Skip to content
FusionBenchmark

Fusion approaches compared

Every family of fusion machine, how each holds its fuel, the best result this dataset holds for it, its main unsolved problem and the organisations pursuing it.

Guide · Updated

A fusion machine has one job: keep a fuel hot and dense for long enough that it releases more energy than was spent heating it. The how fusion works guide explains those three conditions. This guide goes through the ways people try to meet them.

Each entry gives how the approach works, the best result this dataset holds for it (with device and date), the problem that remains open, and the organisations in the dataset that pursue it. Where the dataset holds no measured result for an approach, the entry says so. That absence is information: it means no figure has yet reached the standard the benchmark applies.

Company claims in this guide are attributed to the company. None of them is a measurement unless it says so.

Magnetic confinement

A plasma is made of charged particles, and charged particles spiral along magnetic field lines rather than crossing them. Magnetic confinement uses that to hold a thin plasma, about 10²⁰ particles per cubic metre, away from the walls for seconds or longer. See magnetic confinement.

Tokamak

How it works. A ring-shaped (toroidal) chamber with strong magnetic coils around it, plus a large electric current driven through the plasma itself. The current’s own field twists the field lines so particles stay confined. See tokamak.

Best results held. The highest magnetic-confinement triple product in the dataset: 7.9 × 10²⁰ keV·s·m⁻³ on TFTR, shot 83546, 17 February 1995. The largest fusion energy from one pulse: 69.26 MJ on the Joint European Torus, pulse 104522, 3 October 2023. The highest plasma gain: 0.63 on JET, shot 42976, 31 October 1997. The longest high-confinement plasma: 1,066 seconds on EAST, 20 January 2025. The longest plasma in the dataset: 1,337 seconds on WEST in France, 12 February 2025, a hydrogen plasma whose confinement mode the record does not state.

Open problem. The plasma current can collapse suddenly (a disruption) and damage the machine, and a tokamak cannot drive that current steadily without extra systems. No tokamak has yet reached a plasma gain above 1. ITER is designed for 10, and Commonwealth Fusion Systems’ SPARC for more than 2. Both are design targets.

Who pursues it. Commonwealth Fusion Systems, ITER Organization, General Atomics, Energy Singularity, UKAEA, Korea Institute of Fusion Energy and ASIPP, among others.

Spherical tokamak

How it works. A tokamak squashed into a shape like a cored apple. The tighter shape holds plasma pressure with less magnetic field. See spherical tokamak.

Best result held. An ion temperature of 9.6 keV (about 111 million kelvin) on Tokamak Energy’s ST40, shot 10009, in 2022. ST40’s triple product in the same compilation is 6.3 × 10¹⁸, well below the large conventional tokamaks.

Open problem. There is very little room in the centre of the machine for the central magnet and for shielding against neutrons.

Who pursues it. Tokamak Energy, STEP, UKAEA, ENN and Startorus Fusion.

Stellarator

How it works. Twisted, three-dimensional external coils produce all of the confining field, so no current has to flow in the plasma. That allows steady operation and removes the cause of tokamak disruptions. See stellarator.

Best results held. A triple product of 6 × 10²⁶ K·s·m⁻³ on Wendelstein 7-X in its 2017 campaign, which the Max Planck Institute for Plasma Physics described as a stellarator record. Converted at 11.6 million kelvin per keV, that is about 5.2 × 10¹⁹ keV·s·m⁻³. Wendelstein 7-X also ran for 480 seconds on 15 February 2023.

Open problem. The coils are hard to design and build to the precision required, and the best stellarator triple product in the dataset is about one fifteenth of the best tokamak figure (0.052 against 0.79, in units of 10²¹ keV·s·m⁻³).

Who pursues it. Proxima Fusion, Type One Energy, Thea Energy, Renaissance Fusion, Gauss Fusion, Helical Fusion, Stellarex Energy, the Max Planck Institute for Plasma Physics and Japan’s National Institute for Fusion Science.

Magnetic mirror

How it works. A straight tube of plasma, with stronger magnetic field at each end that reflects particles back towards the middle. See magnetic mirror.

Best result held. None. The dataset lists mirror devices such as GDT at the Budker Institute but holds no performance measurement for any of them.

Open problem. Particles leak out of the ends.

Who pursues it. Realta Fusion, Novatron Fusion Group, Terra Fusion and LINEA Innovations, which is also filed under field-reversed configuration.

Field-reversed configuration

How it works. A ring of plasma that carries its own current and generates its own confining field, with no coil through the middle. See field-reversed configuration.

Best result held. None measured to the benchmark’s standard. TAE Technologies said in July 2022 that its Norman device held stable plasma above 75 million degrees Celsius (about 6.5 keV). Helion said in February 2026 that its Polaris machine reached 150 million degrees Celsius (13 keV) and produced measurable D-T fusion. Both are company statements.

Open problem. Keeping the plasma ring stable for long enough.

Who pursues it. TAE Technologies, Helion Energy, HHMAX-Energy and LINEA Innovations. Helion’s machines also pulse: they compress the plasma repeatedly with magnetic fields and aim to recover electricity directly from its expansion, filed here as pulsed magnetic compression.

Levitated dipole

How it works. A superconducting ring floats inside the chamber, held up by magnetic force, and the plasma forms around it, much as planets hold plasma in their magnetic fields. See levitated dipole.

Best result held. None. The dataset lists the RT-1 device at the University of Tokyo but holds no performance measurement.

Open problem. Keeping a superconducting magnet cold and powered while it floats inside a hot plasma chamber, and showing that confinement scales to fusion conditions.

Who pursues it. OpenStar Technologies, which says it reached first plasma within two years of starting.

Reversed-field pinch

How it works. A ring-shaped plasma in which the magnetic field reverses direction near the edge, allowing confinement with a much weaker external field than a tokamak. See reversed-field pinch.

Best result held. None. The dataset lists MST, RFX, EXTRAP T2R, KTX and RELAX, with no performance measurement.

Open problem. Confinement quality well below that of tokamaks.

Who pursues it. No company in the dataset. The work is at universities and national laboratories.

Inertial confinement

Inertial confinement crushes a small fuel pellet so fast that it burns before it can fly apart. The fuel is held by nothing but its own inertia, for nanoseconds. See inertial confinement.

Laser, indirect drive

How it works. Lasers heat the inside of a small gold or uranium cylinder (a hohlraum), which gives off X-rays that implode the fuel capsule inside. See laser ICF, indirect drive.

Best results held. The first target gain above 1: 1.5 on the National Ignition Facility, shot N221204, 5 December 2022. A later record of 4.13 on shot N250406, 7 April 2025 (8.6 MJ of fusion energy from 2.08 MJ of laser light).

Open problem. NIF was built to study single shots. A power plant would need to repeat them many times a second, with a far more efficient laser and targets that cost very little.

Who pursues it. Lawrence Livermore National Laboratory, Inertia and Longview Fusion Energy Systems.

Laser, direct drive

How it works. Laser beams strike the fuel capsule directly, with no hohlraum in between. See laser ICF, direct drive.

Best result held. A target gain of 0.03 on OMEGA at the University of Rochester, shot 103952, 14 April 2022.

Open problem. The beams must strike the capsule almost perfectly evenly, or the implosion goes lopsided.

Who pursues it. Xcimer Energy, Focused Energy, Marvel Fusion, Blue Laser Fusion, EX-Fusion, HB11 Energy and Innoven Energy.

Heavy-ion beams

How it works. Particle accelerators, rather than lasers, deliver the compression energy as beams of heavy ions. Accelerators are efficient and can fire repeatedly. See heavy-ion ICF.

Best result held. None. The dataset holds no device or organisation pursuing it.

Open problem. No heavy-ion driver has been built at the scale needed to compress a fuel capsule to fusion conditions.

Projectile-driven

How it works. A projectile fired at very high speed strikes a target, and the shock compresses the fuel. See projectile-driven.

Best result held. None. The dataset lists First Light Fusion’s facility with no performance measurement.

Open problem. Reaching the compression that lasers achieve, with a projectile.

Who pursues it. First Light Fusion, which in March 2025 dropped its plan to build its own fusion power plant. It now offers its target and amplifier technology to other fusion companies and to other industries.

Magneto-inertial

Magneto-inertial designs compress a plasma that already carries a magnetic field. The field slows heat loss during the compression, so the plasma needs less density than inertial confinement and less time than magnetic confinement. See magneto-inertial.

Z-pinch

How it works. A large current flows straight along a column of plasma, and the current’s own magnetic field squeezes the column inward. See Z-pinch.

Best result held. A peak electron temperature of 1 to 3 keV (about 11 to 37 million degrees Celsius) on Zap Energy’s FuZE, measured in 2022 and held as claimed. The figure is Zap’s summary of a 2024 paper in Physical Review Letters; it is an electron temperature, not an ion temperature, and the pinch lasts microseconds. The dataset also lists Zap’s newer FuZE-Q, which Zap said produced its first plasmas in June 2022, and Sandia’s Z Machine, with no performance measurement for either. Zap said in 2022 that its modelling puts the Q = 1 equivalent point at about 650 kA of current. That is a model prediction.

Open problem. The column is prone to kink and break up. Zap uses a sheared flow of plasma along the column to hold it steady.

Who pursues it. Zap Energy, Sandia National Laboratories, LPP Fusion and the University of Washington.

Magnetised target

How it works. A magnetised plasma is compressed by a collapsing metal or liquid-metal liner, or by pistons. See magnetised target.

Best result held. None in the dataset.

Open problem. Compressing evenly and fast enough, and repeating it.

Who pursues it. General Fusion and NearStar Fusion. Pacific Fusion, Fuse Energy and Helicity Space are filed under the broader magneto-inertial term.

Electrostatic and beam approaches

How it works. Electric fields, not magnetic ones, accelerate and hold the ions (inertial electrostatic confinement). A related family fires a beam of ions into a solid target. See inertial electrostatic confinement and beam-target.

Best result held. None in the dataset.

Open problem. No electrostatic device has come close to net energy: the electrodes and scattered ions carry energy away faster than fusion adds it. Beam-target designs cannot reach it by construction. Both are proven and useful as neutron sources, which is what several of these companies sell.

Who pursues it. Avalanche Energy, Horne Technologies and Energy Matter Conversion Corporation (electrostatic); SHINE Technologies and Astral Systems (beam-target).

Muon-catalysed fusion

How it works. A muon is a heavier relative of the electron. If one replaces the electron in a molecule of deuterium and tritium, it pulls the two nuclei so close that they fuse at ordinary temperatures. The muon is then usually freed to catalyse another fusion.

Best result. The dataset holds no measurement. Experiments at Los Alamos averaged about 150 D-T fusions per muon. At 17.6 MeV per fusion, that is about 2.6 GeV of fusion energy per muon, against a current estimated cost of about 6 GeV of electricity to make each muon.

Open problem. Muons live for only 2.2 microseconds, and a muon sometimes sticks to the helium nucleus a fusion produces, ending its run. Making muons more cheaply and raising the number of fusions each one catalyses are both needed.

Who pursues it. Acceleron Fusion. See muon-catalysed fusion.

Sources

Related pages

More in Fusion

  • How fusion works What nuclear fusion is, which fuels it can use, the temperature, density and confinement it needs, and why no machine has yet turned it into electricity.
  • Fusion fuels, tritium and neutrons Deuterium-tritium, deuterium-deuterium, deuterium-helium-3 and proton-boron-11 compared: what each reaction releases, where the fuel comes from, and why aneutronic is a relative term.
  • Fusion milestones: what has happened and what is promised The milestones that mark progress towards fusion power, which have been reached and when, and what fusion organisations say they will do next, in their own words.
  • Commercial timeline of fusion power Six fusion power plants are on record, all still planned. The only dated electricity purchase is Helion's May 2023 deal to supply Microsoft from 2028.

All guides · Fusion