Fusion Turned Mercury Into Gold at 10x CERN's Record. The Record Was 29 Picograms.
Marathon Fusion says it beat CERN's gold record tenfold, but CERN's record is 29 picograms, so ten times that is still invisible. The company's own preprint holds the real story: the gold math only closes because every gold atom creates a neutron.
On October 21, a San Francisco fusion startup announced it had beaten CERN's gold production record by a factor of ten. Two million people saw the post within a day, and crypto analysts started pricing the end of gold scarcity, but nobody asked what the record was.
It was 29 picograms. CERN's ALICE experiment spent four years of Large Hadron Collider running, from 2015 to 2018, producing about 86 billion gold nuclei weighing 29 trillionths of a gram. Ten times that is 290 picograms. Marathon Fusion's claim may be true, but it is also, by any physical measure, almost nothing.
That is not the interesting part. Buried in the company's own 2025 preprint is neutron accounting that actually matters: fusion gold only works, even on paper, because making gold creates neutrons, and a plant that spent its neutrons on gold without getting them back would run out of fuel in weeks.
What Marathon says it did
Marathon's post on X read: "Last month, we demonstrated high-rate alchemy: turning mercury into real, stable gold, beating CERN's record by a factor of 10." Alongside the words, the company published gamma-ray spectra showing mercury-197 and gold-197 signals, recorded 8 to 44 hours after irradiation. What it did not publish: the mass, the purity, or the method. No independent verification has appeared.
Physics behind the claim is textbook: the mechanism is real even if the scale is not. A fast neutron from deuterium-tritium fusion, carrying 14.1 million electron volts, strikes mercury-198, about 10 percent of natural mercury. Two neutrons come flying out, leaving mercury-197, which is unstable. After about 64 hours, half of it has decayed by electron capture into gold-197, the only stable isotope of gold. One proton fewer, and mercury becomes gold. Harvard physicists made radioactive gold from mercury with fast neutrons in 1941, so Marathon's novelty is not transmutation at all: it is stable gold, made with fusion neutrons, at a rate the company claims beats CERN's.
Ten times almost nothing
CERN published two numbers for its gold: a peak rate and a total. Its ALICE detector measured gold production at a maximum of about 89,000 nuclei per second during lead-lead collisions. Marathon never said which number its "10x" refers to, so run both readings.
| Reading of "10x" | Rate | Mass after one month | Mass after one year |
|---|---|---|---|
| 10x CERN's peak rate (89,000 nuclei/sec) | 890,000 nuclei/sec | 0.75 nanograms | 9.2 nanograms |
| 10x CERN's Run 2 total (29 picograms) | n/a | 0.29 nanograms | 0.29 nanograms |
Either way, the answer is sub-nanogram. A grain of sand weighs about a milligram; a year at ten times CERN's peak rate produces roughly 100,000 times less gold than that grain. At the demonstrated rate, making a single milligram would take about 109,000 years. Marathon engineer Jake Harter told Cointelegraph the quantities are "right now still microscopic," with "macroscopic milligram quantities" hoped for in 2027. Getting there needs a rate about 109,000 times the one just demonstrated.
None of this makes the claim false, but it does make it uninformative: a factor of ten on a denominator of 29 picograms is a rounding error dressed as a record.
The neutron ledger
Now the preprint, where the real analysis lives. In July 2025, three Marathon researchers published simulations of a tokamak whose blanket contains mercury-198, modeling a 1,500-megawatt-thermal plant that makes 2,909 kilograms of gold a year while still breeding its own tritium fuel, a combination that should be impossible.
Start with the fuel. Deuterium-tritium fusion burns tritium, which barely exists: the world holds about 25 kilograms, and each power plant needs roughly 10. A plant must breed its own replacement tritium by firing fusion neutrons into lithium, one neutron per triton; the paper states the rule plainly: the breeding ratio must exceed 1, with 1.1 to 1.2 the target. Divert half the neutrons to gold, and the plant starves.
Marathon's design escapes the trap through the (n,2n) reaction itself. Each gold atom consumes one neutron and emits two, a net gain of one neutron per atom. Run the ledger, and the paper's headline breeding ratio falls out.
| Ledger item | Neutrons per second |
|---|---|
| Fusion neutrons produced (1.5 GWth at 17.6 MeV per reaction) | 5.32 × 1020 |
| Neutrons feeding gold (2,909 kg/yr = 2.82 × 1020 atoms/sec) | 2.82 × 1020 (53% of fusion neutrons) |
| Neutrons returned by (n,2n) multiplication | +5.64 × 1020 |
| Total neutron supply | 8.14 × 1020 |
| Tritium breeding at the paper's TBR of 1.19 | 6.33 × 1020 |
| Remaining for structural capture and losses | 1.81 × 1020 |
Three checks confirm the ledger. First, if every spare neutron bred tritium, the maximum ratio would be 8.14 divided by 5.32, or 1.53, and the paper's simulated 1.19 sits at 78 percent of that ceiling, and second, the paper's own split puts 0.694 of the 1.19 in the mercury-lithium channel itself. Third, the counterfactual: without multiplication, sending 53 percent of neutrons to gold would leave a breeding ratio of 0.47, and the plant would burn its tritium with no way to replace it. Gold is not a side business bolted onto a fusion plant; it is the neutron multiplier the plant needs to survive.
Milligrams in 2027
Harter's target can be tested against Marathon's own published hardware numbers, and the test is uncomfortable. Marathon's website lists its current neutron source at 1 kilowatt of fusion power with a neutron flux of 1012 neutrons per square centimeter per second, a rate per unit area rather than a total, so the total depends on the size of the irradiated spot. If the spot is about a square centimeter, roughly a fingertip, the total is about 1012 neutrons per second.
At that output, even with physically impossible 100 percent capture, the ceiling is 10.3 milligrams of gold per year. Producing one milligram a year would require 9.7 percent of all neutrons to convert a mercury-198 nucleus, and realistic capture fractions run in the single digits. Milligrams in 2027 sits at the edge of what the advertised source can physically deliver, not comfortably inside it. If the "1 kilowatt" instead means fusion power, the neutron rate is 3.55 × 1014 per second and the ceiling rises to 3.66 grams a year, but Marathon publishes flux, not total rate.
What the gold is worth
Suppose the plant existed: at 2,909 kilograms a year it would produce 93,527 troy ounces, worth roughly $381 million at November 6's closing price of about $4,070 an ounce. Electricity from the same plant, at 40 percent conversion and $60 per megawatt-hour, comes to about $315 million a year, while gold would be worth about 1.2 times the electricity and combined revenue about 2.2 times electricity alone, independently reproducing the preprint's claim that gold "more than doubles" plant revenue. Matching world mine output would take roughly 1,260 such plants.
Two catches sit in the paper's own fine print. First, the mercury is a capital stock, not a consumable: each plant holds 622 tonnes of mercury-198, and a year's gold is only 0.47 percent of that inventory, so the same mercury lasts about 213 years, but assembling the first fill is the hard part. Second, the gold comes out radioactive, and the paper calculates 6.8 years of cooling before the product falls below Class-A low-level-waste activity, 13.7 years before it can be handled without a radioactive label, and 17.7 years per troy ounce before it is less radioactive than a banana.
Limitations
Marathon disclosed no mass, yield, purity, cost, or methodology for the September experiment, so the "10x" claim cannot be verified from public information; the spectra establish that gold-197 was present, not how much, and the preprint is an OpenMC simulation, not peer-reviewed, for a plant that does not exist. The ledger takes the paper's simulated gold rate as given and assumes steady operation at nameplate power, while real blankets lose neutrons to structure and coolant, and the milligram analysis assumes a one-square-centimeter spot, since Marathon has not published target geometry, and prices are November 6 snapshots, not forecasts.
Strongest case against
Consider the strongest case for Marathon, because it is a good one: this experiment was never about mass. Gamma spectra showing mercury-197 and gold-197 after fusion-neutron irradiation prove the (n,2n) pathway runs with fusion neutrons, and that reaction is the exact mechanism the 2025 paper's economics depend on. CERN's 29 picograms is a deliberately absurd denominator, and beating it tenfold is marketing, but the underlying demonstration, if the spectra are what the company says they are, de-risks the one reaction the whole business case needs. Harvard did it with radioactive gold in 1941; Marathon is the first to do it with stable gold-197 from a fusion neutron source. The right question is not "where is the gold" but "does the reaction run at the rate the simulations assume," and that stays open because the company published spectra instead of rates. Until Marathon publishes a mass, the ledger is the only honest way to judge the claim.
What you can do
If you follow fusion, stop grading announcements by their multiples and start grading them by their denominators, because a 10x improvement on 29 picograms is a rounding error. If you evaluate fusion startups, ask for the tritium breeding ratio before the revenue model, since a plant that cannot breed its own fuel is a press release, not a power station. If you invest in gold or mining, the fusion threat is roughly 1,260 power plants away, so watch the neutron source, not the headline: milligram quantities in 2027 would be the first real signal, and anything short of a published mass is marketing. If you work in nuclear engineering, the (n,2n) ledger is the transferable idea: any transmutation that multiplies neutrons changes the fuel-cycle math for every D-T plant.
Bottom line
Marathon Fusion says it made gold from mercury at ten times CERN's record, and the record is 29 picograms, so the achievement fits inside a rounding error: a year at ten times CERN's peak rate would produce 9.2 nanograms, and a milligram would take 109,000 years. But the company's own preprint contains a better story: a fusion plant making 2,909 kilograms of gold a year spends 53 percent of its neutrons on gold, and the design only survives because each gold atom hands a neutron back, lifting the tritium breeding ratio to 1.19. Without that multiplication, the ratio collapses to 0.47 and the reactor starves. The gold is real physics and unproven economics, and the number that matters is not 10x. It is 1.19.