Somebody made gold in a reactor

Alchemists spent two thousand years trying to make gold and failed, and we took the wrong lesson from it.

The lesson we took was that gold is scarce. The lesson available was narrower: nobody had worked out how yet.

What was actually announced

On 21 October 2026, Marathon Fusion said it had produced stable gold from mercury using fusion neutrons — at roughly ten times the quantity CERN recorded in 2025. The demonstration reportedly took place in September.

The physics is not exotic. Fast neutrons from deuterium–tritium fusion strike mercury-198; it becomes mercury-197; that decays to gold-197, which is ordinary, stable, indistinguishable gold. The mercury sits in the reactor blanket, the same place the reaction breeds its own tritium fuel.

Now the caveats, because they are load-bearing and most coverage skipped them. The company published gamma-ray spectra but no mass, no purity figure and no cost. The spectra do not establish how much gold was made, how pure it was, or whether any of it scales. At the time of reporting it had not been independently verified. This is an announcement, not a result you can check.

The numbers, at the right scale

It is worth putting all of this on one axis, because the headline and the reality are several orders of magnitude apart.

CERN’s ALICE experiment produced roughly 29 picograms of gold across four years of running the Large Hadron Collider. Ten times a picogram-scale result is still a picogram-scale result.

The tonnage figures everyone quoted come from somewhere else entirely: a 2025 paper modelling about 2 tonnes of gold per gigawatt of thermal power per year. The flashier “5,000 kg a year” number is derived from a 1.5 GW thermal simulation producing 2,953 kg a year, restated per gigawatt of electrical output — a different denominator, which is how a modelled 2 becomes a marketed 5.

Those are simulations of a plant nobody has built. Marathon’s current reactor runs at around one kilowatt of fusion power. That is a millionth of a gigawatt.

Against that: global gold mine supply was about 3,672 tonnes in 2025 on World Gold Council figures — a record. The US Geological Survey puts the same year nearer 3,300 tonnes, and the gap is a methodology difference worth knowing about rather than averaging away. Either way, a hypothetical gigawatt-scale plant producing 2 tonnes a year would add well under a tenth of one percent of annual supply.

The gold price is not under threat from this. Anyone telling you otherwise is running ahead of the data. What changed is not the supply. It is the category.

What gold’s scarcity was made of

Here is the part worth sitting with, and it has nothing to do with whether this particular startup is right.

Roughly 212,000 tonnes of gold have ever been mined. Nobody promised that. There is no protocol, no issuer, no document anywhere committing the universe to that number or any other. Gold’s scarcity is an observation about how hard it has been to get more — and every observation of that kind is a statement about current cost, not about permanence.

That has always been true. What makes this month interesting is that it stopped being purely theoretical. The constraint on making gold was never a law forbidding it; it was an energy bill nobody could pay. Energy bills move.

So the honest description of gold is not “scarce forever.” It is: scarce as long as making it costs more than buying it. That is a real property, it has held for all of recorded history, and it is probably good for a while yet. It is just a different kind of claim than most people think they are making when they say gold is scarce.

Scarcity by rule

Which brings us to the thing this blog actually cares about.

Blockspace — room in a block to put a transaction — is also scarce, and it is scarce for an entirely different reason. Not because producing more is expensive, but because a rule says how much there is. The amount per block is a parameter. It is published in advance, enforced by every node running the software, and checkable by anyone who cares to look.

You do not have to trust an assay, a vault or a mint. You do not have to believe anybody’s cost estimate. The supply is a number in a specification that thousands of independent machines agree on, and if it changed you could see it change.

That is a genuinely different foundation. One scarcity rests on difficulty, which falls. The other rests on agreement, which holds exactly as long as the agreement does.

The honest counterweight

If the argument stopped there it would be a sales pitch, so here is the other half.

A rule is only as durable as the people who can change it. Parameters get changed. Chains fork. Governance votes pass. Plenty of networks have altered supply schedules, block sizes and fee mechanics, and some did it for excellent reasons. “Enforced by consensus” means exactly that — the consensus can decide otherwise.

So rule-based scarcity is not stronger in some absolute sense. It is differently fragile. A cost curve erodes quietly over decades and you learn about it from a press release, which is roughly what just happened with gold. A parameter change has to be proposed, argued over in public and adopted by people who each choose whether to run the new software. Both can fail. One of them fails where you can watch.

And a rule you cannot audit is worth nothing at all, which is why the verifiability matters more than the number. A supply cap you have to take someone’s word for is just an issuer’s promise wearing a technical costume — a point we have made before about tokenized assets.

Why this is the interesting rush

Call it a gold rush if you like, but notice what is actually being claimed.

In an actual gold rush, people raced to a place because the resource was physically there and whoever dug first got it. The scarcity was geological and the claim was a hole in the ground. What is being competed for now is room in a ledger — and the claim is not a shovel, it is a rule that everyone can read.

That is a strange and underrated shift. For the whole history of money, the hardest thing about a scarce asset was proving the scarcity: assays, vaults, audits, trusted mints, and a long tail of people who turned out to be lying about what was in the basement. Scarcity you can verify yourself, without trusting the person telling you about it, is new. It is newer than crypto prices, and it will matter longer.

None of which makes anything a good investment, and we are not telling you to buy anything. The useful takeaway is a question to ask of whatever you hold: what actually enforces its scarcity — physics, cost, a rule, or somebody’s promise — and what would have to happen for that to stop working?

For gold, the answer turned out to be “a sufficiently cheap neutron source.” That is much further away than the headlines implied. It is also no longer hypothetical, which it was a month ago.

Swop is fully self-custodial — keys are generated and held on your device, and Swop never holds them — and it runs on Solana, Ethereum, Base and Polygon. Tokenized real-world assets on those chains, PAX Gold among them, can be swapped in the app through its existing routing like any other token. Whether you want to hold the metal, the blockspace, or neither, the part we think is worth insisting on is that you can check what you hold and move it without asking anyone.

FAQ

Has someone actually made gold in a fusion reactor?

Marathon Fusion announced on 21 October 2026 that it produced stable gold from mercury using fusion neutrons, at about ten times what CERN recorded in 2025. Treat it carefully: the company published gamma-ray spectra but no mass, no purity and no cost, and it had not been independently verified at the time of reporting. The mechanism is real physics; the quantity and the scaling are not established.

Does this threaten the gold price?

Not on today’s evidence. The tonnage figures are simulations of plants nobody has built — about 2 tonnes per gigawatt-thermal per year in the 2025 paper — against roughly 3,672 tonnes of mine supply in 2025 (World Gold Council; USGS says nearer 3,300). The current reactor is around one kilowatt. The principle moved; the supply did not.

What is the difference between gold's scarcity and blockspace scarcity?

Gold is scarce because making it has been too hard and too expensive — an observation, not a promise. Nobody guaranteed the total would stop near 212,000 tonnes. Blockspace is scarce because a rule says so: a parameter, published in advance, enforced by consensus and checkable by anyone. One depends on difficulty, which falls. The other depends on agreement, which can be renegotiated but not quietly eroded.

Isn't rule-based scarcity just as fragile?

Differently fragile. Rules get changed by whoever controls them — forks, parameter changes, governance votes — and many chains have changed theirs. The difference is visibility: a cost curve erodes silently and you find out afterwards, while a parameter change has to be proposed and adopted in public. Neither is permanent. One you can watch.

What should I take from this?

That “scarce” is not one property. Ask what enforces the scarcity of anything you hold — physics, cost, a rule, or an issuer’s promise — and what would have to happen for that enforcement to stop. The answers differ, and so do the ways they fail.