For decades the recipe for a fusion reactor barely changed: build it enormous. A tokamak traps a cloud of gas hotter than the Sun inside a magnetic field, and for most of fusion’s history the only way to make that field strong enough was to make the machine huge. That is why ITER, the giant international reactor going up in southern France, is the size of a cathedral and has been under construction for well over a decade.<\/p>\n
A company outside Boston bet the recipe was wrong. Commonwealth Fusion Systems, a spinout of MIT’s Plasma Science and Fusion Center, figured that if you could build a strong enough magnet, you could shrink the whole reactor around it. That bet is now a physical object. At a campus in Devens, Massachusetts, CFS is roughly 75% of the way through assembling a compact tokamak called SPARC, and it is aiming to do in 2027 something the giant public machine in France will not attempt until 2039: get more energy out of fusion than you put in.<\/p>\n
The reason tokamaks had to be giant comes down to the magnets. A stronger magnetic field lets you squeeze and hold the plasma in a smaller space, and old superconductors could only push the field so far before you had to compensate by making the machine bigger.<\/p>\n
CFS’s founders bet on a newer material, high-temperature superconducting (HTS) tape woven from a rare-earth compound called REBCO, which carries enormous current and produces a far stronger field. In 2021 the company and MIT proved it, running a model coil up to 20 tesla, a record for a magnet that size and, by CFS’s own measure, the strongest of its kind in the world by a wide margin.<\/p>\n
These magnets are not subtle objects. The record-setting prototype coil weighed about 10 tons and was wound from 168 miles (270 km) of superconducting tape for a single coil. Across the full machine, CFS’s tape order runs toward 10 million meters of the stuff, close to a round trip from Boston to Los Angeles. SPARC uses 18 of these D-shaped magnets, each around three meters tall, to cage the plasma.<\/p>\n
In a separate test in 2024, CFS ran the reactor’s central solenoid, the pulsed magnet that ramps current up and down, to a record stored energy of 3.7 megajoules<\/a>, which the company likes to describe as roughly the energy of five full-size pickup trucks doing 60 mph.<\/p>\n The magnets need something to wrap around, and that something is the vacuum vessel: a sealed, airless steel chamber shaped like a donut, where the plasma actually lives. SPARC’s vessel was built in two halves, each a 48-ton stainless-steel arc, because lowering the whole thing in one piece would have been a nightmare.<\/p>\n The first half arrived months ago. In May 2026, crews at Devens lowered the second 48-ton half into Tokamak Hall by crane and joined the chamber together, the milestone that pushed the machine to roughly 75% complete, according to CFS co-founder and chief science officer Brandon Sorbom.<\/p>\n Once the chamber is sealed, technicians line the inside with panels of tungsten, the metal with the highest melting point on the periodic table, because the plasma it has to face runs hotter than 100 million degrees Celsius. That is several times hotter than the core of the Sun, and the magnetic field’s only job is to hold that plasma away from the walls so it never touches the steel.<\/p>\n When it is all running, SPARC is designed to put out as much as 140 megawatts of fusion power in bursts of about 10 seconds. The first attempt to make plasma at all is targeted for 2027.<\/p>\n To understand why a startup hitting these milestones is a big deal, you have to look at the machine it is racing. ITER is the giant international fusion project in southern France, backed by 35 countries and assembled over decades. It is the closest thing fusion has to a flagship, and it is the reason the running joke about fusion being thirty years away refuses to die. ITER is not even the only big public machine in the mix; in Japan, the world’s largest operating tokamak<\/a> recently powered up its first upgraded systems for a run meant to feed lessons straight into ITER.<\/p>\n ITER has also kept slipping. Under its revised schedule, the project does not start research operations until 2034, does not reach full magnetic energy until 2036, and will not run on its actual fusion fuel, a deuterium-tritium mix, until 2039. That is a four-year slip on the previous plan, and it came with roughly 5 billion euros in extra cost. ITER’s own director-general, Pietro Barabaschi, has played down the old “first plasma” milestone as mostly symbolic and warned that fusion cannot arrive in time to solve today’s climate problems on its own.<\/p>\n That is the gap CFS is driving through. SPARC is a fraction of ITER’s size and privately built, and if it hits net energy in 2027 it would get there more than a decade before the public giant runs its real fuel. It is worth being honest about the asterisk, though. Fusion deadlines move, and SPARC’s have already moved: the machine was originally meant to operate in 2025. Anyone who tells you a fusion date is locked in has not been paying attention to fusion.<\/p>\n Bolting together a fusion reactor is not cheap, and CFS has raised the kind of money that makes the timeline plausible. The company has pulled in more than $2 billion since it spun out of MIT in 2018, including an oversubscribed $863 million Series B2 round<\/a> in August 2025, which it billed as the largest deep-tech or energy raise since its own $1.8 billion round in 2021. The backer list is a who’s-who: the Italian energy major Eni, Bill Gates’ Breakthrough Energy Ventures, Khosla Ventures, the Singapore fund Temasek, and Nvidia’s venture arm, among others.<\/p>\n Google is the name that says the most about where this is headed. The search giant has been an investor in CFS since 2021, and in June 2025 it went further, signing a deal to buy 200 megawatts of power from CFS’s first commercial plant and increasing its stake, in what Google billed as the biggest corporate deal to buy fusion power yet.<\/p>\n The reason a search-and-ads company is buying fusion power that does not exist yet is the same reason it is buying every other kind of clean baseload it can find: AI data centers are pulling electricity off the grid faster than anyone planned, and they want round-the-clock power that does not burn carbon. CFS leaned into that at CES in January, unveiling an AI “digital twin” of SPARC built with Siemens, Nvidia and Google to simulate the machine without cracking it open.<\/p>\n SPARC is not meant to power anything. Like ITER, it is a demonstration machine, built to prove the physics and the engineering work at all. The names are a nested joke: the successor plant is called ARC, for affordable, robust, compact, and SPARC stands for Smallest Possible ARC.<\/p>\n That successor is the one actually meant to sell electricity: a 400-megawatt fusion plant CFS plans to build at the James River Industrial Park in Chesterfield County, Virginia, just outside Richmond, targeting the early 2030s. If it works, it would be the first fusion plant anywhere to put steady power on a grid, enough for roughly 300,000 homes. CFS has already requested a connection to PJM, the largest US grid operator, which is the kind of unglamorous paperwork that signals a company genuinely intends to plug in.<\/p>\n SPARC and ARC are one bet in a much busier field. Bill Gates’ TerraPower is building a sodium-cooled reactor in Wyoming<\/a>, and in Ontario, the first Western grid-scale small modular reactor<\/a> just had its foundation set, both chasing the same late-2020s and 2030 window to put new clean power on the grid. Fusion and fission are different machines solving the same problem, and right now they are all racing the same calendar.<\/p>\n What is real this week is the hardware. There is a sealed 48-ton steel chamber sitting in a hall in Massachusetts, 18 record-class magnets going in around it on a roughly two-week delivery rhythm, and a couple of billion dollars riding on the schedule. Whether the plasma behaves the way the physics papers say it will is the part nobody can promise until they hit the button in 2027.<\/p>\n Fusion has a long, well-earned reputation for blowing past its own deadlines, and SPARC has already slipped once. But for the first time in a while, the joke about fusion being thirty years away has an actual machine bolted together to argue with, and a date attached that a private company, not a committee of governments, is on the hook to hit.<\/p>\n","protected":false},"excerpt":{"rendered":" For decades the recipe for a fusion reactor barely changed: build it enormous. A tokamak traps a cloud of gas … <\/p>\nA 48-ton steel donut goes in by crane<\/h2>\n
The opening ITER’s delays left behind<\/h2>\n
Google, Gates, and more than two billion dollars<\/h2>\n
From a demo in Massachusetts to a power plant in Virginia<\/h2>\n