Why nuclear fusion has been 20 years away for 70 years — and whether that’s finally changing

On December 5, 2022, scientists at the National Ignition Facility in Livermore, California fired 192 laser beams at a target the size of a peppercorn. The fusion reaction produced 3.15 megajoules of energy from 2.05 megajoules of laser light — a ratio of approximately 1.54. The Department of Energy called it a historic milestone. It was.

The number that appeared in the technical record but not in most headlines: it took roughly 300 megajoules of electricity to power those lasers. Full accounting: 300 megajoules in, 3.15 megajoules out. Nobody lied. The scientific Q definition used — laser energy at target divided by fusion energy out — is standard for inertial confinement research, and the NIF was built to prove ignition, not generate power. By its own measure, it proved it.

But there’s a specific choice embedded in how that result was framed, and that choice has characterized fusion’s relationship with its audience for seventy years. The science delivers something precise; the public story is somewhat less so.

What the physics actually requires

The D-T reaction is simple enough to write in a line: a deuterium nucleus and a tritium nucleus collide at extreme temperature, fuse, and release a helium-4 nucleus carrying 3.5 megaelectronvolts plus a neutron carrying 14.1 MeV. Those two products have very different fates, and understanding both is the whole game.

The alpha particle — the helium-4 — stays in the plasma, depositing its energy there and heating the plasma further without external input. This is alpha-particle self-heating, the mechanism that makes self-sustaining ignition possible, and the specific phenomenon the December 2022 NIF shot demonstrated in a controlled setting for the first time. The neutron carries eighty percent of the reaction energy in the opposite direction: through whatever surrounds the plasma, into structural materials, at one of the highest energies produced in any engineered system. The alpha particle enables ignition. The neutron is what makes commercializing it so hard.

Achieving the reaction requires plasma at roughly 100 to 150 million degrees Celsius — ten times hotter than the core of the sun. The sun fuses hydrogen under gravitational pressure so crushing that particles fuse at comparatively modest temperatures; on Earth, with no such pressure, temperature does all the compensating. The engineering consequence: sustaining plasma ten times hotter than the solar core inside a machine whose walls cannot touch it.

The Lawson criterion, established by physicist John Lawson in a 1957 paper in the Proceedings of the Physical Society B, formalizes simultaneous success: the product of plasma density (n), temperature (T), and energy confinement time (τ) must exceed roughly 3 × 10²¹ keV·s/m³ for D-T fusion. All three parameters must hold at once. A pressure fluctuation, a magnetic disruption, a geometry irregularity — any perturbation can cascade to failure, because there is no partial credit in the triple product. Achieving temperature is not remotely sufficient. You need density and confinement time too, simultaneously, while temperature is at peak.

Q — fusion energy out divided by energy used to initiate the reaction — sounds simple until you ask what “initiate” means. Different definitions produce Q values differing by a factor of a hundred for the same experiment. Choose laser energy at the target, and a shot can look like a win. Account for the electricity that powered the lasers, and the same shot looks like a catastrophic loss. The definitional choice isn’t dishonest — different Q values answer different scientific questions — but it creates ample room for the public-facing story and the engineering reality to diverge substantially.

The physics of what D-T fusion requires has been understood since Lawson published. What has taken seventy years is the engineering of doing it at commercial scale, for long enough, in a structure that survives, at a cost that competes. The science is not the mystery. Something else is.

The Q alphabet

Laser-to-target Q measures only the energy the laser beam delivers to the target capsule. December 5, 2022: 2.05 MJ in, 3.15 MJ out. Q ≈ 1.54. This is the Q that was reported.
Driver Q accounts for the lasers' efficiency at converting electricity to laser light — NIF's lasers run at roughly 1% efficiency. Same shot: Q ≈ 0.015.
Wall-plug Q measures total facility electricity consumed versus fusion energy produced. The NIF facility drew roughly 300 MJ to run the shot. Wall-plug Q for the December 2022 result: approximately 0.01.
Commercial electricity requires wall-plug Q high enough to cover turbines, heat exchangers, tritium blankets, neutron shielding, and plant overhead — after all losses. Fission reactors operate at effective wall-plug gains of 30 to 50. No fusion device has reached one.

The institutional architecture of always-almost-there

Lev Artsimovich, the Soviet physicist who led the tokamak program through the 1960s and whose design approach shaped fusion research for the next half-century, said: “Fusion will be ready when society needs it.” The line gets quoted as philosophical patience — the long view on a hard problem.

Read carefully, it’s an indictment.

Society hasn’t needed it badly enough. Urgency drives completion; proximity without urgency produces perpetual proximity. This is the structure that has governed fusion since the beginning. Fusion has been funded as big science — the kind of work requiring international agreements, decades of commitment, timelines set at the frontier of what looks achievable. That frontier advances. But commercial application requires an engineering stack that develops more slowly, and that engineering stack is structurally less fundable. Scientists measure proximity to the next scientific milestone, not proximity to the last engineering step, because those are different destinations and the first is where their incentives point.

ITER is the clearest documentation of this pattern. Proposed in 1985, with formal agreement in 2006, original cost estimate approximately $12 billion. Target for first plasma: 2025. Construction began in Cadarache, France. By 2020, it was internally clear within the project that the 2025 target was unachievable. The French nuclear regulator ASN halted construction in January 2022 over quality control problems with welds in the vacuum vessel. In July 2024, ITER announced a revised schedule: full D-T operations in 2039 — four years beyond the previous 2035 target, and roughly a decade beyond what the original 2006 agreement had implied — plus an additional $5.2 billion in costs, bringing the total to an estimated $22 billion or more.

ITER cannot be cancelled. Political and financial commitments across 35 nations are too deep. When the timeline slips, extension is the only tool available. This isn’t bad faith and it isn’t bad science. It’s a project architecture that converts difficulty into schedule revision rather than resolution, because resolution isn’t available and extension always is.

NIF proves the pattern isn’t ITER-specific. The GAO’s 1996 baseline for the facility was approximately $1.1 billion, with completion projected around 2002. By August 2000, the U.S. General Accounting Office had published a formal report — RCED-00-271 — documenting that key DOE program officials lacked appropriate management skills, that oversight failed to surface escalating costs until months after they were internally documented, and that the total projected cost was likely to reach $3.9 billion. The facility completed in 2009 at approximately $3.5 billion — roughly three times the baseline — approximately seven years late.

Two programs, different decades, different continents, same pattern. Individual mismanagement can be corrected. Structural failure modes recur.

Three problems the excitement hasn’t touched

Beyond institutions and incentives, the three engineering problems that actually determine fusion’s arrival date are these: tritium supply, materials degradation, and wall-plug Q. None can be solved by raising more capital. No current program has solved any of them.

The first is tritium. D-T fusion requires tritium — a radioactive hydrogen isotope with a half-life of 12.3 years — which doesn’t exist in meaningful natural quantities. The world’s entire commercial supply, approximately 20 to 25 kilograms of usable tritium out of a global inventory of roughly 50 kilograms total, comes almost entirely from CANDU heavy-water fission reactors in Canada and South Korea, where it accumulates as a byproduct. A 2018 analysis by Kovari and colleagues in Nuclear Fusion traced what happens next: ITER, when it begins D-T operations in 2039, will consume roughly one kilogram per year. The global CANDU fleet is aging; multiple reactors face retirement across the 2030s and 2040s. Supply is expected to peak this decade, then decline. A single commercial fusion power plant would require several kilograms per year.

The proposed solution is tritium breeding: lithium blankets surrounding the reactor absorb D-T neutrons and produce new tritium via Li-6 + n → He-4 + T. The required tritium breeding ratio — above roughly 1.05, to account for losses — leaves almost no margin. And tritium breeding has never been tested at scale in a fusion device. The lithium blankets haven’t been operated in a neutron environment approaching commercial fusion conditions. Every proposed commercial D-T plant is assuming a process that hasn’t been demonstrated. That’s a technological demonstration that has to happen, and it cannot be bought.

The second problem is materials. The 14.1 MeV neutrons from D-T fusion displace atoms in structural material crystal lattices — causing embrittlement — transmute elements (a tungsten atom can become rhenium or osmium, changing mechanical properties entirely), and drive tritium into the wall itself. Tungsten is the leading candidate for the plasma-facing first wall: high melting point around 3,400°C, low hydrogen permeability, relatively low sputtering rate. But no one has tested tungsten — or any other candidate — under commercial fusion neutron flux conditions, because the facility to do so doesn’t yet exist.

IFMIF-DONES, the International Fusion Materials Irradiation Facility being built in Granada, Spain, is designed to fill this gap. Beam commissioning is projected for 2029, with the facility potentially operational by 2034; materials qualification requires years of data collection beyond that. Even if SPARC demonstrates plasma Q > 1 in 2027, the design of the next device — the one extracting commercial electricity — requires materials data that won’t exist until the mid-2030s at the earliest, regardless of capital availability.

The third problem is wall-plug Q. JET, the Joint European Torus, holds the tokamak Q record: approximately 0.67, achieved in 1997. No magnetic confinement device has reached Q = 1. The NIF’s Q ≈ 1.54 was by the laser-to-target definition; at wall-plug level, the same shot corresponds to Q ≈ 0.01. Commercial electricity requires a wall-plug Q high enough to cover the full plant — turbines, heat exchangers, tritium breeding blankets in operation, shielding, activation management. Fission reactors run at effective wall-plug gains of 30 to 50. The factor of several hundred between the best demonstrated Q in any format and commercial viability is a real engineering project, not a rounding error.

And the target moves. Solar photovoltaics have reached grid-competitive costs of $20 to $40 per MWh in many markets. Wind, advanced fission, and grid storage are on similar curves. The economic threshold fusion would need to beat in the 2040s isn’t fixed at today’s numbers. A technology that takes until 2045 to produce commercial electricity will be competing against energy systems that have had another twenty years of cost reduction. That’s not an argument against fusion — it’s an argument for treating the economic case as live and uncertain rather than assumed.

What private capital actually changed

The argument for private fusion isn’t hype. Parts of it are real.

The magnet work at Commonwealth Fusion Systems is the clearest case. In September 2021, CFS demonstrated a prototype high-temperature superconducting toroidal field coil using REBCO tape — rare-earth barium copper oxide — sustaining a magnetic field of 20 tesla. CFS has since advanced to production-scale magnets for SPARC; in September 2025, a DOE panel validated completion of that milestone under the department’s Milestone-Based Fusion Development Program. The physical significance: in a tokamak, plasma pressure scales with B², magnetic field squared. A 20 T magnet enables the same plasma conditions in a volume roughly one-fortieth that of ITER. Smaller volume means faster build, lower capital per experiment, faster learning cycles. This is a genuine structural change in what private teams can construct.

The incentive structure is also genuinely different. Private capital demands milestones and punishes misses. Helion Energy signed a power purchase agreement with Microsoft in 2023 committing to commercial electricity delivery by 2028 — the first commercial fusion PPA in history — with financial penalty clauses for failure. CFS has raised approximately $3 billion and targets SPARC’s first operations in 2026 and Q > 1 by 2027. TAE Technologies raised $150 million in a June 2025 round, bringing its total to over $1.3 billion. The Fusion Industry Association reported roughly $9.7 billion in total private investment across 53 companies, with $2.64 billion raised in the twelve months to July 2025 — a 178% increase over the prior year. These companies cannot simply extend their timelines. The money runs out.

But capital cannot breed tritium. Stronger magnets cannot qualify wall materials under commercial neutron flux. Milestone-based funding cannot accelerate IFMIF-DONES. The three problems above are not addressed by the private-capital wave because they are not problems that capital, in itself, solves — they require specific technological demonstrations that take time independent of how much money is behind them. The honest version of the private-capital argument: it has accelerated plasma confinement work and demonstrated that compact, rapidly iterated machines are achievable. It has not shortened the timeline on problems that sit downstream of confinement, where capital is not the binding constraint.

The companies, specifically

CFS’s SPARC is a tokamak. That matters. Tokamak physics is the best-understood confinement physics in the field — sixty years of experimental data, a mature theoretical framework, a clear experimental record. SPARC’s plasma physics is not an open question; the question is whether 20 T magnets and compact geometry can achieve it in a commercially viable form. In January 2026, CFS placed the first of SPARC’s 18 toroidal field magnets on the assembly jig, on track for first plasma in 2026 and a Q > 1 demonstration in 2027. If achieved, it would be the first magnetic confinement device to exceed JET’s 1997 record.

That result would be historically significant. It would not be a power plant. SPARC to ARC — CFS’s proposed commercial tokamak — runs through untested tritium breeding blankets, first-wall materials that can’t be qualified until IFMIF-DONES produces data, and a wall-plug Q several hundred times higher than SPARC’s demonstration figure. CFS projects ARC commercial in the mid-2030s. Not impossible. Requires every subsequent step to go right. CFS doesn’t claim to have solved tritium or materials — they haven’t. The risk profile here is engineering risk on established physics. Difficult, but defined.

Helion Energy is a different proposition. Their device uses a field-reversed configuration — a compact torus without the tokamak’s doughnut geometry — aiming long-term at deuterium–helium-3 fusion, which produces no neutrons and would eliminate the materials problem entirely. As an intermediate step, they’re targeting D-T fusion. The Microsoft PPA commits them to 50 MW of commercial electricity by 2028; in July 2025, Helion broke ground on Orion, their demonstration plant, in Malaga, Washington. Their Polaris test reactor is operational but has not publicly demonstrated net electricity production. Commercial electricity by 2028, from a fusion concept that hasn’t demonstrated net electricity at laboratory scale, requires multiple sequential technological leaps — and some of them are physics leaps, not engineering ones. Mike Campbell, a fusion engineering professor at UC San Diego, put it plainly: “I wish them well, but I’ll be happily surprised if they are able to do it by then.”

FRC confinement is significantly less mature as a physics discipline than tokamak confinement. Tokamaks have sixty years of detailed experimental data and a theoretical framework that accurately predicted confinement behavior at progressively larger scales; FRC has neither at the commercially relevant regime. The plasma physics of FRC stability at commercially relevant scales is not fully understood. A PPA with penalty clauses creates financial accountability. It does not create the physics. If the FRC approach works, Helion’s compactness and eventual neutron-free operation are genuine structural advantages. If the plasma physics proves intractable at scale, no amount of investor pressure changes that — and given how little FRC data exists at the relevant regime, intractable remains a live possibility.

TAE Technologies is the highest-risk proposition in the private landscape. Their eventual goal is proton-boron fusion — p-B11 — which produces three helium-4 nuclei and no neutrons, eliminating both the materials problem and the tritium problem at a stroke. This is not an engineering scale-up of known physics. It is an open plasma physics question. P-B11 requires temperatures of roughly 3 billion degrees Celsius, versus 150 million for D-T, and the plasma physics at that temperature has not been demonstrated at laboratory scale. In 2025, TAE demonstrated the first-ever formation of a field-reversed configuration plasma using only neutral beam injection — a milestone researchers had pursued for decades, published in Nature Communications. Their roadmap runs through a net-energy device called Copernicus and a commercial plant called Da Vinci in the early 2030s. Investors include Google and Chevron, both with specific strategic interest in energy outcomes.

The risk here is physics, not engineering. If the bet pays off, two of the three deepest structural problems disappear. If the plasma physics at 3 billion degrees proves intractable, no amount of engineering competence or capital patience closes the gap. TAE’s near-term roadmap, like Helion’s, runs through FRC plasma physics at commercially relevant scales that hasn’t been demonstrated. The 2025 NBI milestone is real and significant. It is also a long way from a net-energy device.

What would have to be true

Commercial fusion electricity before 2050 requires a specific sequence, and investor enthusiasm does not compress it.

Someone has to demonstrate plasma Q > 1 in a sustained, repeatable way — not a single shot, but confinement that holds long enough and hot enough to produce net fusion energy reliably. SPARC’s 2027 target is the most credible near-term candidate. If it works, the late 2020s can check that off.

Tritium breeding then has to be demonstrated in a real fusion neutron environment — a device achieving commercial-scale neutron flux, producing measurable tritium from lithium blankets above the 1.05 ratio. No such demonstration exists. ITER’s design includes a breeding test program, but ITER won’t reach D-T operations until 2039. Private companies aiming for commercial operation before then need their own path to this demonstration, on a timeline no one has yet solved. There’s no workaround and no shortcut: you cannot design a commercial plant around a tritium supply process you haven’t proven produces tritium.

IFMIF-DONES or equivalent must produce qualified materials that survive commercial neutron flux. Beam commissioning starts in 2029; years of data collection follow. Mid-2030s is the optimistic scenario for materials qualification. This timeline is not responsive to funding levels — the neutron bombardment that qualifies a material takes however long it takes, and the data must accumulate in real time, in a real neutron environment, with no computational shortcut available.

After that, a demonstration power plant must produce electricity at competitive wall-plug Q and operate long enough to validate the full system — tritium breeding in a live reactor, activation management, shielding, the factor of several hundred in Q that still exists between the best laboratory result and a competitive grid product. This is the step that converts scientific milestones into an energy technology. CFS targets ARC in the mid-2030s. Whether that’s achievable or whether it moves by another decade will depend on how the preceding steps go.

None of those steps is physically impossible. But they are sequential. Private capital controls the first. It has limited leverage over the rest.

The joke, revisited

Artsimovich said fusion would be ready when society needs it. Private capital has changed the structure of patience that quote implies. Companies in Devens, Massachusetts and Malaga, Washington are operating on timelines with penalty clauses and investors who won’t wait indefinitely. The institutional architecture that converted difficulty into schedule extension — the ITER model — is no longer the only model in play.

But the quote’s harder implication survives. “When society needs it” implies a minimum stack that urgency cannot compress: testing infrastructure with physical timelines no PPA shortens, tritium supply constraints that don’t yield to investor pressure, materials qualification processes that require data you cannot generate until the facilities to generate it are running. What private capital has changed is the accountability structure and the pace of work on the plasma physics. Not the physics of the remaining obstacles, and not the lead times of the infrastructure required to overcome them.

The joke about fusion being twenty years away has been true in every decade since the 1950s. What makes this moment different is not that the joke is finally false. It’s that the joke is, for the first time, actually testable. By the mid-2030s, SPARC will either have demonstrated Q > 1 in a magnetic confinement device or it won’t. IFMIF-DONES will either be producing materials qualification data or it won’t. Helion will have delivered power to Microsoft’s grid, or it will have missed a contractually binding deadline with financial penalties. Prior eras could extend the timeline and call it science. This era cannot. The 2030s will produce answers — not projections, not milestones, not revised baselines.

Whether those results will make the joke obsolete, or simply give it a more precise definition, is genuinely unknown.

Gen AI Haftungsausschluss

Einige Inhalte dieser Seite wurden mit Hilfe einer Generativen KI erzeugt und/oder bearbeitet.

Medien

Plasma of 100 million degrees is confined in a vacuum. This is a view of the vacuum vessel from the inside. Superconducting coils are placed on the other side of the twisted vacuum vessel wall, and the strong magnetic field created by the coils floats the plasma so that it does not come into contact with the wall. – Wikipedia

Wichtige Quellen und Referenzen

John D. Lawson, “Some Criteria for a Power Producing Thermonuclear Reactor,” Proceedings of the Physical Society B, Vol. 70, 1957.

U.S. General Accounting Office, “National Ignition Facility: Management and Oversight Failures Caused Major Cost Overruns and Schedule Delays,” GAO/RCED-00-271, August 8, 2000. gao.gov/products/rced-00-271

M. Kovari, R. Kemp, H. Lux, P. Knight, J. Morris, D.J. Ward, “Tritium supply and use: a key issue for the development of nuclear fusion energy,” Nuclear Fusion, Vol. 58, No. 2, 2018.

Daniel Clery, “Giant international fusion project is in big trouble,” Science (AAAS), July 2024.

U.S. Department of Energy / NNSA, “DOE National Laboratory Makes History by Achieving Fusion Ignition,” December 13, 2022. https://www.energy.gov/articles/doe-national-laboratory-makes-history-achieving-fusion-ignition

Commonwealth Fusion Systems, “US Department of Energy Validates Commonwealth Fusion Systems’ Successful Completion of Magnet Technology Performance Test,” September 30, 2025. https://cfs.energy/news-and-media/us-department-of-energy-validates-commonwealth-fusion-systems-completion-of-magnet-tech/

“Helion and Microsoft announce fusion energy agreement,” CNBC, May 2023.

“Helion Energy breaks ground on Orion fusion plant in Malaga, Washington,” S&P Global / NucNet, July 2025.

Fusion Industry Association, “The Global Fusion Industry in 2025,” FIA, 2025.

Roche, T., et al., “Generation of field-reversed configurations via neutral beam injection,” Nature Communications, Vol. 16, Article 3487, April 12, 2025. https://doi.org/10.1038/s41467-025-58849-5

“The fuel supply quandary for fusion reactors,” Bulletin of the Atomic Scientists, November 2024.

EUROfusion, “Testing fusion materials with a hail of neutrons,” EUROfusion project documentation, 2023–2024.

Windridge, Melanie, “Fusion: Ready When Society Needs It,” IAEA Bulletin, Vol. 62, No. 2, May 2021. https://www.iaea.org/bulletin/fusion-ready-when-society-needs-it

Commonwealth Fusion Systems, “CFS delivers its first fusion magnet — a stronger, smaller design,” The Tokamak Times (CFS blog), January 6, 2026. https://blog.cfs.energy/cfs-delivers-its-first-fusion-magnet-a-stronger-smaller-design/

Fusion for Energy, “DONES Programme collaboration takes shape,” April 11, 2024. https://fusionforenergy.europa.eu/news/dones-programme-collaboration-takes-shape/

Behr, Peter, and Christa Marshall, “Startup begins work on US fusion power plant. Yes, fusion,” E&E News (Energywire), July 31, 2025. https://www.eenews.net/articles/startup-begins-work-on-major-us-fusion-power-plant-yes-fusion/