Inc., is coordinating with DOE on supply and is not a producer. Nano did not appear in any of the three DOE HALEU allocation rounds identified in this research.

Nano's financial statements carry a "Liquidity" paragraph stating they are prepared on a going-concern basis and that continuation depends on securing financing, but management concludes that "sufficient working capital is available to meet the Company's liabilities and commitments as they become due at least for the next twelve months." That is standard development-stage language, not a going-concern qualification from its auditor. It should not be confused with Deep Fission's position, which is materially different.

Grade: exposed. The grade is on the technology-and-economics test, not on solvency: Nano is solvent for years, and that is precisely the point. It carries an $854 million market capitalisation for $569 million of Treasuries and a demonstration project at a university, with no licence of its own, no fuel allocation, no revenue and no disclosed customer order. Falsifier: an NRC construction permit for a Nano-owned reactor at a commercial site, a DOE HALEU allocation, or a step-change in capital deployment toward an actual build would move this to neutral.

Deep Fission (FISN) — an idea, a borehole, and a going-concern opinion

Technology readiness: very low Economic viability: very weak Combined: exposed

Deep Fission belongs in this brief because the mandate's own instruction is to verify listing status rather than assume it: a name filed under "defining privates" became a listed security on 18 June 2026, pricing 2.5 million shares at $16.00 for $40 million gross and approximately $32.9 million net. It closed 31 July 2026 at $9.41, 41% below the offering price, for a market capitalisation of about $568 million.

The concept is genuinely elegant, and it inverts the industry's cost problem. Instead of building a containment structure and a pressuriserThe vessel that keeps a pressurised-water reactor's coolant above boiling point by holding it under pressure. It is large, expensive and safety-critical — and it is one of the components a small reactor cannot shrink in proportion to its output., Deep Fission lowers a pressurised-water core roughly one mile down a 30-to-50-inch borehole and lets the water column supply about 160 atmospheres of operating pressure while the rock supplies confinement and shielding. It uses conventional low-enriched uranium and standard PWR fuel assemblies — deliberately avoiding HALEU, TRISO, sodium and molten salt — which is the most commercially literate design choice anyone in this brief has made. Management targets $152 million of capital cost for a first-of-a-kind 8 MWe unit ($19,000 per kilowatt) falling to $84 million for a 15 MWe nth-of-a-kind ($5,600 per kilowatt), against assumed power purchase agreements at approximately $110 per megawatt-hour and 93% uptime.

The finances are the problem, and they are stark. The auditor's report on the 2025 financial statements includes an explanatory paragraph indicating substantial doubt about the company's ability to continue as a going concern. Net losses were $5.7 million in 2024, $61.0 million in 2025 and $21.3 million in the first quarter of 2026 alone; accumulated deficit at 31 March 2026 was $88.1 million. Cash was $84.8 million at 31 March and approximately $64.3 million at 31 May — roughly $10.2 million a month over April and May. Add the $32.9 million of net IPO proceeds and the company had roughly $97 million in mid-June. At the first-quarter operating burn that is about seven quarters. At the April–May run-rate it is about three. The prospectus states directly that before the offering, management expected available cash "would not be sufficient to operate our business for the next twelve months."

Physically, the company has drilled one data-acquisition well to 6,000 feet. It has not drilled a commercial-scale borehole, built a reactor canister, or loaded fuel. It intends to apply for an NRC commercial licence in the first half of 2027.

Grade: exposed. A $568 million equity value on a $97 million treasury, a going-concern opinion, one test well and a conceptual reactor. The idea may well be the best in the field; the company is the least able to survive to prove it. Falsifier: a successful commercial-scale borehole and prototype emplacement with DOE authorisation, funded by a raise that does not halve the share count, would change the read materially.

8. The incumbents and the defining privates — the reason a pure-play may be beaten by a design it does not own

The competitive threat to the listed developers does not come mainly from each other. It comes from organisations for which a small modular reactor is a product line rather than an existence.

GE Vernova (GEV) — the BWRX-300. GE Vernova's second-quarter 2026 10-Q reports Nuclear Power business-unit revenue of $817 million for the quarter and $1,575 million for the half, against $1,310 million a year earlier, inside a group with $176.3 billion of remaining performance obligationsRevenue a company has contracted for but not yet delivered — the accounting profession's name for a backlog. Unlike a pipeline, it is work customers are already bound to pay for, which is why it is allowed to appear in the accounts at all. at 30 June 2026, up 37% year on year with nuclear cited among the drivers. Two comparability cautions apply and both matter: this is a newly realigned business unit created on 1 January 2026 out of the former Steam Power, and it is overwhelmingly a services and fuel business serving the operating boiling-water fleet. GE Vernova does not disclose BWRX-300 revenue separately, and nothing in this figure should be read as small-modular-reactor revenue. The structural point stands: GE Vernova funds its SMR programme out of an existing profitable business, while every pure-play funds its programme out of dilution. Meanwhile the BWRX-300 is the only Western SMR being built — shafts excavated at Darlington, first unit targeted in service by end-2030 — and NRC staff recommended a construction permit for TVA's Clinch River unit in June 2026 with the mandatory hearing set for 13 August 2026.

Westinghouse, through Cameco (CCJ). Cameco holds 49% of Westinghouse alongside Brookfield, and its second-quarter 2026 disclosure is the most economically informative document in this entire research file — and the most sobering. Westinghouse recorded a net loss of $10 million (Cameco's share) in the quarter against earnings of $126 million a year earlier, and a $56 million net loss over the first half against $64 million of earnings; Cameco's share of adjusted EBITDAEarnings before interest, tax, depreciation and amortisation, with further items management deems one-off stripped out. It measures operating performance while deliberately ignoring the cost of the capital employed — worth remembering in a business whose entire problem is the cost of the capital employed. fell to $163 million from $352 million, the prior-year period having benefited by roughly US$170 million from the Dukovany project in the Czech Republic. This is the company whose technology runs 57% of the world's 417 operating reactors, with a pipeline of up to 91 AP1000 opportunities, and it is loss-making at the net line. Its small modular products are earlier still: over 30 AP300 units "in origination," $0.8–1.2 billion to finish the design, first commercial operations mid-to-late 2030s, and an eVinci whose initial prototype is one megawatt with a commercial five-megawatt variant valued to Westinghouse at $0.2–0.5 billion per unit. Cameco labels all of it "illustrative estimates only and... not a forecast, target or guidance." That labelling is honest, and it is also the point: the most capable reactor vendor in the world will not forecast its own SMR economics.

Holtec (private) — SMR-300. Holtec's claim on credibility is unusual: it is restarting Palisades, the first US commercial reactor to return from decommissioning, and it took the plant to "operations" status on 25 August 2025 and received fresh fuel in October 2025, though restart has slipped past the original end-2025 target. Whatever the SMR-300 turns out to be worth, an organisation that has physically returned a reactor to operational status has demonstrated something no pure-play has. Its SMR-300 licensing is at the beginning: SMR, LLC filed Part 1 of a phased construction permit application — a limited work authorisation and comprehensive environmental report — on 31 December 2025, the NRC docketed it in February 2026, Holtec has asked for approval of Part 1 by 31 December 2026, and Part 2 is expected around mid-2027.

Rolls-Royce SMR (reference). Included for calibration on what a "lead" looks like outside the United States. It completed Step 2 of the UK Generic Design AssessmentThe British regulator's staged review of a reactor design in the abstract, before any site is chosen. Like a US design approval, it clears the design and authorises no construction., was selected by Great British Energy–Nuclear, and on 24 April 2026 signed an early-works contract with ČEZ for the Czech Republic's first SMR programme at Temelín, with memoranda for two further sites. Contracts to start site-specific design work are not orders for reactors either, but they are further along the commercial path than anything a US pure-play holds.

TerraPower — Natrium. The private that has gone furthest. It holds Construction Permit CPAR-1 for Kemmerer Unit 1 issued 9 March 2026, began construction on 23 April 2026, and is the only advanced reactor developer with a commercial construction permit in hand. It is supported by roughly $2 billion of DOE ARDP cost-share and has raised over $1 billion privately including an $830 million round led by Gates and SK. Meta's January 2026 arrangement funds development of up to eight Natrium units with energy rights, targeting first delivery as early as 2032. As a private company it discloses no cash, burn or runway, and this brief will not manufacture them.

Kairos Power — KP-FHR. The best-executing private on milestones and, on the evidence, the most honest about schedule. Construction permits for Hermes and Hermes 2; a $629 million milestone-based DOE agreement under which DOE's $303 million is paid only on demonstrated milestones and Kairos bears the upfront risk; a Google master plant development agreement for 500 MWe by 2035; and the first US utility PPA for a Generation IV reactor, with TVA taking up to 50 MW from Hermes 2 beginning around 2030. It also took a 28-month construction extension on Hermes in May 2026. Both facts are true about the same company and that is exactly why it is the most useful private in the comparison set.

The heat-pipe wave. Radiant, Aalo, Antares, Valar, Deployable Energy, Last Energy. This cohort produced 2026's headlines and deserves neither the enthusiasm nor the dismissal it typically gets. Antares reached zero-power criticality on 4 June 2026 at INL's RACE facility on BWXT-manufactured TRISO with under 120 kg of fuel; Valar followed on 18 June, later reaching 10 kWt; Deployable Energy on 1 July; Aalo-X in the early hours of 4 July. What they demonstrated is that a novel core can be built and taken critical in months under DOE authority rather than years under NRC licensing. That is a genuine and important institutional result. What they did not demonstrate is power generation, grid connection, cost, or a customer. Radiant's Kaleidos, Aalo, Last Energy, Deep Fission, Natura and Terrestrial did not reach criticality by the deadline at all.

Evidence: Centrus press release 1 July 2026; BWXT release 4 June 2026; Oklo 10-Q Q1 2026 (Siemens Energy contract, Kiewit constructor); GEV 10-Q Q2 2026; Cameco Q2 2026 report; Fluor 8-K 23 April 2026; Deep Fission 424B4 18 June 2026. Method: sorted by whether the party is paid in cash today or holds a claim contingent on a future reactor. Synthesis: the profit in the developer layer as of mid-2026 has already been earned — by the suppliers the developers pay and the investors who sold to them.

Incentive mapping — what each participant is actually optimising for

Grades follow from incentives, and the incentives in this layer are unusually legible because almost every participant has written them into a contract or a programme document. Read together, they explain why the developer layer is where the story is priced and the profit is not.

9. The Survivor Scorecard

The scorecard places every developer on the two axes the survivor test uses. Neither axis is a market judgement and neither is a score of quality. Technology readiness is anchored on what a company legally holds and what physically exists: a construction permit and steel in the ground outrank a design approval, which outranks a pre-application, which outranks a concept. Economic viability is anchored on liquidity against consumption, the quality of the order book, and whether the next dollar of capital has to be raised at a price the market sets.

Three features of that plot carry the argument. The upper-right quadrant — ready and viable — contains no pure-play. It contains a product line inside a $176 billion-backlog industrial company. Oklo and X-energy sit on the same rough diagonal from opposite directions: Oklo strong on money and weak on licence, X-energy stronger on licence and fuel but weaker on the durability of its funding. And the lower right is empty — no company in this field has advanced technology and weak funding, because in this cycle the capital arrived before the technology did. That is the signature of a market financing a narrative, and it is not, by itself, a criticism. It is how frontier industries get built. It is also how they get overbuilt.

The bottleneck test is the discipline that separates a position from a moat. For each developer the question is the same: what does the position depend on, can a competitor replicate or overtake it, and what is left if the binding constraint bites? Two answers stand out. Oklo's position survives most of the constraints because its binding one is time, and time is what $2.54 billion buys. NuScale's survives the fewest, because its binding constraint is contractual rather than physical, and a contract cannot be engineered around.

Runway is where the layer separates most cleanly, and the ranking is nearly the inverse of technological progress. Nano Nuclear, which has built the least, has the most time. Deep Fission, whose design is arguably the most commercially thoughtful, has the least. That is the mechanism at work. Companies that raise capital before they spend it look safe on a runway chart precisely because they have not started. A runway figure is a measure of patience, not of progress, and it should be read alongside the technology axis rather than instead of it.

Strip out the cash, and what is the market paying for the business? X-energy carries about $4.6 billion of value above its pro forma net liquidity, Oklo about $4.2 billion, NuScale about $2.1 billion. For X-energy, that premium buys a forty-year fuel licence, a fuel plant under construction, real if unprofitable revenue and a customer permit under review. For Oklo, it buys a DOE-authorised demonstration, a business model that does not need a customer's investment decision, and a mix of funded development, master agreements and letters of intent — none yet a financed build. For NuScale, it buys a design approval, a conditional Romanian project, and a contractual obligation to pay a partner roughly $2.9 billion more if the orders it is waiting for actually arrive. Those are three very different four-billion-dollar propositions, and the market has been pricing them as one asset class.

The single-name reads

Structural reads only. No price targets, entry or exit levels, or position sizing. Metrics from each company's own latest filing as cited in section 7; GE Vernova figures are group and business-unit level and are not BWRX-300 revenue, which the company does not separately disclose.

10. The bear case, steelmanned — then answered

"These companies are overvalued" is not the bear case; anyone can say that. The strongest version is a coherent account of why the entire layer delivers essentially nothing this decade, and it goes like this.

Nuclear's cost problem was never regulatory; it was physical and organisational, and small reactors make it worse, not better. A reactor's capital cost scales sublinearly with output — the pressure vessel, the containment, the control room, the security force and the licensing burden do not shrink in proportion to megawatts. That is why the industry built ever-larger units for fifty years. Shrinking the unit therefore raises cost per kilowatt, and the entire SMR thesis is a bet that factory serialisation more than reverses this. Nobody has demonstrated that bet at any scale. The one Western data point — Darlington — puts the first 300-megawatt unit at roughly US$18,700 per kilowatt including common works. Vogtle Units 3 and 4, the most expensive nuclear project in modern American history, delivered about 2,228 megawatts for what Georgia Power estimated at more than $30 billion including financing: above US$13,000 per kilowatt. The small reactor is starting out more expensive per kilowatt than the disaster it was designed to avoid.

The order book is a mirage. Not one listed developer has an unconditional, financed construction commitment at a disclosed price. X-energy says it has no reactor FID. Oklo's quarterly report describes its commercial task as converting preliminary agreements into PPAs, even as Meta supplies real early-development funding. NuScale's Romanian sponsor adopted a conditional FID whose mandatory conditions had produced no concrete result by July. The gigawatt figures in the press — 12 GW with Switch, more than 11 GWe across X-energy's customers, 5 GWe of Amazon options, and Meta's agreements across several developers — combine options, development funding, prepayment mechanisms and conditional rights. Those instruments are not worthless; they de-risk development for the developer and buy optionality for the customer. They are worthless as firm financed backlog, which is the category the headline gigawatts invite investors to infer.

The fuel does not exist and cannot be conjured. Seven years of American HALEU production totals 1.9 tonnes against a stated 2030 requirement above 40. The bridge is a finite drawdown of the weapons stockpile. New commercial capacity is not expected before 2029 and the binding contractual delivery is 2032. Every HALEU-dependent design is therefore hostage to an administrative allocation, and the first design to be denied fuel will discover that its licence is worthless.

The capital market is the actual reactor coolant, and it is losing pressure. X-energy is 29% below its April IPO price. Deep Fission is 41% below its June IPO price. NuScale sold stock through its own at-the-market programme at a weighted average of $12.01 in the first quarter and closed at $8.42 on 31 July. Oklo is at $38.83 against the $95.50 it sold stock at in the same quarter. Every one of these companies must raise again before it earns anything, and each raise at a lower price transfers more of the eventual upside to whoever provides the last dollar. Meanwhile Vogtle came in at more than double its original estimate on more than double its original schedule, the best-run private in the sector just slipped 28 months on a test reactor, and the demand thesis rests on a data-centre buildout whose own national-laboratory projection carries a band of 9.5% to 15.3% of all US electricity by 2030 — a spread wide enough to be closer to a confession than a forecast.

That is the bear case at full strength. Three parts of it survive scrutiny, and one does not.

What survives. The cost argument survives almost intact. The disclosed evidence in this brief — Darlington, Westinghouse's own AP300 estimate, the capital-recovery sensitivity — supports the bear's central claim that first-of-a-kind small reactors are more expensive per kilowatt than large ones and that nobody has yet shown the learning curve that would fix it. The order-book argument survives entirely; it is not contested here because the companies themselves do not contest it. The fuel argument survives on the numbers, though it is a timing constraint rather than a permanent one: enrichment capacity is a solved engineering problem that requires only money and years, both of which are being applied.

What does not. The bear's implicit conclusion — that the layer therefore delivers nothing and the companies are worthless — does not follow from its own premises, for a reason the bear case systematically underweights. It treats these as revenue-generating businesses being valued on absent revenue, when the two best-funded are better understood as capital pools with call options attached. Oklo consumed $17.9 million of operating cash in a quarter while holding $2.54 billion. It can be wrong about its schedule for five years without needing the market's permission. X-energy is pro forma at roughly $2.0 billion with 25 quarters of runway and a licensed fuel plant that has value independent of whether the Xe-100 is ever built. A closing capital market is fatal to Deep Fission and irrelevant to Oklo for half a decade. The bear case is right about the industry and wrong to apply that verdict uniformly to the balance sheets inside it.

There is also a factual correction the bear should accept. The claim that regulatory reform is cosmetic is not supported by 2026's record: the NRC completed the Kemmerer safety review ahead of schedule, issued the Seadrift environmental finding ahead of schedule, granted an exemption to substitute an environmental assessment for a full impact statement on a commercial power reactor, issued the first commercial advanced-reactor construction permit in its history, and licensed a HALEU fuel plant for forty years. Regulatory velocity has genuinely changed. It simply changes a variable that was not the binding one, which is a different criticism and a more interesting one.

11. Second-order effects — where the money has actually gone

Follow the cash rather than the narrative and the developer layer looks different. Between them, the five listed developers have raised several billion dollars of equity in eighteen months, and that money has not stayed in their hands. It has flowed outward, in cash, to a set of counterparties who are being paid today for work on plants that may never operate.

1. 1The suppliers have already won. Centrus has a $900 million federal contract and a $2.4 billion LEU backlog. BWXT made the TRISO fuel and processed the feedstock for the first new reactor criticality under the DOE programme. Curtiss-Wright is X-energy's selected supplier for multiple Xe-100 systems, Siemens Energy holds a binding power-conversion contract with Oklo, and Kiewit is lead constructor at Aurora-INL. Each is paid in cash out of a developer's equity raise, on delivery, regardless of whose reactor wins. The profit in the developer layer as of mid-2026 has been earned — by the layer's vendors.
2. 2The government has become the allocator, which makes the winner partly a political choice. DOE decides who gets HALEU, whose reactor is authorised under the Pilot Program, and whether X-energy's remaining $692 million of ARDP funding is extended. The rational corporate response is a Washington strategy, a national-laboratory site and a defence application — which is precisely what every developer in this brief has built. The market has repeatedly priced these as commercial validation. They are political positioning, and they are reversible with an administration.
3. 3When the federal balance sheet finally opened, it went to large reactors, not small ones. On 23 June 2026 the Department of Energy announced $17.5 billion of American Nuclear Supply Chain Loans through its Office of Energy Dominance Financing — up to five projects, each jointly owned by Westinghouse and a utility partner, financing long-lead components for ten large AP1000 units, with the stated aim of shortening deployment by up to three years. Not a dollar of it is directed at a small modular design. Nor is it an isolated choice: DOE's own one-year review of the executive orders lists a $1.52 billion loan to Holtec for the Palisades restart, a $1 billion loan to Constellation for the Crane restart, and $26.5 billion of loans to Southern Company — restarts and large light-water reactors, every one. The largest acts of federal nuclear financing in this cycle were bets on the licensed large reactor, placed by the same government that is rationing HALEU to the small ones. That is a revealed preferenceWhat someone's choices show they actually want, as against what they say they want. It is economics' standard corrective to stated intentions: watch where the money goes., and it is not the one the sector narrative implies.
4. 4The hyperscalers have moved the capital risk onto the developers and kept the option. Amazon's 5 GWe, Google's 500 MWe, Meta's 1.2 GW at Pike County, Switch's 12 GW: contingent rights, development funding and prepayments, not obligations to buy reactors. If costs land at two-to-four times target, the developer eats it or the project dies; the hyperscaler simply buys gas. That asymmetry is the single most important commercial fact in the layer and it is almost never stated.
5. 5Two regulatory regimes now exist, and the arbitrage is real but fragile. Four reactors reached criticality in 2026 under DOE authorisation on federal sites, on timelines the NRC could not have matched. That pathway is a genuine national asset and it produces demonstrations, not commercial plants. A single serious incident at a DOE-authorised unit would reprice the whole layer and would advantage precisely the companies that took the slower NRC route.
6. 6Consolidation is the most likely use of the balance sheets, and Oklo has said so. Its own plan of operations includes "evaluating potential acquisition opportunities to strategically accelerate our business." A crowded field of developers, several with excellent engineering and a year of cash, sits next to two companies holding roughly $4.6 billion of liquidity between them. The clearing mechanism for this cycle is not bankruptcy. It is acquisition at distressed prices by the companies that sold stock at the top.

Evidence: Centrus release 1 Jul 2026; BWXT release 4 Jun 2026; Oklo 10-Q Q1 2026 (Siemens, Kiewit, acquisitions); X-energy 424B4 23 Apr 2026 (Curtiss-Wright, Amazon, ARDP); Kairos/Google/TVA announcements; ANS coverage of the Reactor Pilot Program. Method: effects traced from disclosed cash movements and contractual structure, not from market commentary. Synthesis: value in this layer is currently being transferred from equity holders to suppliers, governments and customers — which is what a pre-revenue capital-formation phase looks like from the inside.

The non-consensus conclusions

Five claims that a well-informed reader of the sector press would not already hold, each tied to disclosed evidence and each falsifiable.

1. 1The durable asset is the balance sheet, not the design. Consensus revised: that these are technology companies whose value is their reactor IP. Sixty-plus companies have reactor concepts and four of them took a core critical in a single month. What almost none of them has is $2.5 billion raised at $95.50 a share with no debt. Estimated Falsified if a developer with under $200 million reaches a financed commercial order before a developer with over $2 billion does.
2. 2NuScale's licensing lead is a liability, not an asset, under its current commercial structure. Consensus revised: that the only US design approval makes NuScale the safest pure-play. A $507 million payment on a non-binding agreement, roughly $2.9 billion of contingent partner contributions, about $610 per kilowatt of channel cost, a fully exited EPC sponsor and pending securities litigation about that structure describe an asset whose value accrues substantially to someone else. Known Known Falsified if NuScale discloses a binding module order whose price leaves it a defensible margin after Milestone Contributions.
3. 3Licensing reform is real and almost irrelevant to the outcome. Consensus revised: that the ADVANCE Act and the 2026 rule wave meaningfully de-risk the layer. Every 2026 acceleration is genuine, and the binding constraints — first-of-a-kind construction cost, fuel supply, and a counterparty willing to sign a financed order — are untouched by all of it. Estimated Falsified if a US advanced reactor reaches commercial operation within four years of docketing at a disclosed cost below $10,000 per kilowatt.
4. 4The hyperscaler agreements transferred risk to the developers rather than validating them. Consensus revised: that Google, Amazon and Meta contracts prove commercial demand. Each is structured as contingent rights, options or development funding; X-energy states it has achieved no final investment decision on any reactor, and Oklo describes its task as converting non-binding agreements into power purchase agreements. Known Known Falsified if a hyperscaler discloses a firm take-or-payA contract under which the buyer pays for the agreed output whether or not it takes delivery. It is the strongest commitment a customer can give a plant, because it removes demand risk entirely — and it is the form of commitment nobody in this brief has yet given. obligation for advanced-reactor output at a stated price.
5. 5The HALEU regime hands the Department of Energy a de facto veto over which designs advance — the most under-priced risk in the layer. Consensus revised: that HALEU is a supply problem Centrus is solving. Total US output since 2019 is 1.9 tonnes; the 2026 criticalities ran on NNSA scrap; DOE allocates by undisclosed quantity against published criteria; commercial capacity is a 2029–2032 story. Estimated Falsified if a commercial HALEU market with published pricing and non-government supply emerges before 2030.

12. Falsifiers, and dated predictions

Conviction is worth nothing unless it is stated in a form that can be shown to be wrong. Each prediction below has a date, a testable outcome, and a confidence sized to the evidence rather than to the strength of the thesis.

Evidence: the dated disclosures cited throughout Parts I and II. Method: each prediction is tied to a specific, publicly observable disclosure event so that it can be scored without interpretation. Synthesis: the set is deliberately weighted toward outcomes that would embarrass this brief if wrong — a prediction nobody can lose is not a prediction.

What to watch, and when

Evidence: company announcements and NRC/DOE dockets cited in the evidence register. Method: only dated, externally observable events; company targets are labelled as such. Synthesis: the sector's first genuinely economic datapoint — a delivered unit at a disclosed cost — does not arrive before 2030.

Nobody knows what a tenth Aurora, a twentieth Xe-100 or a fiftieth BWRX-300 costs, because nobody has built the first. The learning curve is not a fantasy; OPG's own 33% decline from unit one to unit four is evidence that it exists. The honest position is that this industry is at the point where the aircraft industry stood before the DC-3 and the gas-turbine industry stood before the F-classA family of large industrial gas turbines, invoked here as the moment an engineering promise became a catalogue item — a machine you order rather than a project you undertake.: the engineering is credible, the cost is not yet, and the difference between those two states has historically been resolved by somebody building twenty of the thing. What can be said with confidence is who is positioned to still be standing when that happens, and who is being paid along the way. On the evidence disclosed through 1 August 2026, the answer to the first question is two companies and an incumbent, and the answer to the second is almost everyone except the shareholders.

Evidence register — dated, tiered, every source opened and read this session

Tier 1 = issuer filings, regulator publications, primary programme documents. Tier 2 = reputable trade, wire and specialist press. Tier 3 = single-source or aggregated data, used only where flagged.

Not verified / not load-bearing — items that carry no grade in this brief

• An SEC enforcement investigation into NuScale. Secondary reporting indicates the SEC's Division of Enforcement described an investigation as "active and ongoing" in a 24 July 2026 response to a records request. The underlying document could not be opened this session, and NuScale's own most recent filing (10-Q, 7 May 2026) discloses no SEC investigation. This item carries no part of the NuScale grade, which rests entirely on the company's own disclosed financial and contractual facts.
• The Guggenheim characterisation of ENTRA1 ("three employees and one investor") circulating through litigation press releases could not be traced to the underlying research note this session. The NuScale analysis rests on the company's own Note 9 disclosure and its own disclosure of the Truedson complaint, not on this characterisation.
• Per-design HALEU core loadings in tonnes. The Idaho National Laboratory report on HALEU requirements for net-zero could not be parsed this session, and no company disclosed a per-core tonnage. The HALEU section therefore quantifies national supply and stated national requirement only, and does not assert how many reactors 21 tonnes would fuel.
• Private-company balance sheets. NRC, DOE and counterparty records establish TerraPower and Kairos milestones, but neither developer publishes filing-grade cash, burn, debt or dilution. Technology stages are verifiable while economic runway is not; no midpoint score is inferred.
• Oklo's first criticality. Oklo received DOE startup authorisation for the Groves isotope test reactor on 23 July 2026, which clears fuel loading and startup testing; no confirmation that criticality has been achieved at Groves, or at Aurora-INL, was found in this session's searches. The Oklo grade assumes none has occurred and would improve if one has. Groves is in any case a low-power isotope test reactor and not the Aurora powerhouse on which the commercial thesis rests.
• TerraPower and Kairos financial condition. Both are private and disclose no cash, burn or runway. Their characterisations rest on regulatory instruments, disclosed DOE agreements and announced funding only; neither is graded on the economic-viability axis, and neither appears on the scorecard.
• Nano Nuclear HALEU allocation. No DOE allocation to Nano Nuclear or its HALEU Energy Fuel subsidiary was found in the three allocation rounds identified. Absence of evidence in this search is not evidence of absence; the point is stated as "not found," and the Nano grade does not turn on it.
• Reported selection of Oklo and X-energy for a federal AI-datacentre nuclear initiative (roughly $200m, reported via wire coverage in July 2026) could not be confirmed against a primary programme document and carries no weight in any grade.