Industries / Frontier science / Fusion energy
Magnetic confinement
This is the work of holding fuel gas, heated far beyond the temperature of the sun's core, inside magnetic fields so that it never touches the walls, for long enough that its atoms fuse.
7
Milestones achieved, from 1951 to 2024.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1951
The stellarator proposed
Lyman Spitzer, an astronomy professor at Princeton University, proposed holding hot gas in a figure-eight-shaped tube with a magnetic field made by coils outside it, and called the design a stellarator. After a review by scientists across the US, the Atomic Energy Commission funded it under the code name Project Matterhorn, the start of Princeton's fusion program; Spitzer had been prompted by what Princeton's lab calls "highly exaggerated" fusion claims from Argentina.
Credit
Lyman Spitzer, Princeton University
1958
ZETA's claim withdrawn
On 25 January 1958 the director of Britain's Harwell laboratory told 400 reporters that ZETA, a ring-shaped "pinch" machine that squeezes hot gas with the magnetic field of a current driven through the gas itself, had heated hydrogen to five million degrees and that its neutron counts showed fusion; under questioning he said he was 90 percent sure. A few months later it was shown that the neutrons came from hot gas escaping the magnetic field and striking the metal casing.
LessonCounting neutrons does not prove that a gas is hot enough to fuse; the temperature has to be measured as well.
Credit
Harwell laboratory, UK
1969
Tokamak temperature checked
A team from Britain's Culham Laboratory, working with a scientist from the Kurchatov Institute in Moscow, measured the electron temperature inside the Soviet T-3 tokamak by Thomson scattering, which reads temperature from laser light bounced off electrons. A tokamak is a doughnut-shaped vessel in which outside coils and a current driven through the gas together hold the gas in place. They found 100 eV to 1 keV, roughly 1 to 12 million degrees Celsius, in line with other methods; JET and ITER below are both tokamaks.
Credit
N. J. Peacock, D. C. Robinson, M. J. Forrest and P. D. Wilcock (Culham Laboratory) and V. V. Sannikov (Kurchatov Institute)
1997
JET's deuterium-tritium records
The Joint European Torus (JET), a tokamak near Oxford run at the UK Atomic Energy Authority's site, produced 16.1 MW of fusion power from a mix of deuterium and tritium, the two heavy forms of hydrogen that serve as fuel, as reported in a peer-reviewed paper in 1999. It released 21.7 MJ of fusion energy that year, 59 MJ in one pulse in 2021, at an average fusion gain (Q, fusion power divided by heating power) of 0.33 over its best five seconds, and a final record of 69 MJ over six seconds from 0.21 milligrams of fuel in its last deuterium-tritium campaign in 2023.
Credit
the JET team and the EUROfusion consortium, with the UK Atomic Energy Authority
2021
A 20-tesla magnet
On 5 September 2021 MIT's Plasma Science and Fusion Center and its spinout Commonwealth Fusion Systems (CFS) brought a large magnet wound from high-temperature superconductor, a material that carries current without loss when cooled, to 20 tesla, a record for a large magnet of its kind. Stronger fields let a tokamak be smaller for the same performance. The team later pushed the magnet to its breaking point to study how it fails, and published the results in six peer-reviewed papers in IEEE Transactions on Applied Superconductivity in March 2024.
Credit
MIT Plasma Science and Fusion Center and Commonwealth Fusion Systems
2022
AI shapes the plasma
Researchers at DeepMind and the Swiss Plasma Center used reinforcement learning, a method in which software learns a task by trial and error, to build a single controller that commands all the control coils of the TCV tokamak. On the real machine it held a range of plasma shapes, including two separate plasmas at once, and cut the design work needed for each new shape, as reported in Nature.
Credit
Jonas Degrave, Federico Felici, Jonas Buchli and colleagues, DeepMind and the Swiss Plasma Center
2024
ITER's schedule reset
ITER, the tokamak that China, the European Union, India, Japan, South Korea, Russia and the US are building in southern France, gave up the first plasma in 2025 promised in its 2016 plan. Its 2024 baseline starts research operation in 2034 and deuterium-tritium operation in 2039, four years later than the old plan for that step, and Physics World reported a cost rise of about €5 billion.
LessonIn ITER's own account, first-of-a-kind parts needed repairs and the planning of some manufacturing and assembly "proved to be too optimistic".
Credit
ITER Organization
Watch for
SPARC aims for net gain
CFS says it is assembling SPARC, a compact tokamak built around its high-temperature superconducting magnets in Devens, Massachusetts, expects to turn it on in 2027, and will then reach Q>1: more fusion power out of the plasma than heating power put in.
Achieved whenSPARC produces more fusion power than the heating power it absorbs, and the measurement is published in a peer-reviewed journal or checked by a team outside CFS.
Claimed by
Commonwealth Fusion Systems, November 2026
Polaris burns deuterium-tritium
Helion, a company in Everett, Washington, says its Polaris prototype became the first privately developed fusion machine to show measurable fusion of deuterium and tritium, and reached plasma temperatures of 150 million degrees Celsius.
Achieved whenthe fusion yield and temperature are published in a peer-reviewed paper or measured by an outside team.
Claimed by
Helion, February 2026
A 43-second stellarator record
On 22 May 2025, the last day of its campaign, the Wendelstein 7-X stellarator in Greifswald, Germany, held a peak value of the triple product (density times temperature times how long the plasma keeps its heat, the standard yardstick of fusion conditions) for 43 seconds. Its operator says this beats tokamak records for long plasma durations, while the tokamaks JT-60U and JET keep the highest values over a few seconds.
Achieved whenthe result appears in a peer-reviewed paper; a Crossref search in November 2026 found none.
Claimed by
Max Planck Institute for Plasma Physics (IPP), June 2025
Open
01
SPARC forecast, filed in advance
Before SPARC first runs in 2027, gather every published prediction of its performance (fusion gain Q and fusion power, scenario by scenario) into one public table. Add an independent prediction made with an open-source code from CFS's published machine parameters, and archive the whole forecast with a timestamp. After SPARC's first campaign, score each prediction against the measured results.
The briefCredit · unclaimed02
Press claims, then papers
A public ledger of fusion performance claims made by press release since 2015, by companies and public labs: temperature, triple product, pulse length, gain or fusion energy. For each, record whether a peer-reviewed paper reporting the same number followed, where, how many months later, and whether anyone outside the organization measured it.
The briefCredit · unclaimed
Open 01 · the brief
SPARC forecast, filed in advance
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Before SPARC first runs in 2027, gather every published prediction of its performance (fusion gain Q and fusion power, scenario by scenario) into one public table. Add an independent prediction made with an open-source code from CFS's published machine parameters, and archive the whole forecast with a timestamp. After SPARC's first campaign, score each prediction against the measured results.
Why it is still open
Published SPARC predictions come from MIT and CFS researchers and close collaborators, for example a 2025 study of SPARC's 12.2-tesla reference scenario run in the ASTRA framework with the TGLF transport model. Since October 2025 CFS has used TORAX, Google DeepMind's open-source tokamak transport simulator, under a research partnership to test SPARC operating plans in simulation, but those runs are not public, and TORAX ships example setups for ITER and the UK's STEP design and none for SPARC (checked November 2026). A November 2026 search found no forecast filed in public for later scoring.
Built with what exists
TORAX (open source, written in JAX); the machine parameters in the published SPARC papers; Zenodo or a similar archive that issues a dated DOI; a coding agent such as Claude Code to set up the runs and parameter scans, with every input checked by a person.
First step
Install TORAX, run its ITER example, then change the geometry, magnetic field and heating to the SPARC reference values given in the 2025 study and compare the result with that study's prediction.
Done looks like
A forecast archived before SPARC's first plasma, holding at least 10 published predictions and one independent TORAX prediction for the 12.2-tesla reference scenario, read by one plasma physicist outside CFS; then, after SPARC's first campaign, a published comparison of each prediction with the Q that CFS reports.
Open 02 · the brief
Press claims, then papers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public ledger of fusion performance claims made by press release since 2015, by companies and public labs: temperature, triple product, pulse length, gain or fusion energy. For each, record whether a peer-reviewed paper reporting the same number followed, where, how many months later, and whether anyone outside the organization measured it.
Why it is still open
The best-known compilation, by Samuel Wurzel and Scott Hsu, plots only published experimental results against the Lawson criterion (the conditions fusion fuel must reach to give net energy); its 2025 update and open code do not track announcements that have no paper yet, such as IPP's June 2025 stellarator record above. A November 2026 search found no public ledger of claims.
Built with what exists
The Crossref and arXiv APIs to find matching papers; an AI agent that reads each press release and drafts a row, with a person checking every row against its sources (see the Note "The jobs agents repeat"); a spreadsheet published with dated links.
First step
Enter the five claims on this page (CFS's 2021 magnet, JET's 2023 record, NIF's April 2025 shot, W7-X's May 2025 record, Helion's February 2026 result) and find the paper for each.
Done looks like
At least 60 claims from at least 15 organizations, each linked to its release and to its paper or marked "none found" with the search date; the median and range of months from claim to paper; every organization shown its rows two weeks before publication, with any reply recorded.