BRENNEN

Industries / Frontier science / Fusion energy

Fuel

This is the work of supplying the hydrogen isotopes a fusion machine burns, making new tritium inside the machine from lithium, and moving fuel around a closed loop without losing it.

6

Milestones achieved, from 1934 to 2023.

3

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1934

    Deuterium fused in the lab

    At the Cavendish Laboratory in Cambridge, Mark Oliphant, Paul Harteck and Ernest Rutherford fired deuterium nuclei at compounds rich in deuterium and saw "an enormous emission of fast protons" even at 20,000 volts, with an equal number of short-range particles: deuterium nuclei were fusing. Britannica credits the experiment with discovering tritium, through the reaction D + D → H + T, and the Atomic Heritage Foundation credits Oliphant with co-discovering helium-3 as well.

    nature.combritannica.comahf.nuclearmuseum.org

    Credit

    M. L. Oliphant, P. Harteck and Lord Rutherford

  2. 1989

    Tritium from Canadian reactors

    Heavy water reactors, which use water made with deuterium, are the only commercial source of tritium, which they make as a by-product. Since 1989 Ontario's Darlington station has run a Tritium Removal Facility that takes tritium out of its heavy water, and JET's tritium for its 1997 campaign came from Canada through the Canadian Fusion Fuels Technology Project. UKAEA researchers write that Ontario's output will cover ITER's needs, put unofficially at 12.3 kg over its life, but may leave only 7.4 kg for a demonstration power plant in 2055.

    Lessonfusion's starting fuel comes from fission reactors that make it by accident, and it decays by about 5.5% a year (a half-life of 12.32 years), so supply shrinks as those reactors retire.

    osti.govscientific-publications.ukaea.ukbritannica.com

    Credit

    Ontario Hydro, now Ontario Power Generation

  3. 1997

    Tritium recycled in a loop

    For JET's first full deuterium-tritium campaign (DTE1, 1997) only 20 g of tritium was on site, yet about 100 g was supplied to the machine and its heating beams over some 220 tritium pulses. A separate plant, the Active Gas Handling System, collected the exhaust, separated the hydrogen isotopes and sent the tritium back at least five times. JET's report called it the first true demonstration that tens of grams of tritium could be processed and recycled safely in a large fusion device, with no unplanned releases and no delays caused by the plant.

    scipub.euro-fusion.org

    Credit

    R. Lässer and the JET Active Gas Handling System team

  4. 2020

    Self-sufficiency judged out of reach

    A power plant must breed (make) at least as much tritium as it burns. A review by Abdou, Riva, Ying, Day, Loarte, Baylor and co-authors modeled the whole fuel cycle hour by hour and concluded that current physics and technology would not let a demonstration plant or later power plants make their own tritium, find enough to start up, and handle it safely. To be self-sufficient with reasonable confidence, the product of burn fraction (the share of injected tritium that fuses) and fueling efficiency had to exceed 2%, exhaust had to be processed within 1 to 4 hours, and the plant had to run more than half the time; below 10% of the time it was impossible. At 2% and 4 hours a 3 GW plant would need about 11 kg of tritium to start.

    Lessonhow much of the fuel burns, and how fast the exhaust is cleaned, matter as much as the blanket.

    doi.org

    Credit

    M. Abdou and co-authors

  5. 2023

    Proton-boron fusion measured

    Fusing a proton with boron-11 is an alternative to deuterium-tritium that needs no tritium. In Japan's Large Helical Device, a team from TAE Technologies and the National Institute for Fusion Science injected boron powder into a hot plasma and fired three 2 MW beams of fast hydrogen into it, and a purpose-built detector counted the alpha particles (helium nuclei) the reaction gives off, at a rate that agreed with calculation. The paper calls it the first such measurement in a magnetically confined plasma.

    doi.org

    Credit

    R. Magee, K. Ogawa, T. Tajima and colleagues at TAE Technologies, the National Institute for Fusion Science and the University of California, Irvine

  6. 2023

    Tritium cleaned out in weeks

    JET's last tritium campaigns, with its metal wall, put far more fuel through the machine than DTE1: just over 373 g of tritium was supplied for the second deuterium-tritium experiment (DTE2) and 630 g for a tritium-only campaign, both run between 2020 and 2022. After the final campaign (DTE3) in 2023, the team baked the vessel and ran deuterium plasmas aimed at the walls, and the tritium share of the plasma reached its target of 0.02% after about four weeks.

    scientific-publications.ukaea.ukscientific-publications.ukaea.uk

    Credit

    A. Widdowson, S. Brezinsek and JET contributors, EUROfusion

Watch for

  1. Breeding tritium at test scale

    UKAEA's LIBRTI program, about £220 million from 2023 to 2030, is turning a building at its Culham Campus into a test bed: a 14 MeV neutron source in a shielded blockhouse, where companies will bring engineering-scale breeder blanket prototypes weighing tonnes to prove that they make tritium under neutrons. In July 2026 Commonwealth Fusion Systems became its first international partner and agreed to build the first test articles.

    Achieved whena prototype tested there produces tritium at the rate its model predicts, and the result is published.

    ukaea.organs.org

    Claimed by

    UKAEA and Commonwealth Fusion Systems, July 2026

  2. A tritium-ready blanket loop

    Kyoto Fusioneering said in February 2026 that it had begun building UNITY-2, a facility for testing blanket components with tritium and for extracting tritium from breeder materials, including tests of the inner fuel cycle, with commissioning scheduled for late 2026. Its non-nuclear twin, UNITY-1, already pumps lithium-lead through a 4 T magnetic field at up to 1,000 °C.

    Achieved whenUNITY-2 is commissioned and publishes measured tritium extraction from a flowing breeder, checked by an outside party.

    indico.ukaea.uk

    Claimed by

    Kyoto Fusioneering, February 2026

  3. Ontario tritium for UK fusion

    In September 2025 Ontario Power Generation and UKAEA signed a memorandum of understanding to work toward a secure, reliable supply of tritium from Ontario's reactors for UKAEA's fusion research, which includes its STEP prototype power plant.

    Achieved whena supply contract is signed and tritium is delivered under it.

    interactive.opg.com

    Claimed by

    Ontario Power Generation and UKAEA, September 2025

Open

  1. 01

    Natural lithium, five blankets

    A simulation study, run only on a computer, of how much tritium five blanket designs breed as the share of lithium-6 in their lithium changes. Model the five main designs (FLiBe molten salt, two lithium-lead designs known as WCLL and DCLL, a helium-cooled pebble bed known as HCPB, and a liquid-lithium design) in one shared reactor geometry, and compute each one's tritium breeding ratio (tritium made per tritium burned) from natural lithium up to 90% lithium-6. Report the lowest lithium-6 share at which each design still breeds enough.

    The brief
    Credit · unclaimed
  2. 02

    Check fuel models against JET

    Set up a plant-level tritium fuel-cycle model, of the kind Abdou and co-authors and Meschini and co-authors used, for JET's tritium plant using only published numbers, and test whether it reproduces what JET recorded: about 100 g supplied from a 20 g site stock in DTE1, about 373 g in DTE2 and 630 g in the tritium-only campaign, the tritium held in the walls, and how fast clean-up removed it.

    The brief
    Credit · unclaimed

Open 01 · the brief

Natural lithium, five blankets

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A simulation study, run only on a computer, of how much tritium five blanket designs breed as the share of lithium-6 in their lithium changes. Model the five main designs (FLiBe molten salt, two lithium-lead designs known as WCLL and DCLL, a helium-cooled pebble bed known as HCPB, and a liquid-lithium design) in one shared reactor geometry, and compute each one's tritium breeding ratio (tritium made per tritium burned) from natural lithium up to 90% lithium-6. Report the lowest lithium-6 share at which each design still breeds enough.

Why it is still open

In May 2026 Ward, Pearson, Scott and Lopes Cardozo argued that the 50 to 100 tonnes of enriched lithium a plant must hold drive its capital cost, that today's enrichment methods are too expensive, not scalable and environmentally risky, and that highly enriched lithium-6 is a controlled substance; they call for blankets that use natural lithium and an urgent reassessment. In 2020 Segantin, Testoni and Zucchetti compared liquid breeders, among them FLiBe, lithium-lead and pure lithium, as bulk liquids in one ARC-like geometry with OpenMC and found the best lithium-6 share for each, but left out solid breeders such as the pebble bed and the structures and coolants of full blanket designs. Politecnico di Torino and Eni compared five blanket designs in one open OpenMC workflow in September 2025, but their abstract reports no lithium-6 sweep. A November 2026 search found no same-geometry curve for full designs of all five kinds.

arxiv.orgiris.polito.itconferences.iaea.org

Built with what exists

OpenMC, an open-source Monte Carlo code that tracks neutrons through a model; the IAEA's FENDL nuclear data library; Paramak to build the geometry; ordinary cloud CPUs.

  1. First step

    Install OpenMC with FENDL data, reproduce one published FLiBe or HCPB breeding ratio, then rerun it with natural lithium.

  2. Done looks like

    For each of the five designs, the breeding ratio at six or more lithium-6 levels from natural to 90%, each with a statistical uncertainty below 0.5%; one published breeding ratio, such as Segantin's for FLiBe, reproduced within 3% as a check; the lowest lithium-6 share giving a ratio of 1.15 reported for each design; all inputs public.

Open 02 · the brief

Check fuel models against JET

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

Set up a plant-level tritium fuel-cycle model, of the kind Abdou and co-authors and Meschini and co-authors used, for JET's tritium plant using only published numbers, and test whether it reproduces what JET recorded: about 100 g supplied from a 20 g site stock in DTE1, about 373 g in DTE2 and 630 g in the tritium-only campaign, the tritium held in the walls, and how fast clean-up removed it.

Why it is still open

Open fuel-cycle codes now check each other. Idaho National Laboratory's TMAP8 reproduces the published results of Abdou et al. (2021) and Meschini et al. (2023), and the open TRICYS platform reproduces Abdou's 292-point data set. These are checks of code against code, not against a plant, and a November 2026 search found no plant-level model checked against a recorded tritium campaign.

tmap8.inl.govgithub.com

Built with what exists

TMAP8 or TRICYS, both open source, or a small Python model; the published JET gas-handling and fuel-retention papers; an AI agent to pull every number from those papers into a table with page citations (see the Note "The jobs agents repeat").

  1. First step

    Build the table from the 1998 JET gas-handling overview, UKAEA's DTE2 gas-supply paper and the 2024 overview of fuel retention in JET.

  2. Done looks like

    A public table of at least 40 published JET tritium figures, each with a page citation; a model that reproduces the tritium supplied and the tritium left in the vessel at the end of DTE1 and DTE2 within 20%, with every input cited; a list of where and why it misses.