Fusion in 2026: A Status Report

Fusion has been “thirty years away” for sixty years, which is the only thing most people remember about it. That was a fair joke for a long time. It is not a fair joke in 2026. The combination of credible scientific breakthroughs in the last four years, several private companies with real funding and serious physics, and a meaningful pull from hyperscale data center electricity demand has changed the field’s trajectory in ways even the field’s most cautious researchers concede.

This guide is the plain-English version. What the 2022 NIF result actually proved. Where the major public projects are. Which private companies are doing real work, and which are mostly raising. The timeline to fusion electricity on the grid, given everything we know in 2026. And the role AI is now playing in solving the problems that kept the field stuck for decades.

For the broader frame on energy and climate tech, the Energy and Climate Tech Frontier hub is the right starting point. This piece goes deep on fusion specifically.

What is fusion, in plain English?

Fusion is what happens when two light atomic nuclei get pushed close enough together to combine into a heavier nucleus. The combined nucleus weighs slightly less than the inputs did. The missing mass became energy, on the famous Einstein equation. Per kilogram of fuel, you get something like four million times the energy of burning coal and four times the energy of fissioning uranium.

The sun does this naturally with hydrogen at the temperatures and pressures of a stellar core. We do not have a stellar core to work with, so the engineering problem is to make a much smaller and hotter plasma that fuses anyway. Two main reactions matter at human scales. The first uses deuterium (a hydrogen isotope with one neutron) and tritium (a hydrogen isotope with two neutrons), produces helium plus a high-energy neutron, and works at roughly 150 million degrees Celsius. The second uses deuterium and helium-3, produces a proton (no neutron), and needs roughly 600 million degrees. The second is harder. It is also cleaner: no high-energy neutrons hammering your reactor walls.

The engineering challenge is confinement. You have to hold the plasma at fusion temperatures long enough for enough reactions to occur that the energy out is greater than the energy you spent confining it. There are three families of ways to do this, and the entire fusion industry organizes around which family you pick.

What was the December 2022 NIF ignition shot, really?

On December 5, 2022, the National Ignition Facility at Lawrence Livermore National Laboratory fired 192 lasers simultaneously at a peppercorn-sized capsule of frozen deuterium-tritium fuel. The capsule imploded, briefly creating conditions hot and dense enough for fusion to occur. The reactions produced 3.15 megajoules of energy from 2.05 megajoules of laser energy that hit the capsule. For the first time in history, a fusion reaction produced more energy than was used to drive it.

That is the headline. It is real, it is historic, and it deserved the press it got. The careful caveat is that “energy used to drive it” referred to the laser energy delivered to the capsule, not the wall-plug electrical energy that produced the laser pulse. NIF’s lasers are inefficient by design (they were built in the 1990s for nuclear weapons stockpile research, not power generation). The wall-plug energy for that shot was roughly 300 megajoules. So the system gain was about one percent, not greater than one. That is scientific gain. Engineering gain, where the wall-plug energy out exceeds wall-plug in, has not been demonstrated by any approach as of 2026.

What changed with NIF’s 2022 result is not that fusion is now a power source. It is that the basic physics of “ignition” (where the fusion reaction itself produces enough energy to sustain further fusion) has been demonstrated. That removed a category of doubt that had hung over inertial confinement for forty years. NIF has repeated and exceeded the 2022 result several times since, with the largest reported yields above 5 megajoules. The follow-on question, which is the entire current research program, is how to translate the physics into an economic power plant.

What are the three main approaches?

Magnetic confinement uses powerful magnetic fields to hold a hot plasma in a doughnut shape (a tokamak) or a more elaborate twisted shape (a stellarator). The plasma circulates inside the magnets at fusion temperatures, never touching the walls. ITER is the most famous example, the Joint European Torus (JET) is the previous generation, and Commonwealth Fusion Systems and Tokamak Energy are private-sector heirs.

Inertial confinement compresses a tiny fuel pellet so quickly that its own inertia holds it together for the brief moment of fusion. NIF uses lasers to drive the implosion. First Light Fusion uses a high-velocity projectile. The advantage is that you only need to confine the plasma for nanoseconds. The disadvantage is repeatability: a power plant needs to fire pellets several times per second, every second, for years.

Alternative confinement is the catch-all for everything else. Field-reversed configurations (TAE, Helion). Sheared-flow stabilized Z-pinches (Zap Energy). Magnetized target fusion (General Fusion). Magnetic mirrors (Realta Fusion). Each of these picks a different point on the cost/risk/timeline curve, and several of them did not look promising five years ago and look meaningfully more promising now thanks to better diagnostics, faster electronics, and machine-learning-assisted plasma control.

Where is ITER, the big public project?

ITER is the international tokamak under construction in Cadarache, France. Thirty-five member countries. A construction budget that has crossed $25 billion. A device the size of a 12-story building. ITER’s purpose is not to produce electricity. It is to demonstrate that magnetic confinement at industrial scale can produce ten times more fusion energy than it consumes in plasma heating (Q greater than 10) on long-duration shots.

ITER has been delayed multiple times. First plasma, originally scheduled for 2016, then 2025, is now targeted for the early 2030s. Deuterium-tritium operation, where the real fusion physics is tested, is later in that decade. The reasons for the delays are well-documented and not mysterious: international procurement complexity, component manufacturing failures, and the inherent difficulty of building something this large for the first time. The science of ITER is sound. The schedule has been the problem.

The fair read in 2026 is that ITER will eventually do what it was designed to do, and the field will learn enormously from it. The era when ITER was the only serious play in fusion is over. The private-sector race is now where most observers expect the first net-energy-on-the-grid demonstration to come from, and ITER’s role has shifted to high-quality public-sector science that informs the whole field.

Which private companies are doing real work?

The 2026 field has roughly a dozen private fusion companies with credible funding and credible physics. The ones to know:

Helion Energy is the company with the most attention-getting commercial commitment. In May 2023, Microsoft signed a power purchase agreement to buy 50 megawatts of fusion electricity from Helion starting in 2028. The agreement is binding (with the standard caveat that a fusion plant must actually exist for power to flow). Helion uses a field-reversed-configuration design that produces electricity directly from the plasma’s expansion against magnetic fields, skipping the steam-cycle heat engine that every other approach needs. The technology is technically the most ambitious choice in the field. The timeline is the most aggressive. Watch this one closely.

Commonwealth Fusion Systems is the MIT spinout building SPARC, a high-temperature-superconductor tokamak in Devens, Massachusetts. SPARC is sized to demonstrate Q greater than 2 (twice as much fusion energy out as plasma heating in) on real deuterium-tritium fuel. The bet is that high-temperature superconductors let you make a much smaller, much cheaper tokamak than ITER, with the same physics. First plasma is targeted for 2026, net-energy operation shortly after. If it works, CFS’s follow-on machine ARC is the commercial product. The company has raised more than $2 billion and is the most-funded private fusion company.

TAE Technologies (formerly Tri Alpha Energy) is the oldest private fusion company, founded in 1998. Their approach is field-reversed configuration aimed at the harder helium-3 reaction, which produces no neutrons and is cleaner if you can make it work. TAE has been quieter in the press than Helion but has its own research reactor (Norman) and serious physics output. The bet is long: aneutronic fusion is the most attractive end-state, and TAE thinks the engineering is solvable on a longer timeline.

Tokamak Energy is the UK company building compact spherical tokamaks with high-temperature superconductors. Their ST40 reactor reached 100 million degrees Celsius in 2022, which is the temperature required for commercial fusion. The follow-on demonstration plant is targeted for the late 2020s.

Zap Energy uses a sheared-flow stabilized Z-pinch, a design that fits in a room and skips the giant magnets entirely. The physics is fundamentally different from a tokamak. Their FuZE-Q reactor is producing real data. Their bet, like several others on this list, is that an unconventional confinement approach can leapfrog the tokamak path on cost.

General Fusion is the Canadian company pursuing magnetized target fusion. The reactor injects a magnetized plasma into a swirling liquid metal vortex, which is then compressed by pistons. The approach trades plasma confinement complexity for mechanical complexity. Their Lawson Machine 26 is operating.

First Light Fusion is the UK inertial-confinement startup that uses a hypersonic projectile to drive implosion instead of lasers. The advantage is engineering simplicity. The disadvantage is yield: projectile-driven implosions have not yet reached the gain figures laser drivers have.

Honorable mentions: Realta Fusion (mirrors), Avalanche Energy (very small modular fusion), Marvel Fusion (laser fusion, Germany), and a handful of smaller programs each doing one specific thing well.

What does Q mean, and what does “break-even” actually mean?

Q is the ratio of fusion energy produced to energy input. Multiple flavors of Q exist, which is what makes the breakdown confusing.

• Q-plasma is fusion energy out divided by plasma heating energy in. NIF reached Q-plasma greater than one in December 2022 (3.15 MJ out, 2.05 MJ in). ITER is designed for Q-plasma of ten.
• Q-engineering is fusion energy out divided by wall-plug energy in (including inefficient lasers, magnets, refrigeration, control systems). NIF’s December 2022 shot was about Q-engineering of one percent. A commercial fusion plant needs Q-engineering well above one, with most analysts saying you want above ten for reasonable economics.
• Q-commercial is the implicit ratio at which fusion competes economically with other electricity sources. This is sometimes called Q-grid. It depends on capital cost, fuel cost, and capacity factor.

Most press coverage uses “break-even” to mean Q-plasma greater than one. The careful version of “break-even” is Q-engineering greater than one. The version that matters for the world is Q-commercial greater than one. Several private companies are now openly targeting Q-engineering greater than one in the late 2020s. None has demonstrated it yet.

When does fusion electricity reach the grid?

The straight answer: the first kilowatts of fusion-generated electricity feeding a grid could happen as early as 2028 (Helion’s stated target) or as late as the late 2030s. The first gigawatts at competitive cost, which is the version that actually matters for displacing fossil generation, is most likely the 2035-2045 window. Anyone claiming more precision than that is overconfident.

The reasons the range is so wide come down to which approach turns out to be most tractable. If Helion’s field-reversed-configuration plus direct electricity conversion works, the timeline collapses dramatically because the engineering path is much shorter than a tokamak. If the tokamak path (CFS, Tokamak Energy) is the one that scales, the timeline extends because tokamaks have to be paired with conventional steam turbines and complex tritium-breeding blankets. If both work, the world gets fusion much faster than any single timeline assumes.

The case for taking the optimistic end of the range seriously: private capital invested in fusion has crossed $7 billion as of late 2024, the pace of physics-paper publication has accelerated, and several engineering milestones that were stuck for decades have moved in the last three years. The case for the pessimistic end: every approach still has unsolved materials problems for plasma-facing components, and every fusion plant ever proposed has to be paired with grid-scale supporting infrastructure that takes years to permit and build.

What are the real bottlenecks?

• Materials. The walls of a fusion reactor get bombarded by 14 MeV neutrons that destroy ordinary structural materials over operational lifetimes. Tungsten composites, vanadium alloys, and silicon carbide composites are all candidates. None is yet a solved problem at commercial-plant scale.
• Tritium. Deuterium is plentiful (every gallon of seawater contains usable amounts). Tritium does not occur naturally in significant quantity and decays with a 12-year half-life. A deuterium-tritium fusion plant has to breed its own tritium from lithium-bearing blankets that surround the plasma. The chemistry and engineering of those blankets is solved on paper but not at scale.
• Plasma instabilities. Hot plasmas are inherently unstable. Edge-localized modes, disruptions, and other instabilities can damage reactor walls if not controlled in real time. This is the area where AI is making the biggest contribution.
• Capital cost. Fusion plants are likely to be capital-intensive even when the physics is solved. The economics versus cheap solar plus batteries, advanced fission, or geothermal will depend on capacity factor and on whether the plants can be built much smaller than ITER suggests.

How does AI fit in?

The most consequential AI contribution to fusion so far has been plasma control. In 2022, a DeepMind collaboration with the EPFL Swiss Plasma Center showed that a reinforcement-learning agent could control the magnetic coils of the TCV tokamak to shape the plasma in real time. Conventional control systems use carefully hand-tuned PID controllers for each plasma configuration. The reinforcement-learning controller learned to handle multiple configurations, including unusual shapes that human engineers had not stably operated. The paper appeared in Nature in February 2022 and is the cleanest single example of AI directly accelerating fusion engineering.

The second-most consequential AI contribution is materials discovery. The plasma-facing component problem (above) is in part a search problem across very large chemical and microstructural spaces. Machine-learning models trained on density-functional theory simulations can propose candidate materials orders of magnitude faster than exhaustive simulation. Several of the materials being tested for ITER and follow-on tokamaks were selected with AI assistance.

The third is operational data. Helion has been clear in its technical talks that part of why its development pace has accelerated is data-driven pulse tuning across thousands of plasma shots. That is a different use of AI than a Nature paper, but at the engineering scale it may matter more.

None of this means AI is the reason fusion is making progress in 2026. The progress is fundamentally about magnets, lasers, materials, and physics. AI is a meaningful accelerant of all those programs and a real source of optimism for the next decade.

Frequently asked questions

Is fusion the same as fission?

No. Fission splits heavy atoms (uranium, plutonium) into lighter ones. Fusion combines light atoms (hydrogen, helium) into heavier ones. Both release energy from mass conversion, but the engineering, fuel supply, waste profile, and safety properties are fundamentally different. Fission powers every nuclear plant on the grid today. Fusion powers none yet.

Is fusion safe?

Yes, by the standards that matter. There is no chain reaction that can run away the way a fission reactor’s can. The fuel inventory in any single plant is small. The high-energy-neutron activation of reactor structural materials is a real waste-disposal issue, but the volumes are far smaller than fission and the radioactive lifetimes are far shorter (decades, not millennia).

Where can I follow the field as a beginner?

The Fusion Industry Association publishes an annual report that is the cleanest single overview of private-sector progress. The IAEA fusion portal is the best public-sector single source. For deeper science, Nature and Physical Review Letters are the venues that matter. For company-specific news, the CFS, Helion, TAE, and Tokamak Energy blogs are the primary sources.

If fusion works, does climate change become solved?

No, but it gets much easier. Even a fully successful fusion industry takes decades to build out at the scale that displaces global fossil generation. Solar plus storage, advanced fission, geothermal, and grid-scale efficiency improvements will all keep mattering. Fusion in the 2030s and 2040s is the long-term tail of the decarbonization curve, not the short-term solution.

Why is fusion only happening now?

Three things changed at once. First, high-temperature superconductors became commercially available, making compact tokamaks economically possible. Second, computing power and AI made plasma control and materials discovery far better. Third, private capital found the field, accelerating engineering at a pace public projects could not. The 2026 fusion moment is the result of those three lines crossing.

The Beginners in AI position on fusion

Fusion is one of the cleanest examples of a technology where pro-technology and pro-human-first point in exactly the same direction. The climate problem is real. The energy demand of the world is rising. The combination of solar plus storage will do most of the work in the next twenty years, but the long tail of decarbonization needs a dispatchable baseload source that is not a fossil fuel and is not constrained by uranium supply. Fusion is the cleanest answer the physics gives us.

The pro-human-first part is what fusion makes possible. A world with abundant, clean, and economically rational electricity is a world with cheaper desalination, cheaper green hydrogen, cheaper direct air capture, cheaper data centers running the AI we want to use, and cheaper everything that runs on watts. That is a world where the next generation has more options, not fewer. The case for fusion is not just thermodynamic. It is moral.

What we recommend to a curious beginner reading on a Saturday morning: follow this field. Read the Fusion Industry Association annual report. Watch the SPARC first-plasma news. Track Helion’s path against its Microsoft 2028 commitment. Hold the excitement and the skepticism at the same time, and remember that the fair version of “fusion is thirty years away” in 2026 is not a joke anymore.

Sources

• Lawrence Livermore National Laboratory NIF ignition program
• ITER official site
• Commonwealth Fusion Systems
• Helion Energy
• TAE Technologies
• Tokamak Energy
• Zap Energy
• General Fusion
• First Light Fusion
• Fusion Industry Association annual report
• IAEA fusion portal
• Degrave et al., “Magnetic control of tokamak plasmas through deep reinforcement learning.” Nature, February 2022. nature.com/articles/s41586-021-04301-9
• Abu-Shawareb et al., “Achievement of target gain larger than unity in an inertial fusion experiment.” Physical Review Letters, 2024 (the NIF Dec 2022 result paper).

You might also like

• The Energy and Climate Tech Frontier, our master hub for fusion, fission, solar, batteries, and the broader energy transition
• Brain-Machine Interfaces in 2026, the parallel frontier in neuroscience
• Figure 02 vs Optimus vs Digit, the humanoid race head-to-head
• If the Mammoth Comes Back in 2032, our biotech thought experiment
• General Technology: The Beginner’s Tour, our cross-domain technology hub
• AlphaFold, the other AI-in-science breakthrough worth knowing