In 2025 two reactors held a piece of a star in place for over 20 minutes and produced no usable power. That paradox, not another unlimited-energy promise, is where fusion actually stands.
In February 2025 a reactor in the south of France held a churning ring of plasma, a gas hotter than the core of the Sun, steady for 22 minutes. It was a world record, and it produced not a single watt of usable power. That contradiction, a triumphant machine that generated no energy, is the truest picture of where fusion actually stands.
For as long as most people have been alive, fusion has been shorthand for a promise that never lands: limitless clean energy, always a few decades away. The story of 2026 is not another version of that promise. It is the quieter, more consequential news that the hardest problems in fusion have changed. The fundamental physics has cleared major milestones. The challenge now is turning those results into a machine that can survive, breed its own fuel and produce affordable electricity.
What actually happened
The record-breakers of the past two years all measured the same thing: endurance.
In January 2025 China's EAST tokamak held a high-confinement plasma for 1,066 seconds, more than doubling its own previous mark. Weeks later, in February, France's WEST reactor went further and sustained a plasma for 1,337 seconds, about 22 minutes. These are extraordinary feats of control. They are also, precisely, duration records rather than energy records: both machines were proving they could hold and steer the plasma, not extract power from it.
Above the national experiments sits ITER, the 35-nation megaproject in southern France meant to be the first machine to produce far more fusion energy than it consumes. In 2025 ITER completed the central piece of the world's largest pulsed superconducting magnet, a 13-tesla solenoid as tall as a building. In the same stretch it confirmed the cost of ambition: full operation on deuterium and tritium fuel has slipped to 2039, and the budget has climbed by roughly 5 billion euros. Meanwhile the private sector is sprinting on a different clock. Fusion companies raised a record 4.5 billion dollars in 2025; Commonwealth Fusion Systems, a spin-out from MIT, aims to show net energy gain around 2027 and is building a 400-megawatt plant in Virginia, and Helion has promised to deliver 50 megawatts to Microsoft by 2028.
The map is genuinely global, which is what makes these benchmarks trustworthy. Japan and Europe switched on JT-60SA, the largest tokamak now operating, in 2023; South Korea's KSTAR has held plasmas above 100 million degrees; and Britain's JET, before it retired in 2023, still holds the record for energy released from a fusion reaction. Add the laser-driven route at the United States National Ignition Facility and you have four continents attacking the same problem from different angles.
Why the hard part moved
For decades the central question was physics: can you confine a plasma hot enough and stable enough for fusion to happen. After the 2025 endurance records and a landmark 2022 result at the United States National Ignition Facility, that question is close enough to answered that the field has moved on. The bottleneck is now the plumbing of an actual power station, and it comes down to three walls.
Heat exhaust. The surface that catches the plasma's exhaust, the divertor, must survive a heat load around 20 megawatts per square metre, a blowtorch that never switches off, for years without failing.
Fuel. The reaction runs on tritium, a form of hydrogen that barely exists in nature. A real plant must breed its own tritium inside its walls, a self-sufficiency no machine has yet demonstrated.
Neutron damage. Fusion floods everything around it with neutrons that embrittle ordinary metals, so the reactor must be built from radiation-tough materials and maintained by robots, because people cannot go inside.
This is why the new heroes of fusion are not plasma physicists but magnet engineers, metallurgists and roboticists. The clearest example is the magnet. High-temperature superconductors let Commonwealth Fusion Systems reach 20-tesla fields and shrink the whole machine, which is how a start-up hopes to beat a global consortium to net gain.
None of this is only a hardware problem. Tritium, the reactor's fuel, is so scarce that the world's civilian stockpile is measured in kilograms, which is why breeding it inside the machine is not a nicety but a precondition for fusion to exist at scale. Solve the wall and the magnet, and the fuel supply is still waiting behind them. The question stopped being can we make a star. It became can we build a machine that survives one, and pays for itself.
Why this matters
You get a better scoreboard. Stop judging fusion by plasma temperature or a one-off record. Judge it by engineering readiness, full-system energy balance and credible build schedules. That is how you tell progress from press release.
Two clocks are running. ITER's slow public megaproject and a pack of venture-backed sprinters are betting on different timelines and different machines. Whoever is right, the late 2020s are when fusion stops being purely a science story.
The prize is real. A working fusion plant would be a source of firm, carbon-free power that runs day and night, the missing piece in a grid full of intermittent wind and sun.
The honest catch
No plant has made net electricity. Nowhere on Earth has fusion put more power onto a grid than it drew. The best energy gain any tokamak has managed is still below break-even, and the National Ignition Facility's celebrated result counts only the energy delivered to its fuel target, not the far larger amount its lasers consumed.
Records are not power. A 22-minute plasma is a control milestone, not a demonstration of usable energy. Confusing the two is the most common mistake in fusion coverage.
The sprint may not finish on time. Private targets for 2027 and 2028 are commitments and press releases, not delivered electrons, and the tritium and materials problems remain genuinely unsolved.
EDITOR'S TAKE
Fusion's reputation problem is that every generation is told it is thirty years away, and every generation is told the truth. What is new is the shape of those thirty years. The physics fights are largely won; what remains is a punishing engineering slog against heat, fuel and neutron damage, plus the money and patience to finish it. That is less romantic than a breakthrough, but it is more encouraging, because engineering problems, unlike deep mysteries, tend to yield to enough good people and enough cash. So stop waiting for one triumphant announcement and start watching the unglamorous scoreboard: energy in versus energy out for a whole plant, a wall that lasts a year, a reactor that breeds its own fuel. When those numbers move, the future arrives, quietly.
Quick questions
Has fusion produced net energy yet?
Not in any way that reaches a wall socket. One United States experiment produced more energy than was delivered to its fuel target, but far less than its machine consumed overall, and no fusion reactor has ever sent net electricity to a grid. The 2025 records were about holding a plasma for a long time, which is a different achievement.
If the physics works, why can't we build a plant now?
Because the surviving problems are engineering, not physics: a wall that can take a relentless blowtorch of heat and neutrons, a way to breed the reactor's own fuel, and the materials and robots to keep the machine running for years. Those are the walls the 2020s are being spent trying to climb.
Sources
ITER: the long-duration plasma records at EAST and WEST (2025).
CEA (WEST): the 1,337-second plasma record, about 22 minutes, on 12 February 2025.
Physics World: ITER's revised schedule to 2039 and the roughly 5 billion euro cost increase.
ITER facts and figures: the performance targets and the engineering systems behind them.
Related from Frontier Signal: our recent deep dive on why pharma is making medicines in orbit. Frontier Signal explains frontier technology in plain English. This is general information, not investment advice.
