In the twelve months to July 2026, private fusion companies raised a record $4.48 billion, up 69% on the year before and lifting the sector's total to $14.24 billion since tracking began in 2021 [source: Fusion Industry Association, 2026]. One startup has promised to sell fusion electricity to Microsoft by 2028. Another says it will switch on a machine that produces more energy than it consumes in 2027. After decades of "always thirty years away," fusion suddenly has dates on the calendar.
So it is worth asking the question that the headlines usually skip: what, exactly, has been achieved, and what has only been promised? The honest answer requires a distinction that most coverage blurs. Producing more fusion energy than a plasma absorbs is a physics milestone. Delivering more electricity to the grid than a whole power plant consumes is a very different feat, and no one on Earth has done it yet. This article keeps those two ideas strictly apart, compares how the leading private companies are trying to close the gap, and lays out the engineering walls that still stand between a demonstration and a power station. It is not investment advice or an endorsement of any company.
One more piece of housekeeping before the numbers start. Every claim in this piece falls into one of three buckets, and they are not interchangeable. The first is a measured, published result: a machine did a thing and the laboratory reported what its instruments recorded. The second is a company target or projection: a date or a performance figure a firm expects to hit, useful as a statement of intent and nothing more. The third is a commercial commitment: a contract or a groundbreaking, which proves a company is willing to sign, not that it can deliver. Where a figure below is a target rather than an achievement, it is labeled as one — and readers will find that most of the impressive dates belong to the second and third buckets.
Table of Contents
- Why fusion is suddenly a 2026 story
- The physics in one paragraph
- The money that changed the mood
- The distinction that decides everything: plasma Q vs engineering Q
- What NIF actually did in December 2022
- The number the celebrations skipped
- Eight ignitions, and what reproducibility proves
- Inertial versus magnetic: why the numbers do not transfer
- One goal, many machines: how the private approaches differ
- Commonwealth Fusion Systems: the bet on stronger magnets
- Helion: the bet on pulses and direct conversion
- The rest of the field, and the strain already showing
- ITER: the public giant resets the clock
- A research device, not a power plant
- What a decade of delay was actually made of
- Follow the money — and separate claims from verification
- What a valuation prices
- What still stands between the lab and the grid
- Why tritium is the hardest of the three
- Back to the gap that matters
- The bottom line: what to watch
- A checklist for the next announcement
Why fusion is suddenly a 2026 story
The physics in one paragraph
Fusion is the process that powers the sun: light nuclei such as the hydrogen isotopes deuterium and tritium fuse into heavier ones, releasing energy. Unlike the fission that runs today's nuclear plants, it splits nothing, produces no long-lived high-level waste chain, and cannot melt down. The catch has always been that forcing nuclei to fuse requires temperatures above 100 million degrees Celsius and a way to hold that plasma together long enough to gain more than you spend.
The money that changed the mood
What changed is money and momentum. The Fusion Industry Association's 2026 report, released in New York on July 13, counted a record $4.48 billion raised by private operators over the prior year, a 69% jump, bringing cumulative private and public funding to $14.24 billion and industry employment above 16,000 people [source: Fusion Industry Association, 2026]. That capital is chasing a wave of confidence: in the association's 2025 survey of 53 companies, 84% of respondents said they expect fusion on the grid before the end of the 2030s, and 53% by 2035 [source: Fusion Industry Association, 2025]. Those figures are expectations reported by the companies themselves, not independently verified schedules — a caveat worth carrying through everything that follows.
The size of the jump is easier to feel with the previous year beside it. The association's 2025 report counted $2.64 billion raised in the twelve months to July 2025, against a cumulative total of $9.766 billion at that point [source: Fusion Industry Association, 2025]. The population being surveyed has grown too: 53 companies responded in 2025, up from 23 when the association began counting in 2021 [source: Fusion Industry Association, 2025]. That growth is worth reading carefully. More companies raising more money is a measure of how much capital believes fusion is close, not a measure of how close it is.
The optimism is not baseless. Since 2022 there have been real, peer-reviewed physics results, and private machines that were paper designs a few years ago are now steel in the ground. But enthusiasm and proof are different currencies, and the rest of this piece is mostly about telling them apart.
The distinction that decides everything: plasma Q vs engineering Q
What NIF actually did in December 2022
In December 2022, the U.S. National Ignition Facility (NIF) at Lawrence Livermore National Laboratory did something genuinely historic. Its lasers delivered 2.05 megajoules of energy to a fuel capsule, and the fuel released about 3.15 megajoules of fusion energy — more out than went in [source: AIP FYI, 2022]. This is "scientific breakeven," or a plasma gain (Q) above 1, and it had never been demonstrated before.
It helps to be precise about what that ratio divides. Target gain — the number physicists usually mean by Q in an inertial-confinement experiment — is the fusion energy released by the fuel capsule divided by the laser energy delivered to that capsule. "Ignition" is the related idea that the fusion reactions themselves heat the fuel enough to keep the burn going rather than fizzling the instant the drive stops. Both are statements about what happens inside a peppercorn-sized target over a few billionths of a second. Neither says anything about the building around it, and that is the whole of the confusion that follows.
The number the celebrations skipped
Here is the part the celebrations tended to skip. Those lasers are only about 0.5% efficient, and to put 2 megajoules onto the target they drew roughly 300 to 400 megajoules from the electrical grid [source: Physics World, 2022]. Measured at the wall plug — the number that matters for making electricity — a target gain of 1.5 corresponds to an engineering gain of roughly 0.01. In other words, the facility consumed on the order of a hundred times more energy than the fusion released, once you count the whole machine. NIF was never built to generate power; it was built for nuclear-weapons stockpile science and to prove the physics of ignition [source: Science, 2022]. To be a power source, a plant would need a gain closer to 100, far more efficient lasers, and the ability to fire repeatedly rather than once.
That 0.5% figure is not a law of nature, and it is the one part of the gap that is plainly fixable. NIF's lasers are 1990s technology; modern laser architectures reach efficiencies of up to about 20% [source: Physics World, 2022]. Swapping in that generation of hardware would shrink the wall-plug penalty by a large factor without touching the plasma physics at all — which is exactly why inertial-fusion advocates treat the current engineering number as a statement about one aging facility rather than about the approach. It is also why the honest version of the sentence is not "fusion is a hundred times short" but "this particular machine, built for a different purpose, is a hundred times short, and no one has yet built the one that would not be."
Eight ignitions, and what reproducibility proves
NIF has kept going, and the physics has held up. By late May 2025 it had achieved ignition eight times, with a shot in April 2025 yielding a record 8.6 megajoules from 2.08 megajoules of laser energy — a target gain of 4.13 [source: LLNL Science & Technology Review, 2025]. That reproducibility matters: it turns a one-off into a phenomenon scientists can study and improve. But it does not move the wall-plug number, and NIF's laser-driven "inertial confinement" is a different route from the magnetic confinement most private companies use. The lesson to carry forward: whenever a company or lab cites a "gain," ask whether it means energy out of the plasma or electricity out of the building. Almost every impressive number you will read is the former.
The individual shots are more instructive than the headline. A February 2024 shot produced 5.2 megajoules [source: LLNL / National Ignition Facility, 2024]. The sixth ignition, in November 2024, came in lower at 4.1 megajoules; the seventh, in February 2025, reached 5.0 megajoules for a target gain of 2.44; the eighth, in April 2025, set the record at 8.6 megajoules with an uncertainty of plus or minus 0.45 [source: LLNL Science & Technology Review, 2025]. The line is not a smooth climb — it dips and then jumps — which is what learning to control a physical system looks like from the inside.
Inertial versus magnetic: why the numbers do not transfer
Two words are worth separating, because cross-company comparisons routinely conflate them. Inertial confinement, NIF's approach, compresses a tiny fuel capsule so violently with lasers that the fuel's own inertia holds it together for the instant it takes to burn; the machine works in single shots. Magnetic confinement, the route most private firms have chosen, uses powerful magnetic fields to hold a diffuse, much longer-lived plasma away from the walls of a vessel — tokamaks and field-reversed configurations are both variants of this idea. NIF is inertial; the majority of the private field is magnetic [source: Physics World, 2022]. That means a NIF record does not validate a tokamak, and a tokamak record would not validate NIF. Reproducing ignition is a physics milestone for the field's understanding of burning plasmas — it is not, by itself, a power plant, and it is not a result any magnetic-confinement company can put in its own column.
One goal, many machines: how the private approaches differ
The private sector is not converging on one design; it is racing along several. The main contenders and their bets:
- Commonwealth Fusion Systems (CFS) — a compact tokamak called SPARC that leans on high-temperature superconducting (HTS) magnets to make stronger magnetic fields in a smaller machine. CFS says SPARC was more than 60% built by early 2026, with first plasma targeted for 2026 and net energy gain (Q>1) for 2027; its own and published projections put SPARC's gain at roughly Q=8–11 on deuterium-tritium fuel [source: Fortune, 2026]. Its first commercial plant, ARC, is planned for Virginia, with construction around 2027–2028 and grid power in the early 2030s [source: Commonwealth Fusion Systems, 2026]. These are company targets, not achieved results.
- Helion Energy — takes a "field-reversed configuration" and a pulsed, magneto-inertial approach that aims to convert fusion energy to electricity directly. Its seventh prototype, Polaris, began operating in late 2024. Helion signed the world's first fusion power-purchase agreement with Microsoft in May 2023, promising at least 50 megawatts starting in 2028, and broke ground on its Washington-state plant in July 2025 [source: CNBC, 2023] [source: S&P Global, 2025]. The 2028 date is a contractual target, not a demonstrated capability.
- TAE Technologies — pursues a field-reversed configuration boosted by particle beams. It raised more than $150 million in June 2025 from investors including Chevron and Google [source: TAE Technologies, 2025].
- Tokamak Energy — a UK company building a compact "spherical" tokamak paired with HTS magnets, with about $335 million raised since 2009 [source: The Fusion Report, 2024].
- General Fusion — a Canadian firm pursuing "magnetized target fusion," a mechanical compression scheme, whose LM26 machine aimed at scientific breakeven by 2026 [source: TechCrunch, 2026].
The diversity is a strength — several independent shots at a hard problem — but it is also a reminder that no design has yet been proven at power-plant scale. Each approach still has to clear the same gap between a hot plasma and a machine that sells electricity.
Commonwealth Fusion Systems: the bet on stronger magnets
CFS is really a bet on one component. A tokamak's performance improves steeply with magnetic field strength, so a stronger magnet lets you shrink the machine — and a smaller machine is cheaper, faster to build, and easier to iterate. In 2021 the company demonstrated a 20-tesla-class high-temperature superconducting magnet, the piece of hardware the whole compact design rests on [source: Commonwealth Fusion Systems, 2026]. Everything downstream of that — SPARC being more than 60% built by early 2026, first plasma targeted for 2026, net gain targeted for 2027, a projected Q of roughly 8 to 11 on deuterium-tritium fuel, and the ARC pilot plant in Virginia with construction around 2027–2028 — follows from the magnet working at scale [source: Fortune, 2026]. Note the tiers stacked in that sentence: one demonstrated component, one machine under construction, and a chain of dates the company has set for itself.
Helion: the bet on pulses and direct conversion
Helion's wager is structurally different. Rather than sustaining a plasma and boiling water with the heat, it compresses a field-reversed configuration in repeated pulses and aims to recover electricity directly from the expanding plasma's magnetic field — skipping the steam cycle that every thermal power station depends on. Its seventh prototype, Polaris, began operating in late 2024, and the company broke ground in July 2025 on a plant it calls Orion, in Malaga, Washington, with operation targeted for 2028 [source: S&P Global, 2025]. The 2023 power-purchase agreement with Microsoft commits at least 50 megawatts from 2028, with Constellation acting as power marketer [source: Helion Energy, 2023]. A signed PPA with named counterparties is a real commercial fact — it just is not a physics result.
The rest of the field, and the strain already showing
The remaining contenders show how wide the spread of maturity is. TAE Technologies pairs its field-reversed configuration with neutral particle beams and reports a stable plasma at 70 million degrees Celsius in a device it calls Norm; its June 2025 round of more than $150 million drew Chevron, Google and NEA [source: TAE Technologies, 2025]. Tokamak Energy's $335 million since 2009 includes a $125 million round, modest next to the sums now moving in the United States [source: The Fusion Report, 2024]. General Fusion is the cautionary case: alongside its LM26 machine and its 2026 scientific-breakeven goal, it cut about a quarter of its staff in May 2025, raised $22 million in a pay-to-play round that August, and took a further $51.1 million through SAFE instruments in November 2025 [source: TechCrunch, 2026]. Emergency structures like those are how a company buys time when the milestones and the money are out of step.
ITER: the public giant resets the clock
The counterpoint to the nimble startups is ITER, the vast international tokamak under construction in southern France and backed by seven members: China, the European Union, India, Japan, Korea, Russia, and the United States. Its scientific aim is to produce 500 megawatts of fusion power from 50 megawatts of plasma heating — a plasma gain of Q=10 — though, crucially, that is thermal power in the plasma, not electricity to a grid; ITER is a research device, not a power plant.
A research device, not a power plant
ITER's headline number is routinely misread as a power rating. The Q=10 goal describes fusion heat produced in the plasma relative to the heat injected into it; it is a scientific gain figure, and it excludes everything the surrounding facility spends on magnets, cooling and controls. ITER also plans a second, less-quoted test: steady-state operation at Q=5, which trades peak gain for duration. That trade is the more relevant one for anyone thinking about power plants, since a grid does not want a spectacular pulse — it wants an output that stays up. Reading ITER correctly means holding both facts at once: it is designed to settle physics questions at a scale no private machine will reach this decade, and it is not designed to sell a single kilowatt-hour.
What a decade of delay was actually made of
ITER also illustrates the field's long habit of slipping schedules. In July 2024 the project adopted a new baseline that pushed the start of research operations to 2034 and full deuterium-tritium operation to 2039, four years later than the previous plan, at an added cost of about €5 billion [source: Physics World, 2024]. The reasons were mundane and instructive: pandemic-hit supply chains, defective components including cracks in thermal-shield cooling pipes, and a decision to swap the first-wall material from beryllium to tungsten [source: ITER Organization, 2024]. If the best-funded fusion project in history can lose a decade to engineering, it sets a sober baseline against which to read private-sector promises of grid power within a few years.
The list of causes repays a second look, because of what is missing from it. Alongside the supply-chain disruption, the cracked cooling pipes and the beryllium-to-tungsten switch, France's nuclear safety authority halted assembly work while component problems were resolved [source: ITER Organization, 2024]. Not one of those items is a failure of plasma physics. They are procurement, welding, regulatory review and materials selection — the ordinary machinery of building something large and radioactive. That pattern is the strongest available evidence for a claim made later in this piece: the binding constraint on fusion has been migrating out of the physics and into the engineering, and engineering delays are measured in years, not press releases.
Follow the money — and separate claims from verification
The investment surge is real, but so is the gap between raising capital and reaching a physics milestone. The blunt fact as of 2026 is that no private fusion company has yet demonstrated scientific breakeven — a plasma gain above 1 — let alone engineering breakeven [source: TechCrunch, 2026]. Every commercial timeline on offer therefore rests on results not yet in hand.
That tension is starting to show. Reporting in April 2026 described "cracks" in the funding boom: General Fusion laid off about a quarter of its staff in May 2025 and needed a $22 million emergency round that August, while some investors worried that companies were preparing to go public before hitting the milestones that would justify their valuations [source: TechCrunch, 2026]. None of this means fusion is a mirage; it means the market is pricing in outcomes that remain unproven. The disciplined way to read any announcement is to sort it into one of three buckets: a peer-reviewed physics result, a company claim not yet independently verified, or a commercial promise contingent on future milestones. Most of what drives the headlines sits in the second and third.
What a valuation prices
TAE Technologies makes the arithmetic unusually legible. Over roughly thirty years the company raised about $2 billion, and its valuation ahead of a merger was also about $2 billion — meaning early backers were, in aggregate, close to break-even on three decades of patience [source: TechCrunch, 2026]. Set that beside the fact that no private company has reached scientific breakeven and the shape of the risk becomes clear: valuations in this sector are being set by capital-market events rather than by machine performance. That is a reason to read a fundraising headline as information about investors, and a shot record as information about physics, and never to let the first stand in for the second.
What still stands between the lab and the grid
Even a machine that reaches plasma gain would not be a power plant. Several engineering walls remain, and by 2026 the field's own experts describe the bottleneck as having shifted from plasma physics to engineering.
The first is fuel. Most designs burn deuterium and tritium, but tritium is radioactive, decays quickly, and barely exists in nature; a fusion economy would have to breed its own inside the reactor. That means a "breeding blanket" achieving a tritium breeding ratio above 1.0 in a fully integrated system — something never validated at scale — because no independent large-scale tritium supply chain exists, with today's inventory coming as a byproduct of fission reactors and not being replenished at the rate a fleet of plants would need [source: Proxima Fusion, 2026]. The second is durability: the materials facing an intense neutron flux must survive continuous operation, and the dedicated neutron test source needed to qualify them does not yet exist [source: TechTimes, 2026]. The third is continuity itself — turning brief, pulsed shots into steady-state operation that runs for months.
Why tritium is the hardest of the three
Of those three walls, the tritium problem has the least room to maneuver, because it is a closed loop with no external supply to fall back on. A plant must breed more tritium than it burns — a breeding ratio above 1.0 — inside a fully integrated blanket that is simultaneously absorbing neutrons, generating heat and surviving them; that combination has never been validated at full scale [source: Proxima Fusion, 2026]. The durability wall has the same structure. Qualifying a material for years of neutron bombardment requires a dedicated neutron test source, and that facility has not been built — the U.S. Department of Energy's own materials roadmap frames these engineering gaps against a 2030s deadline [source: TechTimes, 2026]. Both problems share an uncomfortable property: they cannot be solved faster by making the plasma hotter.
Back to the gap that matters
All of these feed back into the distinction from earlier. Closing the gap from plasma gain to engineering gain means paying for the magnets or lasers, the cooling, the tritium plant, and the conversion losses — and still having net electricity left over. That is the real finish line, and it is further out than any single hot-plasma record suggests.
The bottom line: what to watch
Fusion in 2026 is a field of genuine progress and genuine hype, and the two are easy to confuse. The physics of ignition has been demonstrated and reproduced; record sums are flowing in; private machines are being built. At the same time, no private company has reached scientific breakeven, the wall-plug gap remains enormous, and the hardest engineering — tritium breeding, materials, continuous operation — is largely unsolved.
A checklist for the next announcement
So watch for the milestones that actually move the needle, not the ones that merely make headlines. Does a private machine reach a plasma gain above 1, confirmed by outside review? Does anyone demonstrate energy gain measured at the wall plug, not just at the target? Does a breeding blanket produce more tritium than it burns in an integrated test? Does Helion's plant deliver contracted power to Microsoft in 2028, or does that date move? And does SPARC's 2027 target survive contact with reality? The answers, not the announcements, will tell you how close the sun in a bottle really is.
Two habits make the next few years of headlines readable. The first is to ask which Q a number refers to before reacting to it — energy out of the plasma, or electricity out of the building — because the two differ by roughly a factor of a hundred at the one facility where both have been measured. The second is to ask which of the three buckets a claim belongs in: measured result, company target, or commercial commitment. What survives both questions is still remarkable: ignition is real, it has been repeated eight times, and the money is no longer speculative. What remains unproven is everything between that achievement and a wire carrying fusion electricity into a house.