The Mint and the Promise

What a food delivery company just taught the smartest people in Silicon Valley about the oldest problem in money

Brian Connelly · Piketown Press · July 2026

Two weeks ago, the company that brings China its dinner released one of the largest AI models ever built and gave it away.

Not leased it. Not licensed it with an asterisk and a lawyer attached. Gave it away, MIT license, posted on the internet, free for commercial use, forever. Meituan trained its frontier AI model LongCat-2.0, all 1.6 trillion parameters of it, on 50,000 Chinese-made chips, without a single Nvidia GPU anywhere in the building, and then handed the whole thing to planet Earth like a mint on a hotel pillow.

The coverage got stuck on the chips, and fair enough, the chips are a real story. American export controls were supposed to make this impossible, and here it is, existing anyway, which tells you how well artifact-based control works against people who are motivated and good at math. But the chips are the second most interesting thing that happened. The most interesting thing is the giving away.

A company spent a fortune building something and then destroyed its own ability to charge for it. On purpose. And everyone in Washington and San Francisco is squinting at this like a dog watching a card trick. To understand the move, and what it does to every American AI lab currently charging rent on intelligence, you need a concept that’s about four thousand years older than the transformer.

But first you need to know what, exactly, got given away. Bear with me for three paragraphs. This is the part the tech press always skips because they assume you already know, and the part that matters most if you don’t.

The recipe is the restaurant

An AI model sounds like a machine, or a service, or a brain in a jar somewhere in Virginia. It’s none of those. Strip away the mystique and a model is a file. A gigantic file full of numbers, and nothing else. Those numbers are called weights, and you can think of them as 1.6 trillion little dials, each one turned to a very specific setting.

Nobody sets those dials by hand. You find the settings by burning warehouse-loads of electricity for months, showing the system most of the written internet, and letting the dials adjust themselves a hair at a time until something spooky happens: the file, plus a modest amount of computer to run it on, can write your code, draft your contracts, and explain photosynthesis to your grandson. All the intelligence, every capability the demo videos brag about, lives in the settings of those dials. The training run costs a fortune. The file it produces costs nothing to copy. It is, quite literally, the secret formula, and everything else is packaging.

So when I say Meituan open-sourced the model, I mean they published the formula. Not a description of it. Not a demo of it. The actual dial settings, the finished product of the whole nine-figure effort, downloadable by anyone with a fast connection and a reason. Imagine Coca-Cola spending a hundred years guarding the recipe in a vault in Atlanta, and then one Tuesday morning printing it on the side of every can. Now imagine they did it on purpose, smiling, and the question of this essay is why anyone would smile while doing that.

To answer it, we go to the mint.

The formula, out of the vault.

Seigniorage, or: the king’s cut

For most of recorded history, whoever ran the mint took a cut. You hauled your silver to the king’s mint, the king stamped his face on it, and the coin that came out was worth a little more than the silver that went in. That markup has a name, seigniorage, and it’s one of the oldest rackets in civilization, old enough that we named it after the guy collecting it. The king wasn’t selling you silver. You brought the silver. He was selling you certainty. The stamp says this coin is what it claims to be, and the stamp is worth paying for exactly as long as nobody else can make one.

Read that clause again, because it’s the entire business model. Seigniorage works while the mint is scarce. The moment anybody can stamp coins, the premium is gone and the king is just a guy with a hammer and an inflated sense of his own face.

Now look at what an AI lab actually sells. It mints capability. Electricity and data go in, a file full of dial settings comes out, and the lab charges you the difference between what its file can do and what the best free file can do. Every API dollar is rent on that gap. The frontier labs are mints, the subscription price is seigniorage, and the whole revenue model hangs on the same fragile clause it always did: nobody else can make the stamp.

Meituan made the stamp. Then it did something no king in four thousand years ever did sober. It published the stamp’s design and invited the world to counterfeit at will.

Debasement from outside the mint

When a king wanted to cheat, he debased from inside. Shave the coins, cut the silver with copper, keep the face value, pocket the difference, and let the peasants figure it out at the bread stall. This move is so reliable across history that I’ve written six children’s books and the plot of every one is some grownup discovering it works.

What’s happening to the AI labs is debasement from outside the mint, and it’s a stranger beast, because the debaser doesn’t even collect. Every free release at 90 percent of frontier quality expands the world’s supply of good-enough intelligence and shaves the premium the frontier can charge. The floor rises. The floor is priced at zero. The floor does not go back down. Ever.

And here’s the part that should have pricing strategists chewing their pens. The companies raising the floor cannot be made to feel the pain they’re inflicting, because the model was never their product. Meituan sells dinner. Meta sells ads. Alibaba sells cloud and everything else. For all of them, intelligence is a complement to the actual business, and the oldest play in the strategy book is to commoditize your complements so the money pools in the thing you actually sell. They’re giving away Anthropic’s product the way a casino gives away drinks. The drinks aren’t the business. You cannot win a price war against an opponent who isn’t in the price war. There is no move that makes them blink. There’s nothing back there to blink.

A normal competitor’s price cut gets reversed next quarter, once the board sees the margins. An MIT license cannot be reversed by anyone, including the company that granted it. That’s the detail everyone skims. Meituan didn’t lower a price. It burned the bridge that would let any price ever exist again, and it sold tickets to watch the fire.

What the fork tells you

So the formula is out. The file is copying itself across the internet at the speed of enthusiasm. By the logic of every counterfeiting panic in monetary history, value should be evaporating.

It isn’t. It’s packing its bags and moving. Watch where it goes, because the destination is the whole lesson.

While their formula gets commoditized, notice what the American labs are quietly selling instead. Contracts. Service agreements. Indemnification, which is a lawyer’s word for someone to sue. Safety cases a compliance officer can wave at a regulator. Integration into workflows where a human being answers the phone when the AI agent quietly deletes the wrong database at 2 a.m. Anthropic’s product line has been drifting for two years from selling tokens toward selling finished work, and that drift is not a pivot dreamed up in an offsite. It’s a migration. Money made the same trip, and it took the scenic route.

Gold worked as money because atoms don’t photocopy. The promises wrapped around gold could stay thin, a few assayers and a stamp, because physics did the heavy lifting. Then paper came along, trivially copyable, and value had to retreat into the issuer’s promise to redeem. Then the redemption promise itself got suspended, politely, temporarily, permanently, and value retreated all the way down: a modern dollar is a network of promises about future acceptance, wearing a green costume. Every step that made the token easier to copy chased the value deeper into the commitments behind it, until the token was just the receipt and the receipt was mostly imaginary.

The AI industry is running the same film at 200x speed with the popcorn on fire. The weights are the token, the dial-settings file. The file now copies for free. So price is fleeing the file and taking shelter in the one place it has ever been able to survive, which is the promise.

Because here’s what nobody can fork. Copy Anthropic’s service agreement word for word, every clause, every comma, frame it and hang it in your office, and you have copied nothing. The agreement was never the words. It’s Anthropic standing behind the words, with a legal department and a balance sheet and something to lose. A promise is only a promise from someone. It carries its speaker the way a signature does, which is exactly why no printer can counterfeit it. Artifacts get copied. Commitments get kept. Price can only live where keeping is possible.

One honest caveat before a sharper reader finds it for me. A counterfeit dollar adds nothing to the world; money’s value is purely positional, so copying it just steals from everyone holding the real ones. A copied model still works. It writes the code, drafts the contract, explains photosynthesis. Copying money destroys value. Copying intelligence destroys price while multiplying use, the way air is priceless and also free. Same destination, value pooling in the promises, but the road is abundance instead of emptiness. Which, if you’re a lab, is worse. Counterfeiters you can arrest. Try arresting air.

Five thousand years, compressed into five

So the labs are learning, on a venture capital timeline, what monetary history took fifty centuries to teach. Value flees copyable artifacts and hides in commitments. Fine. Except monetary history has a second lesson, and it’s the one with teeth: commitments can be revised by whoever holds them.

That’s the whole tragedy of money in one sentence, and I mean whole. The stones stayed put at the bottom of the lagoon; the promises about the stones got renegotiated by men in nicer clothes. The iron money worked until the promise behind it didn’t. Every time value migrated into a commitment structure, it became hostage to whoever ran the structure, and that party, given enough time and one bad war, always revised. Counterfeiting attacks the token. Debasement revises the promise. Two failure modes, five thousand years, a perfect record. Not one issuer, ever, who held the pen and declined to use it.

Which brings us, late and on purpose, to Bitcoin.

Bitcoin’s artifact is the most copyable object our species has ever produced. The code is open source, the ledger is public, and you can clone the entire thing before lunch. Thousands have. Every fork copied the token perfectly and captured a rounding error of the value, and those forks are the cleanest natural experiment ever run on where monetary value actually lives. Not in the token. In the commitment structure: the accumulated proof of work, the consensus around one chain, and underneath everything, the founding act itself.

Because Satoshi’s real invention wasn’t digital scarcity. It was a new species of promise. Every issuer in history kept the pen. Satoshi made a commitment, then structurally destroyed his own ability to take it back, dropped the pen down a well, and walked away, leaving a promise standing that nobody, including its maker, can edit. The first commitment in monetary history that is neither copyable nor revisable. Both failure modes, closed by one design, by an author whose greatest flex was leaving.

The pen, down the well.

Meituan, knowingly or not, borrowed half the move. The MIT license is irrevocable, and the irrevocability is what makes the gift dangerous. A model that could be yanked back would earn cautious adoption, a fling. A model that provably can’t gets married. The weapon was never the parameters. It was the burned bridge behind them.

The American labs are betting the world pays for lawyers, not law books, and it’s a decent bet; law books are nearly free and lawyers most certainly are not. But they should study what happened to every previous institution that collected rent on promises, because the pattern has a final act. First your artifact gets copied and the value flees into your commitments. Then somebody shows up holding a commitment that can’t be revised, and asks the market a very old question in a very new context.

Why rent a promise from someone who can break it?

Somebody already answered that one, in 2009, in nine pages, with no keynote and no logo. The intelligence market is about to hear the answer again, and this time it won’t take five thousand years to sink in.