Fusion Energy Startups Explained: Why Private Fusion Is Accelerating

Fusion companies raised a record $4.48 billion in the 12 months to July 2026, according to the Fusion Industry Association (FIA), bringing the total reported over six years to $14.24 billion. On July 30, 2026, Commonwealth Fusion Systems alone announced another $1 billion, lifting its total to $4 billion. None of these companies sells a watt of fusion electricity yet. This article explains what changed to make investors confident, how the competing reactor designs work, and which milestones are verified and which are still promises.

What fusion requires, in plain terms

Nuclear fusion is the process of forcing two light atomic nuclei, usually isotopes of hydrogen, to merge into a heavier nucleus and release energy. It powers every star, including the Sun. The reaction most companies are chasing combines deuterium and tritium (two heavy forms of hydrogen) into helium and a high-energy neutron. The U.S. Department of Energy notes that researchers focus on deuterium-tritium because it releases a large amount of energy and works at lower temperatures than other fuels (DOE).

"Lower" is relative. ITER, the international fusion project in France, says deuterium-tritium fusion in a laboratory needs about 150,000,000 degrees Celsius (270 million degrees Fahrenheit), roughly ten times hotter than the Sun's core at about 15,000,000 degrees Celsius (ITER). The Sun gets by on gravity, which squeezes its core hard enough to fuse hydrogen at lower temperatures. A reactor on Earth has no such gravity, so it compensates with heat. How the Sun manages it is covered in How the Sun Works: Nuclear Fusion Powered Star.

At those temperatures, matter exists as plasma, an electrically charged gas of stripped nuclei and free electrons. No wall can touch it without being destroyed, which makes confinement the central engineering problem. A deeper treatment of the physics is in What Is Plasma? The Fourth State of Matter Explained.

The fuel supply has an asymmetry that matters later. Seawater holds about 33 grams of deuterium per cubic meter, so deuterium is abundant. Tritium is not. It "occurs only in trace quantities in nature," and ITER puts the global inventory at around 20 kilograms (ITER, fusion fuels). The plan is to breed tritium inside the reactor: neutrons escaping the plasma strike lithium in the blanket wall and produce it.

ITER estimates that a 1,000-megawatt fusion plant would need about 250 kilograms of fuel per year, split equally between deuterium and tritium. A coal plant of the same size burns roughly 2.7 million tonnes of coal per year. At any moment, less than one gram of fusion fuel sits in the reactor vessel.

Fusion differs from fission in kind, not just in degree. Fission splits heavy nuclei and can sustain a chain reaction, as explained in Nuclear Fission Explained: How Nuclear Power Works. A fusion plasma has no chain reaction to run away. If confinement fails, the plasma cools and the reaction stops. For the broader safety picture of nuclear power, see Is Nuclear Power Safe? The Real Risks and Benefits.

Three kinds of "net energy" that headlines blur

Most confusion in fusion news comes from the word "gain." There are at least three different things a company or laboratory can mean, and they are separated by orders of magnitude.

Kind of gain	Question it answers	Status as of October 2026
Scientific (target) gain	Does the fusion energy out exceed the energy delivered to the fuel?	Achieved at NIF; record target gain of 4.13 on April 7, 2025
Engineering (wall-plug) gain	Does the whole facility put out more electricity than it draws?	No facility has published such a result
Commercial gain	Can the electricity be sold at a competitive price?	Not demonstrated

The distinction is easy to see with a campfire analogy. Scientific gain asks whether the logs gave off more heat than the match you struck. Engineering gain asks whether the fire gave off more heat than you spent driving to the forest, cutting the wood, hauling it home, and running the fan. Commercial gain asks whether anyone would pay for the warmth. A fire can pass the first test and fail the other two by a wide margin.

Magnetic confinement: tokamaks, stellarators, and the magnet shift

The FIA's 2026 survey found that 48 percent of the companies it covers pursue magnetic confinement, 21 percent inertial confinement, and 14 percent a hybrid called magneto-inertial fusion (World Nuclear News on the FIA 2026 report). Magnetic confinement is the older idea: build a cage of magnetic field that holds the plasma away from the walls, the way a magnetic bottle might hold a superheated gas.

A tokamak is a donut-shaped chamber in which external magnets and an electric current driven through the plasma together form the cage. It is the most-studied design in the world and the one ITER uses. ITER's seven members are China, Europe, India, Japan, Korea, Russia, and the United States.

ITER's delay is the backdrop for private fusion

ITER's 2024 rebaselining moved the schedule substantially. Operation with a deuterium-only plasma is now planned for 2034, compared with a 2016 plan that had first plasma in 2025. Full plasma current follows in 2036, and deuterium-tritium operation, the phase that matters for energy, moved from 2035 to 2039. The delay adds about €5 billion. ITER's director-general Pietro Barabaschi cited the pandemic, quality problems, and over-optimistic planning for a first-of-its-kind project, and the cracks found in cooling pipes for the thermal shields and a brief regulatory halt in early 2022 were part of the story (Physics World, July 2024).

ITER's toroidal-field magnets are made of niobium-tin and niobium-titanium and generate a 12-tesla field. Each of the 19 coils is 17 meters tall and weighs about 360 tonnes. Those are impressive numbers, and they are also the limit of what low-temperature superconductors can do. For the physics of why superconductors matter at all, see What Is Superconductivity?, and for where the phenomenon was first seen, Onnes and the Discovery of Superconductivity (1911).

How stronger magnets shrink a reactor

The startup case rests on one technical shift: high-temperature superconducting (HTS) magnet tape, which can carry current at stronger fields than ITER's materials. On September 5, 2021, a large HTS electromagnet built by Commonwealth Fusion Systems and MIT was ramped up to a field of 20 tesla for the first time, compared with ITER's 12 tesla.

Why does that matter so much? The pressure a magnetic field can hold in check rises steeply with field strength, so even a modest increase in field lets engineers build a much smaller machine for the same plasma. MIT's announcement put a number on it: the HTS magnet design equals the performance of an apparatus "40 times larger in volume" built with conventional low-temperature superconducting magnets (MIT News, September 8, 2021). A smaller machine is cheaper to build and faster to iterate on, which is the heart of the startup pitch.

CFS is building SPARC in Devens, Massachusetts, as a demonstration machine. TechCrunch reported on July 30, 2026, that the company "now expects" SPARC to achieve scientific breakeven in 2027. The follow-on plant, ARC, is planned for Chesterfield County, Virginia, where CFS says it is "on track to put power on the grid in the early 2030s." Google has committed to buy 200 megawatts from ARC, half its output according to TechCrunch, and Eni has agreed to buy more than $1 billion of its electricity (TechCrunch, July 30, 2026; CFS press release).

The SPARC claim to watch is the breakeven year. It is a company expectation, and it is scientific breakeven (plasma output versus heating input), not electricity delivered to a grid.

Stellarators and the long-pulse records

A stellarator also uses magnetic confinement, but it twists its coils into a complicated three-dimensional shape so the magnets alone provide stability. It needs no large current through the plasma, which avoids a class of sudden instabilities, but the coils are hard to design and build. Computer optimization has made the design more practical, which is why startups are betting on it.

Type One Energy is the clearest U.S. example. On August 31, 2026, the Tennessee Department of Environment and Conservation issued the company a byproduct material license, described as the first in Tennessee to operate a commercial fusion machine. It lets Type One break ground on the first phase of Project Infinity at the Tennessee Valley Authority's Bull Run Energy Complex near Oak Ridge. The plan is a stellarator prototype first, then Infinity Two, a 400-megawatt pilot plant the company expects to start up by 2034 (American Nuclear Society, September 2026).

Public laboratories keep setting records for how long a plasma can last. On February 12, 2025, France's WEST tokamak sustained a hydrogen plasma for 1,337 seconds, about 22 minutes, with 2.6 gigajoules of heating energy injected and extracted. The CEA researchers noted that one of seven klystrons powering the heating antenna stopped around 1,040 seconds and the plasma moved into a new regime, and that a deuterium plasma the next day lasted 921 seconds (CEA-IRFM). Duration records matter because a power plant must run for months, not seconds, though they are set with lower heating power and at lower temperatures than a plant would need.

Inertial confinement: what NIF proved and what it did not

The National Ignition Facility (NIF) at Lawrence Livermore National Laboratory uses up to 192 laser beams to compress a tiny fuel capsule so fast that fusion ignites before the capsule can fly apart. This is inertial confinement: the fuel's own inertia holds it in place for the billionth of a second that matters. The laser physics behind it is in How Lasers Work: Coherent Light Explained.

On December 5, 2022, NIF delivered 2.05 megajoules of laser energy to the target and produced 3.15 megajoules of fusion energy. Since then the numbers have climbed. On April 7, 2025, a shot yielded a record 8.6 megajoules (plus or minus 0.45) from 2.08 megajoules of laser energy, a target gain of 4.13. On June 20, 2026, NIF achieved ignition for the 11th time, yielding 7.9 megajoules at a gain of about 3.8 (LLNL).

The wall-plug number is the part that rarely makes headlines. LLNL director Kim Budil said the NIF laser needed to draw "a whopping 300 MJ from the wall" to produce the 2.05 megajoule pulse in 2022. Budil also noted that the laser fires once a day at a single target, that making target capsules takes about two weeks but design-to-delivery takes about seven months, and that a power plant would need millions of identical capsules daily. She said a plant would likely take "a few decades" of research (Optica, December 2022). NIF was built for stockpile stewardship and science, not as a power prototype.

Startups are working on those gaps from different directions. Inertia Enterprises was among the largest raises of the past year, at $450 million, according to the FIA. Pacific Fusion uses rapid, high-energy electrical pulses rather than lasers or magnets to compress its fuel, and its first system, in New Mexico, is designed to reach net facility gain (Pacific Fusion).

Helion's different bet

Helion Energy uses a pulsed field-reversed configuration. Two plasma rings are fired at each other, merge, and are compressed by magnetic fields. When fusion occurs, the hot plasma expands and pushes back against the magnetic field, and that change in magnetic flux induces current that can be captured directly as electricity. The design aims to skip the steam turbine that every tokamak and laser plant still needs.

On February 13, 2026, Helion announced that its seventh-generation Polaris prototype had become the first privately developed machine to demonstrate measurable deuterium-tritium fusion and to reach 150 million degrees Celsius, up from 100 million degrees with the earlier Trenta prototype. Helion calls 100 million degrees the industry threshold for a commercially relevant machine, and its release quotes outside experts who reviewed the data (Helion Energy). The results are company-reported, and the release presents expert comments as endorsements, not as an independent verification process.

The commercial promise is Orion, Helion's first commercial machine, in Malaga, Washington, where Helion began building in July 2025. It is meant to supply fusion electricity to the grid for Microsoft. Helion's results so far come from prototypes, and Helion says it will keep raising Polaris's temperatures to test deuterium-helium-3 fusion, the fuel it plans to use commercially.

That points to the reading rule for this whole sector. "Begin operations" and "deliver power to a customer at full output" are different milestones, so read the exact wording of any target.

The private fusion landscape at a glance

Company	Approach	Verified milestone	Stated commercial target
Commonwealth Fusion Systems	HTS-magnet tokamak (SPARC, then ARC)	$4 billion raised as of July 2026	Scientific breakeven in 2027; ARC power on the grid in the early 2030s
Helion Energy	Pulsed field-reversed configuration with direct electricity recovery	Company-reported D-T fusion and 150 million °C on Polaris, February 2026	Orion plant for Microsoft under construction in Malaga, Washington; no delivered power yet
Type One Energy	Stellarator	Tennessee fusion license issued August 31, 2026	Infinity Two 400 MW pilot plant by 2034
Pacific Fusion	Pulser-driven inertial fusion	First system under way in New Mexico	Net facility gain first; no plant date given
Inertia Enterprises	Inertial fusion	$450 million raised in the year to July 2026 (FIA)	No date given here

Every company in this table is pre-revenue. Targets are company statements, not independently verified outcomes. Sources: FIA via World Nuclear News, TechCrunch, Helion, American Nuclear Society, Pacific Fusion, and company announcements cited above.

Why capital is moving faster than the physics

The FIA's 2026 report counts $4.48 billion raised in the year to July 2026 across 56 companies, the highest annual total in the six years it has reported. The largest raises were Commonwealth Fusion Systems at $863 million, Proxima Fusion at $518 million, Helion at $465 million, and Inertia Enterprises at $450 million. Five companies have a power purchase or offtake agreement or similar commitment, including Microsoft with Helion and Google with Commonwealth Fusion Systems, and six have siting agreements. Seventy-one percent of surveyed companies expect a plant delivering power to the grid "by the 2030s" (World Nuclear News).

Several developments plausibly explain the shift, though no single study ranks them. HTS magnets changed the size calculation for tokamaks. NIF's repeated ignition shots changed the question for laser approaches from "is it possible?" to "can it be repeated and made efficient?" Power purchase agreements gave investors a counterparty. And the buyers are the same companies that need new electricity: data centers are a major driver of new demand, which How AI Data Centers Use Electricity explains.

Money, though, does not repeal physics. The FIA's own respondents identify funding as the key short-term challenge, and neutron-resilient materials, power efficiency, and tritium self-sufficiency as the longer-term ones. Machine learning is also entering the field as a control tool for plasmas, as covered in Machine Learning and Nuclear Fusion: Teaching AI to Tame the Sun.

What still has to be solved before fusion reaches the grid

Four hurdles get less attention than ignition headlines.

• Tritium self-sufficiency. With only about 20 kilograms of tritium in the world, a power plant has to breed its own. The neutrons escaping the plasma react with lithium in the blanket to produce tritium, and ITER plans to test the concept with test blanket modules. No plant-scale breeding blanket has been demonstrated.
• Materials. The wall closest to the plasma takes a continuous bombardment of high-energy neutrons. ITER's rebaseline switches its first wall from beryllium to tungsten, which is closer to what a demonstration power plant would use.
• Power conversion. Heat from the reaction must become electricity efficiently, and the plant must run its magnets, lasers, cooling, and fuel systems on a fraction of its output. This is the engineering-gain problem from the table above.
• Repetition rate and cost. NIF fires once a day. A laser plant would need to fire many times per second, with millions of capsules a day, as Budil described.

Fusion would also need to fit into a grid that already has other tools. Storage, for one, is a different answer to the same question of reliable clean power, discussed in Grid-Scale Battery Storage: The Missing Piece of the Energy Transition.

Where the confusion usually comes from

Treating ignition as a power plant. NIF's gain is measured against the laser light that struck the target. The facility drew far more from the grid. Both statements are true, and the ignition result is still a landmark.

Treating "fusion startup" as one technology. Tokamaks, stellarators, lasers, pulsed-magnetic systems, and hybrids share a goal and little else. A result from one design says little about another. For a wider look at the science, see Nuclear Fusion in 2026: Is It Finally Working? and Nuclear Fusion Energy Explained: The Power of the Stars.

Believing dates. Fusion has a long record of schedule slips. ITER's first operation slipped from 2025 to 2034 and its deuterium-tritium phase from 2035 to 2039. Treat any year as a target and ask what the milestone actually measures.

Assuming money equals progress. Funding signals confidence, not results. The $14.24 billion raised over six years is a large number, but the FIA's own respondents still name funding as the main near-term challenge.

The takeaway

Private fusion is moving faster than public fusion for identifiable reasons: stronger magnets, a laser facility that has repeatedly reached ignition, and buyers willing to sign contracts for power that does not exist yet. The verified milestones are real, including NIF's record gain of 4.13, WEST's 22-minute plasma, ITER's completed magnets, and the first state fusion license for a commercial machine. The unproven ones are the ones that sell electricity: whole-plant net energy, tritium breeding, and materials that survive years of neutron exposure. Until a plant shows those on a grid, read every fusion date as a hypothesis with a budget attached.

Frequently Asked Questions

Q: What is nuclear fusion and how is it different from fission? A: Fusion merges light nuclei, usually hydrogen isotopes, into a heavier nucleus and releases energy. Fission splits heavy nuclei such as uranium. A fusion plasma has no chain reaction, so if confinement fails it cools and stops, while fission plants need active control to prevent a runaway reaction.

Q: Has any fusion reactor achieved net energy gain? A: Scientific gain, yes: NIF delivered 2.05 megajoules of laser energy and produced 3.15 megajoules on December 5, 2022, and reached a record target gain of 4.13 on April 7, 2025. Engineering gain, where the whole facility makes more electricity than it uses, has not been published by any facility. NIF's laser drew about 300 megajoules from the wall for the 2022 shot.

Q: Why are private fusion companies raising billions if fusion power does not exist yet? A: The FIA counted $4.48 billion raised in the 12 months to July 2026 across 56 companies. Investors cite stronger HTS magnets, NIF's ignition results, and power purchase agreements such as Microsoft with Helion and Google with Commonwealth Fusion Systems. The money pays for demonstration machines, not for proven power.

Q: When will fusion power reach the grid? A: Company targets cluster in the early 2030s: Commonwealth Fusion Systems says ARC will put power on the grid in the early 2030s, and 71 percent of surveyed companies expect a plant by the 2030s. ITER's deuterium-tritium operation is planned for 2039. These are targets, not outcomes.

Q: What is the difference between a tokamak and a stellarator? A: A tokamak confines plasma with external magnets plus a current driven through the plasma itself, which is simpler to build but prone to sudden instabilities. A stellarator uses only external, twisted three-dimensional magnets, which avoids that current but demands very precise coil manufacturing.

Q: Why is tritium such a problem for fusion? A: Tritium is rare and radioactive, with a global inventory of roughly 20 kilograms according to ITER. Power plants have to breed it from lithium inside the reactor using the fusion neutrons, and no plant-scale breeding blanket has been demonstrated yet.

Sources

• Fusion Industry Association, The Global Fusion Industry in 2026, as reported by World Nuclear News.
• Commonwealth Fusion Systems, press release, July 30, 2026, and TechCrunch, July 30, 2026.
• Lawrence Livermore National Laboratory, How ICF works and NIF records, and Optica, "Ignition accomplished: NIF looks to the future," December 2022.
• Physics World, ITER rebaseline, July 2024; ITER, What is fusion and fusion fuels.
• Helion Energy, February 13, 2026 announcement.
• American Nuclear Society, Type One Energy license, September 2026; CEA-IRFM, WEST record, February 2025.
• U.S. Department of Energy, DOE Explains: Nuclear Fusion Reactions; MIT News, HTS magnet test, September 2021.

---

Read next:

• How GPUs work: why AI needs them
• Why semiconductors are so hard to make
• How AI models are trained: the full process