A startup says it turned mercury into gold. Here's what that really means

What happened

On October 21, 2026, Marathon Fusion posted a three-sentence announcement on X. The first line is the one that spread:

The same post says the company enriched isotopes in what it calls the world's most powerful plasma centrifuge, and that it is now building a fusion reactor in San Francisco "to achieve record power density." In other words, the reactor that would make gold at scale isn't running yet.

The rest of the thread changes the subject to medicine. "Millions of lives depend on rare medical isotopes," Marathon wrote next, and a later post describes using the same kind of reaction to turn zinc into copper isotopes used in cancer diagnostics.

What's missing matters as much as what's there. Neither the post nor Marathon's alchemy page gives a date, an amount of gold, or any data from the demonstration. Everything about the experiment itself is, for now, the company's claim.

The idea isn't new to Marathon either. In July 2025, three of its researchers, Adam Rutkowski, Jake Harter and Jason Parisi, posted a paper laying out exactly this route from mercury to gold inside a future fusion power plant. And the record Marathon says it beat belongs to CERN, which reported on May 8, 2025 that the Large Hadron Collider had been turning lead into gold.

How it works

What makes gold gold is a single number: every gold atom has 79 protons in its nucleus. Mercury, its neighbor on the periodic table, has 80. Change that number and you change the element.

Chemistry can't do it. Chemical reactions only shuffle the electrons on the outside of atoms; the nucleus never notices. To change the proton count you need a nuclear reaction, and that is what fusion supplies. When deuterium and tritium, two heavy kinds of hydrogen, fuse together, each reaction throws off a neutron carrying about 14 million electron volts (14 MeV) of energy. That's a very fast particle with no electric charge, so it can plow straight into a heavy nucleus.

Step through what happens when one of those neutrons meets mercury-198, the isotope (version of mercury with a particular number of neutrons) that Marathon's paper builds on.

Notice where the path ends on the chart: directly below where it started. Mercury-198 and gold-197 have the same 118 neutrons. The net effect of the two steps is that one proton disappeared, which is precisely the one change that turns mercury into gold.

The second step isn't instant. Mercury-197 has a half-life of 64.14 hours, about 2.7 days: in any pile of it, half the atoms become gold in that time, then half of what's left in the next 2.7 days, and so on. Drag the slider to watch a batch of 100 atoms age.

How CERN did it differently

CERN's lead-into-gold result worked another way. In the Large Hadron Collider, lead nuclei that narrowly miss each other get hit by intense bursts of light from each other's electric fields. Sometimes that knocks three protons out of a lead nucleus (lead has 82), leaving gold. CERN says about 86 billion gold nuclei were made this way at the collider's four big experiments during its second run, from 2015 to 2018, with a peak rate of about 89,000 nuclei a second at the ALICE experiment. That is 29 picograms, or 29 trillionths of a gram, and each gold nucleus lasted only a tiny fraction of a second. Marathon's route is different in kind: it ends in ordinary, stable gold atoms that stay put.

Ten times what, exactly?

"Beating CERN's record by a factor of 10" can be read two ways, and Marathon hasn't said which it means or published an amount.

Either way, a record "by a factor of 10" over CERN is a record in a contest measured in specks you could never see. That says nothing bad about the physics. It just means the demonstration, as described, is proof that the method works, not a source of gold anyone could sell.

The scale: why gold isn't about to crash

The real claim to weigh is the one in Marathon's 2025 paper: a fusion power plant whose neutrons pass through a layer of mercury could make "approximately 2 t/GWth/yr," about 2 tonnes of gold a year for every gigawatt of fusion heat, without cutting its electricity output, the authors say.

That's a projection for power plants that don't exist yet. To see what it would take to matter, set the amount of fusion heat below and compare it with the world's gold.

At the paper's rate, one gigawatt of fusion heat buys 2 tonnes a year, about 0.06% of what the world's mines dug up in 2025. Matching a single year of mining would take around 1,650 gigawatts of fusion heat devoted to it. Fusion power plants of that kind don't exist yet, and Marathon's own reactor is still under construction.

Marathon's own business case depends on gold staying valuable, too. Its alchemy page pitches gold as a co-product that "can approximately double the revenue" of a fusion plant.

What we checked ourselves

The physics comes from CERN, the IAEA's nuclear data tables and Marathon's own paper; the mining figures come from the US Geological Survey. We didn't discover anything here. We did rerun the arithmetic behind the scale claims with a short script, to check the numbers in our video before we posted it:

• 290 picograms. Ten times CERN's 29-picogram total is 290 picograms, or 2.9 × 10^-10 grams.
• 1,650 gigawatts. USGS's 3,300 tonnes of 2025 mine output divided by the paper's 2 tonnes per gigawatt-year is 1,650 gigawatts of fusion heat.
• 2.67 days. The IAEA's 64.14-hour half-life for mercury-197, divided by 24, is 2.67 days.

All three matched. The peak-rate reading in "Ten times what, exactly?" is extra arithmetic shown in full on this page, using CERN's published rate.

Reactions

Much of the reaction on X was "gold is dead" hype and crypto jokes. Two widely followed voices took it more seriously.

Timeline

1. 2015 to 2018

   During the Large Hadron Collider's second run, collisions at its four big experiments make about 86 billion gold nuclei from lead: 29 picograms in all.

2. May 8, 2025

   CERN announces that its ALICE experiment has measured the lead-to-gold conversion.

3. July 17, 2025

   Marathon researchers Rutkowski, Harter and Parisi post their paper projecting about 2 tonnes of gold per gigawatt of fusion heat per year.

4. September 2026

   Marathon's demonstration, by the company's own account ("last month"). No date, amount or data has been published.

5. October 21, 2026

   Marathon posts its claim on X. It passes 5.4 million views within a day.

6. October 22, 2026

   Reactions from Aakash Gupta and Brian Roemmele; we post our video explainer on X and YouTube.

What it means

Transmutation is real. Turning one element into another is long-established nuclear physics, and turning mercury into gold with fast neutrons follows from well-measured nuclear data. Nobody needs to take that part on faith.

The demonstration is a claim until there's data. Marathon hasn't published how much gold it made, how it measured it, or how pure it was. A paper or independent measurement would settle that; a post on X doesn't.

The money is in the scale, and the scale is far away. Even on the paper's own numbers, gold from fusion would take hundreds to thousands of gigawatts of fusion heat to rival mining, and Marathon's first reactor is still being built. Nothing in this announcement changes how much gold there is to buy and sell today.

Watch the isotopes. Marathon's thread now leads with medical isotopes made by the same kind of reaction, where tiny amounts can be valuable. Aakash Gupta argues that is the nearer business, with gold as the bigger, later prize.

Questions people are asking

Updates

1. October 24, 2026

   This page went up. It covers everything known as of October 22, 2026, when Marathon's claim was spreading.