In January 2026, Commonwealth Fusion Systems installed the first of 18 superconducting magnets for SPARC, its experimental tokamak currently being assembled in Devens, Massachusetts. Six months later, the reactor was approximately 75% constructed. This is not a victory press release, it is an industrial schedule holding firm, with suppliers, signed purchase contracts, and a Department of Energy report published June 9, 2026 that precisely identifies what remains to be solved. Nuclear fusion is no longer a physicist’s promise. It has become an engineering project under delivery pressure.

For sixty years, nuclear fusion answered to the same joke: it will be available in twenty years, and always has been. What 2026 reveals is that the timeline has changed in nature. SPARC’s first plasma is now targeted for 2027. The sector has raised $11.4 billion in private capital. Eni and Google have signed power purchase agreements for fusion electricity. The risk is no longer physical in nature; it is industrial.

The Essentials

  • First SPARC magnet installed in January 2026; tokamak approximately 75% assembled by June 2026; first plasma targeted for 2027
  • Commonwealth Fusion Systems has signed power purchase agreements with Eni and Google; Nvidia is a technology partner
  • DOE identifies in June 2026 two unresolved engineering bottlenecks: the neutron source and tritium breeding blankets
  • $11.4 billion in private capital invested in fusion worldwide as of June 2026, with the majority in the United States
  • ITER, the international public project based in France, faces structural delays; no European startup is in a comparable buyer position

The NIF Result Changed the Terms of the Debate

To understand why private billions are arriving now, we must return to December 2022. That month, the National Ignition Facility at Lawrence Livermore National Laboratory produced more energy from fusion than the laser energy delivered to the target for the first time in controlled laboratory fusion, representing a genuine scientific gain, distinct from overall energy gain. The result was modest in absolute value—approximately 3.15 megajoules produced for 2.05 injected via laser—but it should be emphasized that the 2.05 MJ delivered to the target itself required approximately 300 MJ of electricity to be produced by the laser system. The experiment settled a question of principle that the scientific community had debated for decades in this specific context: controlled fusion can produce a net gain relative to the energy deposited on the target. The physics works.

This demonstration had an immediate effect on venture capital markets. Before 2022, investing in fusion amounted to betting on undemonstrated science. After, the residual risk became one of engineering and cost, not fundamental feasibility. This shift in category, from speculative physics to industrial engineering, is what opened the floodgates of private financing. The $11.4 billion raised globally in 2026 cannot be understood without this precedent.

Commonwealth Fusion Systems, a MIT spinoff founded in 2018, made the central technological bet of this decade: high-temperature critical superconducting magnets, capable of producing magnetic fields of unprecedented power (20 teslas) in a compact volume. The HTS magnet developed with MIT was demonstrated in 2021. SPARC, the tokamak that results from it, is designed to be approximately ten times smaller than ITER while aiming at the same objective: a plasma that sustains itself and produces net energy. The first magnet installed in January 2026 is not a symbolic stone. It is piece number 1 of a machine for which 18 are necessary to achieve the targeted performance.

Buyers Who Sign Before the Machine Runs

What distinguishes the current cycle from previous ones is not only the volume of capital raised, it is the nature of the commitments. Eni, the Italian energy group, and Google have signed power purchase agreements with Commonwealth Fusion Systems. These contracts have no value if SPARC does not work, and both buyers know this. They are making an industrial bet on a technology that has not yet produced a commercial watt.

This type of commitment is structurally similar to the renewable electricity purchase agreements that enabled financing of the first offshore wind farms in the 2000s-2010s. Without a guaranteed buyer, no developer could raise funds to build. With a long-term contract, demand risk disappeared, and financing became possible. Fusion is following the same path, with one difference: the contracts are signed not after an industrial demonstration, but before. This is a greater risk-taking, and a stronger signal of buyer conviction.

Nvidia entered as a technology partner, without a purchase agreement but with a direct interest: data centers consume quantities of energy that are exploding with the scale-up of artificial intelligence. A source of decarbonized, dense, and controllable electricity would be a structural response to a constraint threatening the sector’s growth. The link between AI’s energy demand and industrial interest in fusion is not rhetorical; it is financial.

What the DOE Identified as Unresolved

The Department of Energy report published June 9, 2026 is the year’s most useful document on the subject precisely because it refuses smug optimism. It identifies two engineering bottlenecks that do not yet have demonstrated solutions at industrial scale.

The first concerns the neutron source. An operating tokamak produces high-energy neutrons, which is the very mechanism by which fusion releases energy. These neutrons traverse the reactor walls and must be captured to heat a coolant fluid, which produces steam, which drives a turbine. The problem is that these neutrons damage materials over time. No commercial facility has ever experienced this flux of fusion neutrons over a long duration, and data on material degradation at this scale remains limited. The DOE notes that the testing necessary to qualify materials cannot be conducted in current facilities.

The second bottleneck is the tritium blanket. D-T fusion, deuterium-tritium, the reaction most accessible in terms of temperature, consumes tritium as fuel. Tritium is rare, radioactive, and its global civilian production numbers in kilograms per year, which is insufficient to fuel a reactor fleet at scale. The theoretical solution is the breeding blanket: a layer of lithium around the plasma that, bombarded by fusion neutrons, regenerates the tritium consumed. In theory, the reactor produces its own fuel. In practice, no breeding blanket has yet demonstrated a regeneration rate greater than 1, meaning no system has yet proven it produces more tritium than it consumes.

These two problems are not insurmountable. But they are real, they are known, and the DOE report deserves to be read for what it is: a rigorous assessment of a technology that is advancing, not a list of problems blocking it.

The Calendar Shift and What It Says

SPARC was supposed to reach its first plasma in 2025. The objective is now 2027. Two years of slippage on a project of this complexity is not a warning; it is the norm in major nuclear engineering projects. What matters is not the gap from the initial schedule, it is the nature of progress: the tokamak is 75% assembled, magnets are being installed, contracts are signed.

The slippage reveals something more interesting than the delay itself. Commonwealth Fusion Systems chose not to further compress the schedule to meet a date. In a sector where announcements often precede achievements by decades, this form of discipline is significant. Investors who have committed private capital at nine or ten figures are not financing a date in a press release; they are financing documented technical progress. The fact that the schedule slips without financing withdrawing says something about stakeholder confidence in the actual trajectory.

ARC, the commercial reactor designed to succeed SPARC, is already in the design phase. If SPARC demonstrates first plasma in 2027 and achieves targeted performance in the following years, ARC could be operational in the 2030s. This is not a certainty; the two DOE bottlenecks will need to be resolved along the way. But the capitalistic trajectory is that of an industrial project, not a research program of indeterminate duration.

ITER and the European Question

While American startups are installing magnets in Devens, ITER continues to accumulate delays in Saint-Paul-lès-Durance. The international project, which France hosts and the European Union is the principal funder of, has seen its costs exceed 20 billion euros and its schedule slip by several years. The first plasma, originally planned for 2025, is now expected at best for 2035 according to the latest official estimates, with some internal sources mentioning 2039.

ITER is not a competitor to SPARC. The two projects have different objectives: ITER is an international scientific program intended to demonstrate fusion physics at large scale; SPARC is a private industrial project intended to demonstrate the viability of a compact tokamak. But ITER is funded largely by European public funds, and its delays pose a strategic question that no one in France seems eager to formulate clearly: Europe is investing billions in a scientific program whose industrial returns it will not capture if American startups arrive first on the commercial market.

No European startup is currently in a position to be a serious buyer or competitor in the race for commercial fusion. Initiatives exist—Proxima Fusion in Germany, Novatron Fusion in Sweden, Renaissance Fusion in France—but their capitalization remains incomparable to American financing. The question is not whether Europe can still position itself. It is whether it will do so before the window of industrial competitiveness closes, as it has closed in semiconductors, in space launchers, and in several segments of renewable energies.

The comparison with lithium is useful here. Like Chile with its mineral rent, Europe has a strategic asset: the location of ITER, decades of public research in plasma physics, laboratories of excellence, but risks not transforming it into an industrial position. Holding the resource is not enough. One must still decide to do something with it.

What Is at Stake Over the Next Ten Years

Fusion will not feed the electrical grid by 2030. But the industrial schedule of SPARC means that central technical questions will have experimental answers in the next five years. If first plasma is achieved in 2027 and targeted performance is confirmed, the two DOE bottlenecks will become R&D priorities for the entire industry. Tritium blankets and neutron sources are not theoretical problems; they can be worked on in parallel with reactor assembly.

What changes in this cycle, compared to previous ones, is the structure of incentives. When Eni signs a power purchase agreement for fusion electricity, it mobilizes its engineering teams to understand what they will buy and how to integrate it into their energy system. When Google commits, it has a direct interest in the bottlenecks resolving quickly; each year of delay is another year of energy bills for its data centers. This coupling between industrial buyers and R&D schedules did not exist ten years ago.

The great uncertainty remains the cost per kilowatt-hour. No actor publishes reliable projections at this stage, and estimates circulating in the sector vary by a factor of ten depending on assumptions. SPARC is not designed to be economic; it is designed to be demonstrative. ARC will be the first test of genuine economic competitiveness. Investment decisions in the 2030s will be made on data that SPARC will produce between 2027 and 2032.

The question for public decision-makers, not only in Europe but everywhere energy mix choices are negotiated over thirty years, is whether to integrate fusion into long-term scenarios before achieving technical certainty, or wait for certainty to arrive and lose the capacity to influence the technology. It is the same question posed by the first offshore wind turbines in the early 2000s: those who bet early built an industry; those who waited for proof bought from others.

Sources

  1. TechCrunch, Commonwealth Fusion Systems installs reactor magnet, lands deal with Nvidia (January 2026)
  2. DOE Fusion S&T Roadmap official (June 9, 2026)
  3. DOE, National Laboratory Makes History Achieving Fusion Ignition (December 2022)
  4. Physics World, ITER fusion reactor hit by massive decade-long delay and €5bn price hike
  5. CFS + Eni, official press release (September 22, 2025)
  6. IEEE Spectrum, CFS SPARC 75% complete (June 2026)
  7. ITER, location in France (official iter.org)
  8. DOE Fusion Energy Strategy 2024
  9. Department of Energy, Report on fusion engineering bottlenecks, June 9, 2026 (published by DOE, no stable URL available)
  10. National Ignition Facility, Lawrence Livermore National Laboratory, results of the December 2022 ignition experiment, according to official laboratory publications
  11. Wikipedia — SPARC (tokamak), Commonwealth Fusion Systems (factual reference pages)
  12. Fortune — coverage of private fusion financing and Eni/Google contracts (2025-2026)