Fusion's industrial path after the physics

Public attention to fusion follows physics milestones: ignition, record plasma temperatures, and the promise of net energy gain, the point at which a plasma releases more fusion energy than the heating that sustains it. These milestones matter, but they describe a laboratory result. A power plant must also convert heat to electricity, run for months with high availability, breed its own fuel, survive years of neutron bombardment, win a licence, connect to a grid, and sell power at a price a buyer will pay. Each requirement is a separate engineering and institutional problem, and several are less mature than the plasma physics.

The industry now behaves accordingly. Commonwealth Fusion Systems (CFS) has applied to the PJM grid operator for interconnection of a planned 400-megawatt plant in Virginia, signed offtake agreements with Google and Eni, and joined the UK's LIBRTI programme to test tritium blanket technology, all while its demonstration machine is still under construction (Utility Dive 2026). Helion is building a plant for Microsoft before its prototype has published a measure of net energy gain. Money, offtake, and permits are moving ahead of the technology that must justify them.

The useful frame is a chain with seven links: high-temperature superconducting tape, magnets, the machine, tritium, the blanket, the licence, and the grid connection and customer. Net energy gain tests one link. A plant is as strong as the least mature of the seven, and the weakest links are in materials and fuel, not in plasma physics.

For decades fusion was a public programme. ITER, the international tokamak under construction in France, is its benchmark, and its record shows how long industrialisation takes. Its 2024 baseline set the start of research operations in 2034, full magnetic energy in 2036, and deuterium-tritium operation in 2039, three to four years later than the 2016 reference, with an estimated 5 billion euros added to a budget already the largest in experimental science (Physics World 2024). By September 2026 six of nine vacuum vessel sector modules were in the tokamak pit, a three-year thermal shield repair had finished, and sector module 9, due in the pit in early December, was the critical path (ANS 2026).

The private wave began in the 2010s on a different premise: that high-temperature superconducting magnets would let a tokamak be much smaller and therefore much cheaper and faster to build. The Fusion Industry Association's 2026 survey counts 56 companies, up from 23 in 2021, with 28 in the United States and four each in the United Kingdom, Germany, and China. Companies raised 4.48 billion dollars in the 12 months to July 2026, a record and 69 percent above the prior year, taking cumulative funding since 2021 to 14.24 billion dollars (FIA 2026). Five companies have raised more than 1 billion dollars, all American.

The stated expectation is the 2030s: 71 percent of companies expect their first plant to deliver commercial electricity in that decade, and respondents estimate that a plant needs on average 2.7 billion dollars to become commercially viable (FIA 2026). The distance between that figure and the amounts raised shows that most of the industry is still funding the first plant, not a fleet.

The leading programmes make different bets. CFS raised 1 billion dollars in a round announced on 30 July 2026, taking total funding to 4 billion dollars, with pension funds and sovereign wealth funds among the investors. Chief executive Bob Mumgaard said SPARC is about 80 percent complete and the company expects to show net energy gain in 2027, and that a string of ARC plants would need more than 1 billion dollars with 10 billion dollars the scale of capital needed to pursue fusion seriously (Utility Dive 2026). ARC, a 400-megawatt plant planned for Chesterfield County, Virginia, is targeted for the early 2030s, with Google contracting 200 megawatts and Eni more than 1 billion dollars of power.

Helion pursues a pulsed approach intended to convert energy more directly. The company began construction of its Orion plant in Malaga, Washington in July 2025 under a power purchase agreement with Microsoft targeting 2028 delivery of at least 50 megawatts (S&P Global 2025), and reportedly became the first company to receive licences to build and operate a fusion power plant, from the Washington State Department of Health in June 2026 (Industrial Sage 2026). In the sources reviewed, Helion has not published a fusion gain figure for its prototype, so the 2028 date rests on schedule rather than on demonstrated performance.

The private tokamak thesis rests on one material. High-temperature superconducting magnets made from REBCO tape generate much stronger fields than their low-temperature predecessors, and because fusion power rises steeply with field strength, a machine can shrink dramatically. That moves the constraint from physics to manufacturing. A prototype compact device is estimated to need about 10,000 kilometres of tape, while the whole world's annual production capacity has been a few thousand kilometres (Commercial compact fusion 2025).

The supply response is under way. A 2025 review counts roughly 15 high-temperature superconductor wire vendors with combined annual capacity above 5,000 kilometres of 12-millimetre-wide equivalent, with pulsed laser deposition producers making more than 3,000 kilometres a year and accounting for over half of global output (Commercial compact fusion 2025). Cost is the second gate. The Department of Energy's ARPA-E describes commercial tape at about 300 dollars per kiloampere-metre under fusion magnet conditions and says it must fall to roughly 10 dollars to be cost-competitive (ARPA-E n.d.). Sources put current prices between 100 and 300 dollars, and these figures are several years old.

Fusion demand is now what justifies the factories. The risk is that every private tokamak depends on the same few tape suppliers, the same cryogenic and quench-protection engineering, and a supply base in which a small number of producers hold most of the capacity.

Most near-term fusion designs burn deuterium and tritium, and tritium is almost nonexistent in nature. Civilian stocks are estimated at 25 to 30 kilograms and decline about 5 percent a year through radioactive decay. Canada is the largest civilian source, producing about 2 kilograms a year as a by-product of heavy-water reactor operation, and most of that output is earmarked for ITER over the next decade (FAS 2024). Starting a single commercial fusion reactor may require up to 10 kilograms, and operating a 1-gigawatt plant for a year may consume more than 55 kilograms.

The arithmetic is severe. A stock of 25 to 30 kilograms is enough to start only a handful of commercial plants even if none of it were committed to ITER, and none of it can be stockpiled without losing about a twentieth a year. The only way to a fleet is for plants to breed their own fuel, which means a lithium blanket that produces more tritium than the plant burns. No full-size blanket has been demonstrated. An industry expert from the Electric Power Research Institute put the point bluntly: blanket technology is immature globally (Utility Dive 2026). The US government's marginal supply from Tennessee's Watts Bar reactors costs about 50,000 dollars per gram, against about 30,000 dollars for Canadian tritium, and is reserved for military use (FAS 2024).

This makes tritium a chokepoint that does not yet appear in most investors' models. The FAS analysis argues that planned refurbishments of ageing Canadian reactors and ITER delays have kept open a window for several more years in which supply exceeds demand. After that window, plants that cannot breed fuel compete for a shrinking stock. Fusion's fuel logic is therefore the reverse of its marketing: the fuel is plentiful in principle, because lithium and deuterium are abundant, and scarce in practice, because the intermediate isotope is not made at scale.

Regulation is moving faster than technology. In the United States the Nuclear Regulatory Commission proposed a rule in February 2026 that regulates fusion machines as a category of particle accelerator under the byproduct material framework of 10 CFR Part 30, consistent with Section 205 of the ADVANCE Act of 2024, rather than under the licensing regime for fission reactors, with a comment period that closed on 27 May 2026 (Federal Register 2026). The choice avoids reactor-style licensing and its costs. The United Kingdom follows a similar logic, with the Environment Agency and the Health and Safety Executive rather than the nuclear regulator overseeing fusion, a draft national planning policy statement for fusion in 2026, and a state-backed STEP prototype at West Burton with first operations expected around 2040 (DESNZ 2026).

The grid is the less visible gate. CFS has submitted an interconnection application to PJM for ARC, which places a future fusion plant in the same queue as the generation and data-centre loads that have driven that market to its price cap (Utility Dive 2026). A plant that can generate for the grid still has to wait for the grid.

State competition shapes the outcome. China Fusion Energy Co., a state-backed company under the China National Nuclear Corporation, was launched in 2025 with initial capital reported at about 1.6 billion dollars from state-owned enterprises (Caixin 2025), and its BEST tokamak at Hefei is scheduled for completion by the end of 2027 with a target of roughly five-fold energy gain, a claim that is self-reported and has had limited independent verification (Interesting Engineering 2025). The contrast with the American approach is one of structure: private capital moves quickly and concentrates risk in a few firms, while a state programme can commit to a long build without a quarterly funding test.

The strongest objection is that the industrial gates are engineering and will be passed as they were in other technologies. High-temperature tape has fallen in price and rising demand is expanding capacity. Blanket designs are in test programmes, regulators have chosen a lighter framework than fission, and capital is plentiful. Net energy gain in 2027 would be a decisive proof of the physics, and each successful step would draw in suppliers faster than caution predicts. This view is credible, and the rapid drop in the funding challenge, reported as the top short-term obstacle by 67 percent of companies this year against 84 percent last year, supports it (FIA 2026). What it cannot supply is a blanket and a fuel cycle that nobody has shown at scale, or the evidence that structural materials can withstand years of neutron bombardment while keeping a plant available for most of the year. A demonstration plant that runs for weeks is a different asset from one that sells power with the reliability a utility or data-centre buyer expects, and the gap between them is where earlier energy technologies have lost the most time and money.

Three paths follow. In staged commercialisation, SPARC and Helion demonstrate their physics by 2027 to 2028, tape supply scales, and first plants in the early 2030s start on external tritium and add blanket demonstration plants, producing a few plants by the mid-2030s. In a fuel-constrained path, physics succeeds but tritium and blankets lag, so plants wait on supply and the industry consolidates around those with secured fuel. In a state-led path, a publicly funded programme, most plausibly China's, reaches an integrated plant first, and Western companies compete on cost and export. The signposts are observable: SPARC's published gain, a demonstrated breeding ratio, the NRC's final rule, ITER's sector welding, tape price and volume, and whether Orion delivers power in 2028.

The record supports a selective conclusion. Fusion reaches the grid first where a developer secures the whole chain: tape at an acceptable price, a fuel cycle that does not depend on a fixed stock, a blanket tested in parallel with the machine, a regulator with a workable framework, and a buyer with a grid connection. It stalls where a company proves its physics and then waits for fuel, materials, or a queue position. A first plant is likely to be a demonstration with imported tritium. Fleets depend on breeding.

For governments, the priority is the unglamorous infrastructure: tritium production and breeding test facilities, tape manufacturing, and regulatory certainty, rather than additional plasma experiments. For developers, the discipline is to treat fuel and blankets as the critical path and to secure offtake and interconnection on realistic dates. For investors, the safest exposure is in the enabling layers, tape, cryogenics, blanket materials, and tritium handling, which earn revenue whichever plant design prevails, while plant developers carry the binary risk.

Fusion's physics has become a necessary condition and no longer a sufficient one. The first company to sell electricity will have solved a supply chain, not only a plasma.