On October 21, 2026, at 1:14 PM PT, San Francisco fusion startup Marathon Fusion posted on X: "Last month, we demonstrated high-rate alchemy: turning mercury into real, stable gold, beating CERN's record by a factor of 10." A follow-up post in the same thread added that it was done "using fusion neutrons" (X, Oct 21 2026). By Oct 24 the first post showed about 8.5 million impressions in X's public metrics.
Is it true? Is it new? And should anyone holding gold be nervous? The short answers are "plausibly," "partly," and "no." Here is the long version.
Who is Marathon Fusion?
Kyle Schiller and Adam Rutkowski, a former SpaceX engineer, founded Marathon in 2023. The company builds fuel-cycle systems for future fusion plants. It won an ARPA-E award in 2023, and its backers include Bill Gates' Breakthrough Energy and the mining group Anglo American (Euronews, Oct 23 2026). Marathon is not a power-plant company yet. Its own website lists a 1 kW fusion power target for the "high power density" reactor it is building in San Francisco (marathonfusion.com, accessed Oct 24 2026).
The idea: a gold mine inside a fusion reactor
The concept comes from a July 2025 preprint by Rutkowski, Jake Harter and Jason Parisi, "Scalable Chrysopoeia via (n, 2n) Reactions Driven by Deuterium-Tritium Fusion Neutrons" (arXiv:2507.13461, v1 Jul 17 2025, v3 Aug 3 2025). Chrysopoeia is the old alchemists' word for making gold.
Here is the physics, step by step:
- Fusion makes fast neutrons. When deuterium and tritium fuse, each reaction releases a 14 MeV neutron.
- Mercury-198 absorbs one and spits out two. Every D-T fusion plant needs a "blanket" around the plasma that catches neutrons and breeds new tritium fuel. Many designs include a "neutron multiplier," a material that takes in one neutron and releases two. Marathon proposes using mercury-198 as that multiplier. In this (n, 2n) reaction, Hg-198 loses a neutron and becomes mercury-197.
- Mercury-197 decays into gold. Hg-197 is unstable, with a half-life of about 64.1 hours per the paper. It decays by electron capture, turning one proton into a neutron. That leaves 79 protons, which makes it gold, specifically gold-197, the only stable isotope of gold.
The clever part is that the extra neutrons still help breed tritium. On paper, then, gold becomes a by-product of a blanket the reactor needs anyway, and Marathon's simulations say it doesn't cut electricity output (arXiv:2507.13461).
The numbers Marathon claims
All of these come from Marathon's own paper or website. They are modeled, not measured.
- Yield: "about 2 t/GWth/yr," meaning roughly 2 tonnes of gold per year per gigawatt of thermal fusion power, from neutronics simulations (arXiv abstract). The Oregon Group reports that a modeled 1.5 GWth tokamak made 2,953 kg/yr, which the company markets as about 5 t per gigawatt of electricity (Oregon Group, Oct 22 2026). Watch the units: 2 t per GW thermal and 5 t per GW electric are the same claim expressed two ways.
- Revenue: The paper says gold could "double the revenue" of a fusion plant. Its economic case assumes gold at $100,000/kg and 3 t/yr of gold, worth about $300M a year. With that added, a plant that only breaks even on electricity at $50/MWh gets a net present value "well over three billion dollars" (arXiv:2507.13461, Sec. 5). Marathon's alchemy page frames it as $223M/yr from electricity alone vs. $482M/yr with gold for a gigawatt-scale plant (Deep Dive, Oct 22 2026).
- Feedstock: The tokamak design needs 100–400 tonnes of Hg-198 per GWth, enriched to 90%. The reference case holds 622 tonnes. Only about 0.3–1.2% of that converts to gold each year (arXiv:2507.13461). Hg-198 makes up only about 10% of natural mercury (9.97%, per Oregon Group), so this calls for industrial-scale isotope enrichment that doesn't exist today.
"Real, stable gold," but it comes with a waiting period
The gold-197 itself is stable. The catch is that mercury contains other isotopes too, and some of them turn into radioactive gold. The worst offender is gold-195, made from trace mercury-196, with a half-life of 186.1 days. The paper's own cooldown estimates for the resulting gold are:
- ~6.8 years to qualify as Class A low-level waste, the least hazardous category
- ~13.7 years before it no longer needs radioactive labeling under NRC rules
- ~17.7 years to be "less radioactive than a banana" per troy ounce
(arXiv:2507.13461, Sec. 4.1.2)
Purer feedstock with less mercury-196 shortens the wait. The paper estimates about 1.7 years saved for every tenfold cut in starting gold-195. Marathon CTO Adam Rutkowski told the Financial Times in July 2025 that the gold might need to be stored for 14 to 18 years to be labeled "completely safe" (FT, Tom Wilson, Jul 22 2025, FT English text via FT Chinese; FT.com original is paywalled). For a central-bank vault, that may not matter. For a wedding ring, it does.
What did the September experiment actually show?
This is where the claim gets thin. The public evidence so far is a gamma-ray spectrum posted on X. It shows labeled mercury-197 and gold-197 signatures measured 8–44 hours after irradiation, which fits the proposed reaction (Oregon Group). Marathon has not published the numbers behind its demo. It has not disclosed:
- how much gold it made, by mass
- purity or isotope mix
- the neutron source or how long the irradiation ran
- recovery efficiency or cost per gram
(Deep Dive; Forschung und Wissen, Oct 22 2026)
Euronews put it plainly: the claim "has not been independently verified."
"10× CERN": a low bar
In May 2025, CERN's ALICE experiment reported that near-miss lead–lead collisions at the Large Hadron Collider had produced about 86 billion gold nuclei during Run 2 (2015–2018). That is 29 picograms (2.9 × 10⁻¹¹ g), made at a peak rate of about 89,000 nuclei per second. Those nuclei fragmented "in a tiny fraction of a second" (CERN, May 8 2025).
Marathon hasn't said whether "10×" refers to rate or total, though "high-rate" suggests rate. Either way, ten times a few picograms is still an invisible amount. ALICE deputy spokesperson Kai Schweda said that to make even one gram this way, "the LHC would have to run 30 billion times longer – in other words, for around 100 billion years" (LHC Deutschland interview, 2025). The meaningful difference is that Marathon's gold is a stable isotope that stays put, while CERN's flew into the beam pipe and broke apart.
Alchemy has worked before, at a loss
- 1941, Harvard. R. Sherr, K. T. Bainbridge and H. H. Anderson bombarded mercury with fast neutrons and made gold. All of it was radioactive: Au-198 with a 65-hour half-life, Au-199 at 78 hours, and others (Physical Review 60, 473, Oct 1 1941).
- 1980, Berkeley. Glenn Seaborg's team turned several thousand bismuth atoms into gold using carbon and neon beams, published in 1981 (Aleklett et al., Phys. Rev. C 23, 1044; Live Science). Scientific American reported that Seaborg estimated it "would cost more than one quadrillion dollars per ounce," at a time when gold sold for about $560 an ounce (Scientific American, Jan 31 2014).
- 2025, CERN. ALICE quantified the 29 picograms described above.
Marathon's pitch is that a fusion plant's flood of neutrons finally makes the rate matter.
Hype vs. physics
| The hype | The physics |
|---|---|
| "Startup turns mercury into gold" | The nuclear reaction is real, well understood, and has been demonstrated since 1941. What's new is the proposal to use it at power-plant scale in a fusion blanket. |
| "Lab gold will crash gold prices like lab diamonds" | Mines produced a record 3,672 t in 2025 (initial estimate, subject to revision; WGC Gold Demand Trends FY2025), and about 222,600 t have been mined in history (WGC, end-Q2 2026). At 2 t per GWth, the paper itself says the world would need about 1.65 TW of fusion to match 2024 mine output. |
| "Real, stable gold" | Gold-197 is stable, but the batch carries radioactive gold-195, so it needs roughly 7–18 years of storage. |
| "10× CERN's record" | No mass has been published. Ten times picograms is still picograms. |
| "Fusion plants that print money" | No fusion plant sells electricity today. Helion's Orion plant in Washington aims to supply at least 50 MWe to the grid "from 2028," CFS is planning its ARC plant in Virginia, and ITER isn't scheduled for deuterium-tritium operation until 2039 (World Nuclear Association, updated Oct 23 2026). Marathon's own reactor targets 1 kW. |
| "Gold doubles fusion revenue" | That depends on $100,000/kg gold, hundreds of tonnes of enriched Hg-198 and a working D-T plant. The paper models revenue but not the full production cost. |
Karl Lackner of the Max Planck Institute for Plasma Physics told the society's magazine that using mercury as a multiplier "probably wouldn't be done in the first generation of fusion reactors." He said enriching it to 90% is "a technically demanding process." Even if all the estimated 600,000 tons of mercury in Earth's crust were used, it "would produce only 600 tons of gold per year," so "gold would therefore not become worthless, contrary to some media claims" (Max Planck Research, "Five Questions," issue 2025). Princeton Plasma Physics Lab physicist Ahmed Diallo told the FT in 2025: "On paper it looks great" (FT, Jul 22 2025).
Why it still matters
Fusion's hardest problem may be economics rather than physics. Electricity is a cheap commodity, and Marathon's paper says plainly that a plant selling power at $50/MWh is "almost exactly at breakeven." A high-value by-product that comes from the neutrons the reactor already makes could change how investors model fusion. Gold has a catch, though. Its price is set by the gold market, so the subsidy only works while gold stays expensive.
Marathon's nearer-term bet is more modest: medical isotopes. The company says its experiments have made radioactive copper from zinc (Deep Dive). In a company video, it says its first reactor will be available to partners in 2027 for medical isotope production and materials qualification. That 2027 date is a company claim, and it has not been independently verified.
FAQ
Is this gold real? Gold-197 made this way is chemically and physically identical to mined gold. What hasn't been shown publicly is how much Marathon made.
Is it radioactive? Freshly made batches would contain radioactive gold isotopes, mainly Au-195. The paper estimates 6.8–17.7 years of storage, depending on the safety standard.
Could it crash the gold price? Not on any foreseeable timeline. It would take hundreds of gigawatts to terawatts of fusion capacity that doesn't exist yet.
When could fusion gold be sold? Only after a commercial D-T fusion plant runs with a mercury blanket, plus years of cooldown. No date has been announced.
Is this the first time anyone made gold? No. Physicists did it in 1941, 1980 and at CERN. The new claim is about rate, stability and a path to scale.
Has anyone verified the experiment? Not publicly, as of Oct 24 2026.
Sources
Marathon Fusion; Rutkowski, Harter & Parisi, arXiv:2507.13461 (2025); Marathon Fusion on X (Oct 21, 2026); CERN (May 8, 2025); Physical Review 60, 473 (1941); Physical Review C 23, 1044 (1981); Euronews (Oct 23, 2026); The Deep Dive (Oct 22, 2026); The Oregon Group (Oct 22, 2026); Forschung und Wissen (Oct 22, 2026); Max Planck Research; Financial Times (Jul 22, 2025), via FT Chinese; Scientific American (Jan 31, 2014); Live Science; LHC Deutschland; World Nuclear Association (updated Oct 23, 2026); World Gold Council, Gold Demand Trends FY2025; World Gold Council, How much gold.
