How fusion works

Fusion is the joining of two light atomic nuclei into a heavier one. The product weighs slightly less than the two nuclei that went in, and the missing mass is released as energy. The Sun runs on it. A fusion power plant would try to do the same thing on Earth, fast enough and efficiently enough that the energy released pays for the energy spent making it happen.

Fission, the process in every operating nuclear power station, works the other way: it splits a heavy nucleus such as uranium-235. The two share the word “nuclear” and very little else. A fission reactor starts itself once enough fuel is gathered in the right shape. A fusion device has to be forced, continuously, to keep going.

The fuels

Every atomic nucleus carries positive charge, so two nuclei repel each other. To fuse, they must approach fast enough to overcome that repulsion. The lightest nuclei, the forms of hydrogen, repel least, which is why almost every fusion fuel is built from them.

• Deuterium–tritium (D-T). Deuterium is hydrogen with one extra neutron; tritium has two. Each D-T reaction releases 17.6 MeV, carried by a helium nucleus (3.5 MeV) and a neutron (14.1 MeV). About 80% of the energy therefore leaves in the neutron. D-T fuses at the lowest temperature of any practical fuel, which is why every near-term attempt at net energy uses it. See deuterium–tritium.
• Deuterium–deuterium (D-D). Deuterium alone. It needs far higher temperatures than D-T, but deuterium is plentiful: ITER gives 33 grams in every cubic metre of seawater. See deuterium–deuterium.
• Deuterium–helium-3 and proton–boron-11. These release most of their energy as charged particles rather than neutrons, and are much harder to ignite. They are covered in Fusion fuels and neutrons.

Tritium barely exists in nature. It is radioactive, with a half-life of 12.32 years, and ITER puts the world’s inventory at around 20 kilograms. A D-T power plant would have to make its own tritium from lithium, using the neutrons its plasma produces. That too is covered in the fuels guide.

The three conditions

At fusion temperatures, electrons are stripped from their nuclei and the fuel becomes a plasma: a hot, electrically charged gas. Getting useful energy out of a plasma needs three things at once.

Temperature. The nuclei must move fast enough to overcome their repulsion. Plasma physicists give temperature in kiloelectronvolts (keV) rather than degrees, because the number then describes the energy of the particles directly. One keV is about 11.6 million kelvin (1,000 eV divided by Boltzmann’s constant, 8.617 × 10⁻⁵ eV per kelvin, gives 1.16 × 10⁷ K). A headline “100 million degrees” is therefore about 8.6 keV. At these magnitudes, kelvin and Celsius differ by only 273 degrees, so the distinction does not matter. See ion temperature.

It matters which particles are hot. Fusion depends on the ions, the nuclei. Electrons in the same plasma can be at a different temperature, hotter or cooler depending on how the plasma is heated: microwave heating warms the electrons first, while neutral beams can leave the ions far hotter. A figure quoted as “the plasma temperature” without saying which is a warning sign.

Density. The more nuclei packed into each cubic metre, the more often they collide. The fusion rate rises with the square of density. Magnetic confinement devices work at about 10²⁰ particles per cubic metre, around 250,000 times thinner than air at sea level (about 2.5 × 10²⁵ molecules per cubic metre). Inertial confinement devices instead crush their fuel to far higher density than any solid, for a few billionths of a second. See plasma density.

Confinement time. A hot plasma loses energy constantly, by radiation and by particles and heat escaping. The energy confinement time, written τ_E, measures how long the plasma holds on to its energy: the energy stored in it divided by the rate at which it is losing energy. It is not the same as how long the machine runs. China’s EAST tokamak held a plasma for 1,066 seconds on 20 January 2025, but the longest energy confinement time this dataset records for EAST is 0.054 seconds. See energy confinement time.

The triple product

The three conditions trade against each other. A denser plasma needs less confinement time; a hotter one can tolerate more loss. The product of all three, density × ion temperature × energy confinement time, is called the triple product, and it is the fairest single measure of how close a plasma is to producing net energy.

For D-T, a plasma heats itself (the state called ignition) once the triple product reaches about 3 × 10²¹ keV·s per cubic metre, at a temperature of about 14 keV. The highest measured value in this dataset for a magnetic-confinement device is 7.9 × 10²⁰, from the TFTR tokamak on 17 February 1995: about a quarter of the ignition threshold. The triple product guide explains the leaderboard in detail.

Two ways to hold a plasma

Nothing solid can touch a plasma at 10 keV without cooling it and being damaged. There are two broad solutions, and several hybrids.

Magnetic confinement uses magnetic fields to hold a low-density plasma away from the walls for seconds or longer. The charged particles spiral along field lines instead of travelling straight to the wall. The tokamak, a ring-shaped device, is the most studied design. See tokamak and magnetic confinement.

Inertial confinement compresses a small pellet of fuel so fast that it burns before it can fly apart. Its own inertia holds it together for nanoseconds. The National Ignition Facility in California does this with lasers. See inertial confinement.

Magneto-inertial designs combine the two: they compress a plasma that already carries a magnetic field. The fusion approaches guide goes through every family and the companies pursuing each.

Why it is hard

The basic physics has been understood for decades. The difficulty is doing everything at once, repeatedly, in a machine that survives.

• The plasma resists being held. Plasmas develop turbulence and instabilities that carry heat to the wall faster than simple theory predicts. Much of the last seventy years has gone into understanding and suppressing them.
• Output has not yet exceeded input in a way that pays for a power plant. On 5 December 2022 the National Ignition Facility released 3.1 MJ of fusion energy from 2.05 MJ of laser light, a target gain of 1.5. The facility stored about 400 MJ of electricity to fire that shot. Counted against that, the gain was about 0.0078 at most. The fusion gain guide explains why both numbers are true.
• The neutrons damage the machine. D-T neutrons carry 14.1 MeV each. They pass through the plasma’s magnetic cage, heat the surrounding structure and, over years, weaken the materials they strike.
• The fuel has to be bred. A D-T plant must make tritium at least as fast as it burns it, which no device has yet done.
• Electricity is the goal, and no fusion device has produced any. Every result so far is heat or particles. Engineering gain, electricity out over electricity in, is zero for every machine ever built.

The benchmark records what each device has actually measured, with the denominator stated for every gain figure, so these gaps can be read directly rather than inferred from announcements.

Sources

• Abu-Shawareb et al., “Achievement of Target Gain Larger than Unity in an Inertial Fusion Experiment”, Physical Review Letters 132, 065102 (2024)
• Lawrence Livermore National Laboratory, NIF power conditioning system
• ITER Organization, Fuelling the fusion reaction
• Chinese Academy of Sciences, EAST sets new record (21 January 2025)
• Wurzel and Hsu, “Continuing progress toward fusion energy breakeven and gain as measured against the Lawson criteria”, arXiv:2505.03834 (2025)
• Wikipedia, Lawson criterion
• Wikipedia, Fusion power (reaction energies)
• Wikipedia, Tritium