Commercial Fusion

Fusion power means forcing light nuclei together, capturing the energy released, and selling the resulting electricity at a price a buyer accepts. Those are three achievements and the field has completed one of them. Keeping them apart is the whole analytical discipline the subject requires, because the word breakthrough is applied to all three interchangeably and the distances between them are not comparable.

Three thresholds exist. Scientific gain — fusion energy out against energy delivered into the plasma or the target — has been crossed, repeatedly, in two confinement schemes. Engineering gain — electricity out against electricity in, counting the cryoplant, the magnets, the heating systems, the drivers and the pumps — has never been demonstrated anywhere, by anyone, at any scale. Commercial gain is not yet a research question, because nobody has reached the rung below it.

The framing under test is that fusion has become an engineering and financing problem rather than a physics one. That is close to right and it omits a third category. Magnets, cryoplants, vacuum vessels, power conversion and the assembly of a burning-plasma machine are engineering, and private teams are executing them faster than the public programme did. But the fuel is not an engineering problem and it is not a financing problem. There is no amount of money that buys tritium at fleet scale, because tritium is a decaying by-product of a shrinking fleet of heavy-water reactors; no operating facility on Earth produces the neutron spectrum against which a first wall must be qualified; and the comparator a 2040s plant has to beat is what the grid will cost in the 2040s, not what a gas plant costs now. Fusion has stopped being a plasma-physics problem and has become a fuel, materials-qualification and supply-chain problem. That is a real change in the subject's character and it is not the change the framing describes.

Established Ignition is real, replicated, and its headline number excludes the plant. The National Ignition Facility achieved target gain above unity on 5 December 2022 — 3.15 MJ of fusion yield from 2.05 MJ of laser energy delivered to the target, a target gain of about 1.54. The best shot on the public record is 7 April 2025: 8.6 MJ from 2.08 MJ delivered, a target gain above four, enabled by a continuous-gradient-doped capsule, with the laboratory reporting ignition achieved nine more times after the first. Established The word “delivered” is doing the work. NIF's laser system draws roughly 300 MJ of electricity to put 2 MJ on target — about 150 times what reaches the capsule. The April 2025 record therefore returned 8.6 MJ against roughly 300 MJ drawn: a wall-plug ratio near 2.9%. Established Magnetic confinement carries the same distinction: JET's final tritium campaign set a record of 69.26 MJ over about six seconds from 0.21 mg of fuel in 2023–24, against 59 MJ in 2022 and 22.7 MJ in 1997. Those are fusion-energy records. None of them is a net-electricity result.

Established NIF is not a power-plant prototype, and its own leadership says so in the shot rate. The facility's director framed the forward question as whether the laboratory “can begin to do it more than one time a day”. An inertial fusion plant needs several shots per second against a mass-produced target. Frontier That is not a physics gap; it is a manufacturing gap of roughly five orders of magnitude in repetition rate, and no funded programme has demonstrated the target factory it implies.

Established The largest public machine slipped a decade and stopped quoting a total cost. ITER's July 2024 baseline moved start of research operations to 2034, deuterium–deuterium operation to 2035, full magnetic energy to 2036, and the deuterium–tritium phase to 2039. Against the previously advertised 2025 first plasma that is roughly a nine-year slip on the first milestone and four years on D-T, attributed by the organisation to pandemic supply shocks, first-of-a-kind manufacturing, component repair, and planning that “proved to be too optimistic”. Established Added cost is reported at about €5 billion, and the more telling fact is what was not reported: the Director-General declined to give a total project cost, on the ground that the in-kind contribution structure makes one meaningless. The 2001 redesign was costed near €5 billion and the 2010 estimate near €15 billion; nothing comparable is on the current public record. Frontier A project whose members cannot state its cost is not a project whose schedule can be independently audited, and this is the largest fusion programme in existence.

Frontier The defence offered by ITER's own constituency concedes the point while making it. The Max Planck plasma-physics director called the delay “less than we feared” and argued that “we are not aware of any project that will analyse the challenges as comprehensively as ITER in the foreseeable future”. That is an interested party whose researchers use the machine, which is precisely why the concession embedded in “less than we feared” carries weight. Established The reframing is substantive rather than cosmetic: the new baseline deliberately trades an early symbolic plasma for a more complete machine at first operation, and the Director-General's own word for the 2025 milestone was that it “was rather symbolic”.

Established Private capital is now the larger story, and the sector's only census is run by the sector. The Fusion Industry Association reports 56 companies, $4.48 billion raised in the twelve months to July 2026, $14.24 billion cumulative since 2021, five US companies past $1 billion each, and over 16,000 people employed. The same survey reports that 71% of companies expect a first plant delivering commercial electricity in the 2030s, and that respondents put the average capital needed for commercial viability at $2.7 billion, on a range from $100 million to $10.9 billion. Frontier A hundred-fold spread in what practitioners think their own industry costs is not a forecast; it is a measurement of how little is known — reported by the party with the strongest incentive to present coherence.

Established The best-capitalised effort is assembling hardware on a stated schedule. Commonwealth Fusion Systems had the first of 18 toroidal-field magnets installed as of early 2026, with first plasma targeted later that year and net energy in 2027, on a design gain above 10 at 50–100 MW of fusion power; its ARC plant in Chesterfield County, Virginia is described at roughly 400 MW electric for the early 2030s, against which Google signed a 200 MW offtake announced 30 June 2025, the largest corporate fusion offtake on record. Frontier A power purchase agreement against a plant that has not been designed to a construction permit is a financing instrument and a signal of intent, not evidence about physics or schedule, and the price was not disclosed. Established Helion reported 150 million °C on its Polaris prototype in February 2026, is under contract to Microsoft data centres, and targets first electrons to the grid in 2028 in Washington State. Frontier Trade coverage records skeptics questioning the start date, the approach and “its relative lack of scientific updates”, and no third-party verification of the Polaris temperature claim has been published. Speculative A 2028 grid date from a company with no peer-reviewed net-energy result is a commercial commitment, and this brief treats it as one.

Established The failure record is the most informative body of evidence in the sector, because these are conceded losses rather than contested forecasts. General Fusion cut at least 25% of staff in May 2025, took a $22 million emergency round in August 2025, listed on Nasdaq on 13 July 2026 after a SPAC merger in which redemptions left under $30 million of a possible $230 million, and holds about $150 million cash against more than $600 million raised since 2002; its LM26 breakeven milestone moved from 2026 to 2028 or later and its first plant to about 2035. Established First Light Fusion announced on 3 March 2025 that it was abandoning Machine 4 and the power-plant path entirely, pivoting to selling amplifiers and targets into other people's machines, and its lead investor cut the valuation from £236 million to £100 million — a 60% write-down. Established TAE Technologies deleted an entire device generation, cancelling Copernicus and stating it would move directly from the Norm experiment to a first-of-a-kind plant. Frontier Deleting a demonstration step is presented as acceleration; it reads equally as capital discipline under funding pressure, and no external evidence distinguishes the two.

Established Tritium is the constraint popular treatments omit, and the arithmetic is unforgiving. Tritium does not occur naturally in usable quantity, has a 12.3-year half-life, and the world civil stock is estimated at 25–30 kg, produced as a by-product of heavy-water reactors at roughly 2 kg a year in Canada, at about $30,000 per gram on the Canadian market and near $50,000 per gram for US production. The stock decays at about 5% a year. A single commercial reactor may need up to 10 kg simply to start, and a 1 GW plant may consume more than 55 kg a year in operation. Frontier Put those published figures together: a 30 kg stock decaying at 5% loses about 1.5 kg a year against roughly 2 kg a year of production — a net accumulation on the order of half a kilogram annually. The entire world civil inventory is enough to start about three plants and would be consumed by one of them inside seven months of operation. Established A university policy assessment records fewer than four kilograms a year from roughly 30 CANDU-style reactors globally, and the fleet is not growing.

Established Breeding has never been demonstrated in a closed loop. Every credible design requires the plant to make its own fuel from lithium in a blanket surrounding the plasma. The most detailed public analysis — by Type One Energy with the University of Wisconsin, and therefore an interested party — sets a viability target tritium breeding ratio of 1.35, computes 1.30 for its helium-cooled pebble-bed baseline in three-dimensional neutronics, and notes the operational requirement is a TBR under 1.05 for self-sufficiency with an operating inventory near 675 g. Its own caveats are the useful part: the model is “a simplified neutronics model of an early design iteration”, material behaviour in the relevant environment carries “relatively high uncertainty”, and pressure drop was not accounted for. Established No facility anywhere has bred more tritium than it burned.

Established Materials qualification runs on a clock nobody can shorten with money. Fusion first walls face 14.1 MeV neutrons and damage rates above 15 displacements per atom per year in reduced-activation ferritic-martensitic steel. Fission reactors cannot substitute as test beds, and the reason is spectral rather than budgetary: fission neutrons average around 2 MeV, below the thresholds that transmute iron and generate helium, so in the programme's own words “gas is not efficiently generated” — and helium embrittlement is exactly the failure mode a fusion structural material must be qualified against. Spallation sources are an order of magnitude short in flux at the relevant energies. Established The facility built to close the gap, IFMIF-DONES at Granada, is an accelerator-driven neutron source and is not yet operating. Handwave A schedule that puts a commercial plant in the early 2030s while the facility that would qualify its first wall has not begun operation is asserting that qualification will not be on the critical path. That assertion is rarely defended and it is almost never stated.

Established The economics have a threshold and it is not fossil fuel. The most complete peer-reviewed costing models tenth-unit levelised costs of $150+/MWh for an early large plant, $83–135/MWh for an advanced large plant at a 7% discount rate, and $101–145/MWh for an advanced small plant at the 75th unit, against first-of-a-kind small-reactor capital costs of $29,670–34,500 per kWe. Its own threshold is explicit: “for fusion to be competitive beyond 2040, costs will likely need to be at or below ~$80–100/MWh”, conditional on a fleet programme rather than single projects, capacity factors moving from 54–56% to 75–85%, higher cycle efficiency and proportionate regulation. Established The comparator is not gas: a university policy assessment puts early fusion above $0.15/kWh against utility-scale solar at $0.03–0.09/kWh. Frontier A technology arriving in the 2040s must beat what solar, wind, storage and firm alternatives cost in the 2040s, and every year of slip moves that target further away. This is the same structural argument Hydrogen Economies makes about hydrogen and Energy Storage Revolutions makes about long-duration storage, and it is the strongest general result in this category.

Established The subject's real open questions are no longer about whether a plasma can burn. They are about which confinement concept survives contact with a fuel cycle, a wall and a price — and the field's long tail of alternatives deserves ranking by what each has actually shown rather than by how often it appears in a press release. What follows is the live hypothesis space, flagged individually.

Frontier Compact high-field tokamak. The best-capitalised wager, and the physics behind it is standard: high-temperature superconducting REBCO tape allows higher fields, and fusion power density scales steeply with field, so a burning-plasma device can be far smaller than ITER. Frontier What would settle it is SPARC reaching Q above one in D-T, then ARC demonstrating net electricity with a breeding blanket attached — the second half of which no funded machine has yet attempted. Frontier ITER-class conventional tokamak is the same physics without the field advantage and with the most complete engineering integration ever attempted; its physics case is frontier and its 2039 schedule is speculative.

Established Stellarators are the strongest non-tokamak result and the field corrected its own record about them. Wendelstein 7-X sustained a record triple product for 43 seconds on 22 May 2025, at over 20 million °C rising to 30 million, with about 90 hydrogen pellets injected, and separately achieved 1.8 GJ of energy turnover in a 360-second discharge, up from 1.3 GJ in 2023. Established The operating institute then walked the first claim back: an update of 30 June 2025 records that the decommissioned JET tokamak sustained comparable triple products for up to 60 seconds on previously unpublished data, and now describes the two machines as joint leaders. Frontier That correction is unusually good practice and it is the reason the figure can be trusted. Stellarators have no disruption class and are inherently steady-state; what would settle the concept is one with a breeding blanket reaching Q above one, which no funded machine targets.

Frontier Sheared-flow-stabilised Z-pinch has moved from physics to repetition, which is the right problem to be working on and is not gain. Zap Energy's Century platform runs at 0.2 Hz — twelve shots a minute — through over 100 consecutive shots in its record run, with DOE certification of more than a thousand consecutive shots in February 2025, using 2,500 pounds of circulating liquid bismuth as conductor, plasma-facing surface and coolant, at 39 kW delivered to the chamber from 57 kW input. Established Those are power levels three to four orders of magnitude below a plant, and the platform has not claimed fusion gain. Frontier Field-reversed configurations have a genuine new result: TAE reports the first FRC formed using only neutral beam injection, a thirty-year objective, claiming up to 50% reduction in reactor length and complexity, with the Norm device being upgraded toward 100 million °C. Helion's pulsed FRC-compression variant is the one architecture that could in principle demonstrate net electricity without a steam cycle, through direct recovery. Frontier Magnetised target fusion — General Fusion's magnetic compression — now has a financing status rather than a physics one, its breakeven shot deferred on funding grounds. Speculative Laser inertial fusion energy has the physics done and nothing else: what would settle it is a driver above 1% wall-plug efficiency running at several hertz against a target factory, and no such driver exists. Speculative Projectile and hypervelocity inertial fusion was abandoned by its own developer as a power-plant route in March 2025, and that abandonment is the strongest datum about it.

Speculative Aneutronic proton–boron-11 is the concept whose appeal inverts its difficulty. The cross-section is real and the reaction is genuine; a p-B11 plant would eliminate the tritium problem and most of the neutron-damage problem at a stroke. It also requires plasma conditions roughly two orders of magnitude harder than deuterium–tritium, and no device has approached them. Handwave Nobody has demonstrated net energy in any fuel; proposing to skip to the hardest one, on the ground that it avoids the consequences of the easiest one, is a coherent hypothesis with no supporting device result.

Speculative Inertial electrostatic confinement — the fusor lineage, including Polywell and its descendants — reliably produces fusion neutrons on a bench and is sold commercially as a neutron source. Handwave As a power claim it has never had a credible path past bremsstrahlung and grid losses, and no design has published an energy balance approaching unity. It is real fusion and it is not a power source, which is exactly the distinction the three thresholds draw. Handwave Beam-target and colliding-beam schemes fail on the same standard objection: in any non-thermal distribution, scattering losses exceed fusion yield by orders of magnitude. The objection is a textbook result and has never been answered.

Speculative Muon-catalysed fusion is the most instructive minority concept, because its failure is quantitative rather than vague. Muons catalyse D-T fusion at room temperature and this is not disputed. The Los Alamos benchmark achieved about 150 catalysis cycles per muon, alpha-sticking sits near 0.45%, and muon production costs about 5 GeV each at pion facilities. A 2026 preprint computes that Q = 1 requires at least about 284 fusions per muon and proposes a four-part scheme that might reach more than 500 cycles and Q above two — while stating plainly that these are “model-based projections rather than experimentally demonstrated performance benchmarks”. Handwave Every published route past the sticking limit does its work in a term that has not been measured.

Frontier Two live options are systematically under-covered. The first is fusion as a neutron source rather than a power source: a machine well below Q = 1 is already a useful producer of 14 MeV neutrons for medical isotopes, materials testing and fusion–fission hybrid proposals, and this is the one commercial application fusion can serve today. What would settle it is a customer paying for neutrons at a price that funds the machine. Frontier The second is state-led scale-up: China's BEST tokamak at Hefei is targeted for completion end-2027, had a 400-tonne, 18-metre Dewar base installed in October 2025, and states goals of burning D-T plasma and generating electricity from fusion for the first time. Speculative If BEST does what it says on the schedule it says, the Western compact-tokamak lead evaporates and the interesting question becomes state capacity rather than venture capital. No cost figure is public.

Established The first bottleneck is upstream of the blanket and is almost never mentioned: lithium-6 and beryllium. Lithium-6 is 7.4% of natural lithium and breeding blankets need it enriched; the historic industrial enrichment route was the mercury-based COLEX process, whose legacy contamination makes revival awkward and whose capacity the world allowed to lapse. Beryllium, the neutron multiplier most blanket designs assume, has global production near 170 tonnes a year against a maximum capacity around 350–400 tonnes and an official resource base near 100,000 tonnes. Speculative A fleet of plants each needing tonnes of beryllium and enriched lithium would be competing for a supply chain roughly the size of a single specialty-metals company, and this pack located no published analysis modelling that competition at fleet scale.

Established The second is the qualification facility, and it is public infrastructure private capital does not fund. No operating facility produces a fusion-relevant 14.1 MeV spectrum; fission reactors cannot substitute for a stated spectral reason; IFMIF-DONES is not operating. Frontier This is the same structural problem Advanced Fission identifies for fission — a qualification queue rate-limited by irradiation facilities that public bodies own — and it applies to fusion with less existing data behind it, not more. Established That brief's terminal claim is that fission's constraint is paperwork while fusion's is physics. The friendly amendment this brief makes is that fusion's constraint has largely stopped being physics too, and what remains is the same class of constraint fission faces, on a thinner data base.

Established The third is superconducting tape, and it is the one place where the engineering claim is genuinely strong and still supply-limited. REBCO tape is the whole basis of the compact-tokamak wager. It is also a supply chain: SPARC alone requires on the order of 10,000 km of tape, sector demand moved from hundreds of kilometres a year to thousands in a few years, and a leading supplier's reference production run in 2021 was just over 300 km in nine months. Frontier The United States imports essentially all its yttrium, 93% of it from China across 2020–23. Speculative A tape shortage is a plausible binding constraint on how many compact tokamaks can be built in the 2030s, and no public analysis this pack located quantifies the ceiling. The materials science itself belongs to High Temperature Superconductors; what is owned here is the fusion demand curve running into it.

Frontier The fourth is specific to inertial confinement and is a factory rather than a physics problem. A plant needs several shots a second against a target that is currently hand-made and fired about once a day, plus a driver whose wall-plug efficiency exceeds 1% — NIF's is near 2.9% measured as yield against draw, which is not the same quantity and is worse on the axis that matters. Handwave Every inertial power-plant roadmap contains a target factory nobody has built and a driver nobody has demonstrated, and the gap between them and the demonstrated system is presented as scale-up rather than as invention.

Frontier The fifth is the balance of plant, which is where scientific gain and engineering gain diverge. The peer-reviewed costing's own route to competitiveness runs through capacity factors rising from 54–56% to 75–85% and higher thermodynamic cycle efficiency. Speculative Those are availability and heat-engine numbers for a machine class whose maintenance regime has never been demonstrated, in which the first wall is activated, the coolant is tritiated, and remote handling is the only access route. The UK programme states net energy, tritium self-sufficiency and maintainability as a single combined goal precisely because the three cannot be demonstrated separately and then assumed to compose.

Established Nothing on this map produces a result this brief waits on. Fusion's binding constraints are a fuel inventory, a neutron-irradiation facility, two upstream materials supply chains and a price — all of them recorded below as typed requirements rather than as edges to other briefs. No typed depends-on edge is claimed, and the reason is not that fusion is independent of everything: it is that the things it depends on are held by governments, mining companies and buyers rather than by researchers.

Established What it waits on from research is narrow and nameable. A tritium breeding ratio measured rather than computed, in a blanket built rather than modelled; irradiation data on reduced-activation steels at fusion-relevant helium production; a REBCO tape process at industrial yield; and a demonstrated tritium extraction and recycle loop closing at the throughputs a burning plasma implies. Frontier None of these is a discovery. Each is a measurement campaign that would take years on facilities that mostly do not exist. Speculative The one place where a genuine scientific unknown remains is high-fluence material behaviour: the extrapolations from fission and spallation data carry, on the qualification programme's own account, uncertainties that only a 14 MeV source can close.

Established The highest-value experiment in the subject is the one nobody has run: close a tritium loop. Breed tritium in a blanket, extract it, purify it, inject it, burn it, and account for the inventory — at any scale, with any breeding ratio, published. No facility anywhere has bred more tritium than it burned, and the field's most detailed public analysis computes 1.30 against a 1.35 viability target on a self-described simplified model of an early design iteration. Frontier A measured TBR from a built module, even below unity, would convert the single largest assumption in every commercial roadmap from a calculation into a datum.

Established Second: operate IFMIF-DONES and publish the first fusion-spectrum irradiation curves. The whole materials case currently rests on extrapolation from a spectrum the programme itself says cannot substitute. Frontier This is a public-infrastructure decision rather than a research design, which is why it is recorded below as an industrial-capacity requirement rather than as an experiment somebody could choose to run cheaply.

Frontier Third: the milestones already scheduled, treated as the tests they are. SPARC reaching Q above one in D-T would be the first burning plasma in a private machine and would validate the field-scaling wager. BEST's end-2027 commissioning would be the first attempt at fusion electricity anywhere. ITER's 2039 D-T campaign remains the only integrated test of a full-scale machine. Speculative Each of these is a date rather than a design, and the field's own record on dates is the most predictive datum available.

Frontier Fourth, and cheapest: independent verification. Helion's 150 million °C has no published third-party confirmation; the wall-plug efficiency of every private device is unpublished; and the sector's census is run by its trade association. Established A standing, independently staffed measurement protocol — agreed thresholds, agreed instrumentation, published raw data — would cost a rounding error against $14.24 billion and would resolve more disagreement than any single machine. Frontier The LENR literature, discussed in section 14, is the worked example of what a field looks like after thirty-five years without one.

Established Magnets are the achievement, and they are real. High-temperature superconducting REBCO tape delivers the fields that make a compact burning-plasma device geometrically possible, and the first of eighteen toroidal-field magnets is installed in the leading private machine. Frontier The engineering question is no longer whether the magnet works but whether the tape supply, the joint technology and the quench protection survive a fleet.

Established The blanket is the component that has to do four incompatible jobs at once. It must breed tritium at a ratio above 1.05 net of every loss, multiply neutrons, extract heat at a temperature a turbine can use, and shield the magnets — while being repeatedly replaced by remote handling after neutron damage accumulating at over 15 dpa a year. Speculative No blanket meeting all four requirements has been built and operated. The helium-cooled pebble-bed baseline that has been analysed most carefully comes with its authors' statement that pressure drop was not accounted for and that material behaviour under the relevant environment carries relatively high uncertainty.

Established The fuel cycle is a chemical plant handling a radioactive gas that permeates metals. An operating inventory near 675 g with a burn fraction of a few per cent means the great majority of injected tritium comes back unburned and must be recovered, purified and re-injected on a timescale short against a 12.3-year half-life and against permeation losses. Frontier Every percentage point of loss in that loop has to be repaid by the breeding ratio, which is why the difference between a computed 1.30 and a required 1.05 is not comfortable margin — it is the entire allowance for a plant that has never been operated.

Established And the plant has to be available. Levelised cost at the competitive threshold requires capacity factors of 75–85%, which for a machine whose first wall is a consumable and whose maintenance is entirely remote is an assertion about a maintenance regime rather than a specification of one. Frontier The UK's STEP at West Burton targets operations in the 2040s against a £220 million initial package and aims to demonstrate net energy, tritium self-sufficiency and maintainability together, its chief engineer describing it as “a big extrapolation from today's machines” and its director saying “until we do STEP we won't know everything”. Frontier A national programme stating that it does not know what it will find, in the 2040s, is not consistent with private schedules promising grid electricity in 2028.

Within this map: Advanced Fission, which owns fuel cycles, code-qualified materials and the irradiation-queue argument this brief borrows and restates as a fusion problem; High Temperature Superconductors and Superconducting Infrastructure, which own REBCO as a materials and systems question where this brief owns only the fusion demand curve into it; Energy Storage Revolutions, which lists firm generation as the alternative to storing anything, making fusion an adjacency rather than a dependency; Small Modular Reactors, whose first-of-a-kind cost record is the closest available analogue; Nuclear Waste Solutions, since activated first walls are a waste stream nobody has costed; and Ultra Efficient Computing Energy Systems, which explains why data-centre buyers are the ones signing fusion offtakes.

Hydrogen Economies is the methodological neighbour rather than a topical one, and the boundary matters. That brief adjudicated the hydrogen-as-carrier thesis as failing on three legs and holding on one, and concluded that it depends on nothing on this map. Fusion does not compete with hydrogen and this brief makes no claim about hydrogen. What the two share is an instrument — a sector's own withdrawal record is stronger evidence than its announcements, and a scale ratio settles arguments a trend line cannot — and one substantive result: a technology arriving in the 2040s must beat 2040s alternatives. Fusion's verdict here is less severe than hydrogen's, because fusion has no deployed competitor beating it on physics the way heat pumps beat hydrogen heating.

This brief supersedes, and does not replace, the carbon programme's treatment, which reaches the same subject from the emissions side and asks what fusion contributes to the 37 Gt denominator. That page remains the shorter route in. Everything here on tritium arithmetic, breeding ratios, neutron qualification, the alternative confinement concepts and the LENR literature is new to this brief.

Outside the map: plasma physics and magnetohydrodynamics; nuclear materials science, which supplies the displacement-damage framework; the specialty-metals industry, which supplies beryllium, lithium-6 and yttrium; and weapons-physics stewardship, which is why NIF exists and why its shot rate is what it is.

Established The most consequential institutional fact in the subject is that the qualification infrastructure is public and the schedules depending on it are private. IFMIF-DONES is a state project at Granada; the irradiation data it would produce underwrite every private first wall; no private company is funding it and none could shorten it. Frontier The same asymmetry appears with tritium: the civil stock exists because Canadian utilities operate heavy-water reactors for reasons that have nothing to do with fusion, and the fleet is shrinking. A commercial sector's critical path runs through decisions made by public bodies for unrelated reasons, and the sector's roadmaps do not price that.

Established Regulation moved, and in fusion's favour. The US Fusion Energy Act of July 2024 placed fusion under particle-accelerator regulation rather than fission licensing — defensible on hazard grounds, since a fusion device has no critical mass and a small prompt inventory, and materially shortening approvals. Frontier The tritium inventory and the activated structure are still radiological hazards, and the regulatory settlement has not been tested against an operating plant, because there is not one.

Established Academic capacity is thin in a way that constrains everything downstream. The median number of fusion faculty in a US programme is 2, against 16.5 for fission. A sector employing 16,000 people and planning fleets is drawing on a training pipeline an order of magnitude narrower than the comparator technology's. Established And the sector's own numbers come from the sector: the Fusion Industry Association is the only body running the count, which makes its census simultaneously the best available data and an interested one. Frontier Where its figures are cited here, they are cited as the industry's self-report — and the most useful of them, the hundred-fold spread in estimated capital requirement, runs against the interest in presenting coherence.

Frontier The live ethical question is allocation, and the sector does not usually state it as one. $14.24 billion of private capital and substantial public programmes are committed to a technology that will not contribute materially to emissions reduction before 2040 on any evidence in this pack. Established That is not an argument against fusion research, which is cheap against its potential payoff and produces spillovers in magnets, materials and plasma processing. Frontier It is an argument for stating the opportunity cost explicitly rather than allowing a 2030s grid-electricity claim to do the work of one — because a policymaker who believes firm carbon-free power arrives in 2032 makes different decisions about what to build now.

Speculative The supply-chain ethics are unexamined because the fleet is hypothetical. Beryllium is toxic in respirable form and its production is concentrated; lithium-6 enrichment historically ran on a mercury process whose legacy contamination is why the capacity is gone; yttrium supply is 93% import-dependent from a single country for the United States. Speculative A fusion fleet would be a new claimant on three constrained and geopolitically awkward materials markets, and no published assessment this pack found treats that as an ethical question rather than a procurement one.

Established And there is a claim-discipline duty that this field discharges unevenly. Ignition was reported accurately by the laboratory and inaccurately almost everywhere else; the wall-plug ratio near 2.9% was rarely printed alongside the gain figure; power purchase agreements against undesigned plants were reported as evidence about schedules. Frontier A field whose credibility rests on a single well-executed physics result has a strong interest in not spending that credibility on financing announcements, and the withdrawals of 2025–26 are the first instalment of the bill.

Established The terminal position is narrower than the framing and still substantial. The plasma-physics question of whether fusion can release net energy has been answered, twice, in two schemes. Every remaining question is about fuel, materials qualification, supply chains and price. That is a genuine change in what the subject is, and it is not the same as “an engineering and financing problem”, because the fuel problem is not solved by engineering or by financing.

Frontier If it works, the prize is a firm, fuel-light, siteable, carbon-free generator with no long-lived fission-product inventory and no meltdown mode — which is a different kind of asset from anything on the grid, and the reason the wager is worth making even at these odds. Speculative The scale of the prize also explains the field's characteristic distortion: a payoff large enough to justify a small budget at low probability is also large enough to justify overstating the probability, and those two facts are hard to separate from outside.

Frontier The most likely near-term civilizational contribution is not electricity at all. A machine well below breakeven is already a useful 14 MeV neutron source, and neutrons are scarce: medical isotopes, materials qualification for both fusion and fission, and hybrid proposals all want them. This is the one fusion application that could be sold this decade, and it is the least discussed. A sector that funded itself partly on neutron sales would be a sector with revenue, external customers and independently audited performance — three things it currently lacks.

Speculative And the long-run question is whether the fuel cycle can be industrialised at all. A world with a fusion fleet is a world that enriches lithium-6 at scale, mines and refines beryllium at multiples of current output, and operates a closed tritium economy in which every plant makes its own fuel with no external market to fall back on. That is not a technology; it is an industrial system, and no part of it exists. Handwave Any account of a fusion-powered civilisation that skips from a burning plasma to a fleet is skipping the part that has never been done.

These horizons track fuel inventories, qualification facilities and machine commissioning dates rather than plasma performance:

• 10 yr: Frontier The dated tests land in this window. SPARC's D-T campaign, BEST's end-2027 commissioning with a stated goal of first fusion electricity, and Helion's 2028 grid claim all resolve one way or another. Established ITER does not: its D-T phase is 2039. Speculative Expect at least one further private withdrawal on the pattern already set — a deleted device generation, an abandoned power-plant path, a breakeven milestone slipping past a funding round — because the sector's own capital estimate spans a hundred-fold range and some of those companies are wrong about their own costs. Handwave A commercial plant delivering electricity to a grid in this window would require a breeding blanket, a qualified first wall and a fuel inventory that do not exist, and no published roadmap explains how all three arrive.
• 25 yr: Frontier This is the window in which the tritium question is settled or the fleet question is abandoned. A measured breeding ratio above unity from a built blanket, plus IFMIF-DONES irradiation curves, would move fusion from a physics result to an engineering programme. Speculative Absent both, the plausible outcome is a small number of demonstration machines, no fleet, and a subject that has proved its physics and not its supply chain. Speculative The stellarator line and the state-led tokamak line are the two most likely to still be running, for opposite reasons — steady-state operation and state balance sheets.
• 50 yr: Speculative If fusion contributes materially to primary energy, it does so in this window and not before, and the competitive question is whether it beats what the 2070s grid costs rather than what the 2020s grid costs. Speculative Aneutronic p-B11 either has a device result by then or is a permanent minority position; nothing in the current record distinguishes those. Handwave Projections of fusion's share of world energy at this horizon are outputs of models whose input assumption is that the fuel cycle closed, which is the thing being asked.
• 100 / 250+ yr: Handwave Beyond useful forecasting. The one durable observation is structural rather than predictive: deuterium is effectively unlimited and tritium is not, so a very long-run fusion economy is either a breeding economy that closed its loop or an aneutronic one that solved a harder plasma problem. Handwave Which of those a civilisation ends up with is not extrapolation from any measurement now on the record.

• Depends on Nothing on this map. No brief in this corpus produces a result commercial fusion waits on: scientific gain is demonstrated, and what remains is a fuel inventory, an irradiation facility, two materials supply chains and a price. No typed depends-on edge is claimed, and the reason is that the binding constraints are held by governments, miners and buyers rather than by researchers.
• Requires (not on this map) Tritium is the sharpest constraint on this map for this subject and it is a supply chain rather than a technology. The world civil stock is 25–30 kg, produced at roughly 2 kg a year as a by-product of heavy-water reactors and decaying at about 5% a year, for a net accumulation near half a kilogram annually — against up to 10 kg to start a single plant and more than 55 kg a year to run a 1 GW one. Every plant must therefore breed all of its own fuel from its first year, at a ratio nobody has measured, in a blanket nobody has built, from lithium-6 that is 7.4% of natural lithium and must be enriched by a mercury-based process the world stopped running, using a beryllium multiplier whose global production is about 170 tonnes a year against a 350–400 tonne capacity. Alongside that sit two industrial capacities: an operating 14.1 MeV irradiation source, since fission spectra average near 2 MeV and do not generate the helium that embrittles a first wall, and IFMIF-DONES at Granada is not yet running; and REBCO tape at thousands of kilometres a year against a 2021 supplier benchmark of just over 300 km in nine months, with 93% US yttrium import dependence on a single country. The last is a price rather than a thing: the field's own peer-reviewed costing puts the competitiveness threshold at roughly $80–100/MWh beyond 2040 and models $83–150/MWh for a tenth unit, against utility solar already at $0.03–0.09/kWh. None of the five is a research result and none is shortened by capital.
• Enables In principle a firm, siteable, carbon-free generator would relieve the firm-power requirement that Energy Storage Revolutions quantifies and would supply the cheap electricity Hydrogen Economies identifies as two-thirds of green hydrogen's cost stack. No typed enabling edge is claimed, because an enabling relation conditional on a plant that has never produced a watt is a hypothesis rather than an edge, and asserting it would let a 2040s possibility carry weight in a 2020s dependency graph.
• Adjacent Plasma physics and magnetohydrodynamics; nuclear materials science; the specialty-metals industry for beryllium, lithium-6 and yttrium; and within this map Advanced Fission, High Temperature Superconductors, Nuclear Waste Solutions and Ultra Efficient Computing Energy Systems.

Handwave “Fusion has achieved net energy.” It has achieved target gain: more fusion energy out of a capsule than laser energy delivered to it. NIF's laser draws roughly 300 MJ of electricity per 2 MJ shot, so the record 8.6 MJ sits against about 300 MJ drawn — a wall-plug ratio near 2.9%. Established No device anywhere has produced net electricity, or net energy at the wall plug: not NIF, not JET, not W7-X, not any private machine. Quoting a target gain as though it were a plant result is the field's characteristic error and it is made by advocates and critics alike.

Frontier “A signed power purchase agreement shows the schedule is credible.” Google's 200 MW ARC offtake and Microsoft's Helion contract are financing instruments against plants that do not exist, at prices that were not disclosed. Established They are evidence that large buyers will pay a small option premium for firm carbon-free power; they are not evidence about physics, engineering or dates. Frontier The sector's own survey has 71% of companies expecting commercial electricity in the 2030s and an average capital requirement spanning $100 million to $10.9 billion. Both cannot be right.

Handwave “Tritium breeding is solved engineering awaiting scale-up.” The most detailed public analysis computes a breeding ratio of 1.30 against a 1.35 viability target, on what its own authors call a simplified neutronics model of an early design iteration, with material behaviour at relatively high uncertainty and pressure drop unaccounted for. Established No facility has ever bred more tritium than it burned. The gap between a computed number and a measured one is the whole subject here.

Speculative “Aneutronic fuels avoid the hard problems.” They avoid tritium and most neutron damage, which is real and important. Handwave They do so by requiring plasma conditions roughly two orders of magnitude harder than the fuel nobody has yet run a power plant on. The claim that p-B11 is easier because it avoids neutrons inverts the difficulty, and the honest statement is that it trades a materials and fuel-cycle problem for a confinement problem nobody has approached.

Handwave “Fusors and inertial-electrostatic devices are fusion, so they could be scaled to power.” Fusors demonstrably fuse and are sold as neutron sources; that is the whole of the evidence. No inertial-electrostatic scheme has published an energy balance approaching unity, and none has a credible answer to bremsstrahlung and grid losses. Handwave The same standard objection retires beam-target and colliding-beam power claims: scattering losses exceed fusion yield by orders of magnitude in any non-thermal distribution, which is a textbook result that has never been answered.

Handwave “Cold fusion was suppressed rather than tested.” This is the claim this section exists for, and the answer is unusually strong evidence rather than dismissal. Established The 1989 Fleischmann–Pons claim of excess heat from deuterium-loaded palladium was not replicated, and the mainstream objection is physical rather than cultural: the claimed heat is not accompanied by the neutron and gamma signatures that deuterium fusion at those rates must produce. Established The 2004 US Department of Energy review is the most-cited data point on both sides and says less than either wants: eighteen reviewers split approximately evenly on whether excess power had been demonstrated, the panel found the claims “intriguing but not convincing” with much published work poorly documented, and the reviewers were nearly unanimous that well-designed proposals should be eligible for funding. Frontier That is the honest terminal position of the mainstream evaluation: not proven, not funded as a programme, not formally excluded.

Established What changed the evidence is a funded, competent, well-designed replication attempt that found nothing. Google funded a programme from 2015 to about 2018 recruiting four principal investigators and roughly 30 researchers across about 12 projects, published as Berlinguette et al. in Nature in 2019. Three independent lines were run: electrochemical and solid-state loading of palladium to high deuterium concentrations, high-sensitivity calorimetry on hydrogenated palladium, and pulsed-plasma ion bombardment for screening effects. It achieved palladium hydride loading near 0.96, reported no evidence of the effect, produced 28 peer-reviewed publications and an unrelated useful result in deuterated pharmaceuticals, and was reported as concluding that researchers “did not find credible evidence of anomalies associated with cold fusion”. Frontier This is the highest-quality negative result the field has, and its provenance is what makes it strong: privately funded, therefore free of any incentive to find nothing, run by people who published their protocols and who argued in the same paper that the parameter space of highly hydrogenated metals remains under-explored and worth instrumenting. A well-resourced group that wanted to find an effect, designed carefully to find it, and did not, is much stronger evidence than any amount of dismissal.

Established The field was not defunded on that basis either. ARPA-E subsequently funded $10 million across 8 projects to “determine whether low-energy nuclear reactions could be the basis for a potentially transformative carbon-free energy source” and to break the stalemate, with the agency's framing conditional throughout. Frontier No programme-level result establishing an effect has been published from it as of this brief's date, and this brief does not treat the absence of a published positive as a published negative.

Speculative The commercial LENR sector's own census is the most damaging document available, and its provenance is why. A 2026 landscape report by a physicist sympathetic to the field counts 30 organisations across 12 countries and about $42.9 million disclosed as received 2020–2026, with companies reporting coefficients of performance mostly in the 1.5–5 range and one claiming about 24 kW thermal per module. Established Its own conclusion is the finding: no organisation in the survey has publicly documented a commercial deployment supported by broadly available independent performance data, the highest-rated case rests on “company-reported, third-party-witnessed” operation rather than independent audit, and the author names the threshold that would settle it — an independently audited, sustained, multi-week, megawatt-scale demonstration. Frontier That an advocate's census reaches that conclusion raises its weight considerably, because the interest runs against the finding. Handwave Every energy-source claim built on the nickel–hydrogen commercial lineage rests on company-reported coefficients with no independent audit, and the relevant fact is the absence of the audit rather than the presence of the claim.

Handwave Why the field persists is explicable without any of it being true. The 2004 panel's even split is quotable indefinitely; calorimetry at the few-watt level is genuinely difficult and a null result is always attackable as insufficiently sensitive; and the payoff if true is so large that a small expected probability justifies a small budget. Frontier The last of those is a defensible argument for the ARPA-E programme and it is not an argument that the effect exists. Speculative One mild signal is worth recording: after finding nothing in cold fusion, the Google-funded group pivoted toward nuclear excitation and photonuclear routes — a well-run programme choosing a different problem is weak evidence about relative promise, and it is offered here as exactly that.

Frontier “ITER is a €20 billion project.” Nobody knows. The 2001 redesign was costed near €5 billion, the 2010 estimate near €15 billion, the 2024 rebaseline added about €5 billion — and the Director-General declined to state a total, on the ground that in-kind contributions make one meaningless. Established Any single figure quoted for ITER's cost is a construction, not a disclosure.

Frontier “Fusion is part of the 2030s decarbonisation plan.” Nothing in this evidence base supports a material emissions contribution before 2040. Established The competitiveness threshold in the field's own costing is $80–100/MWh beyond 2040, against utility solar at $0.03–0.09/kWh today and falling. Speculative The opportunity-cost question — capital and attention that could deploy technologies working now — is a live allocation argument that the sector rarely states as one, and stating it is not an argument against funding fusion research.