Seven billion dollars for an AI "button"? 

On Sunday evening, Bloomberg drops the news: Stripe, the company that processes payments for half the sites where you've entered your card details, is about to acquire OpenRouter for more than seven billion dollars.

Seven billion. For a three-year-old company you've probably never heard of.

And the funniest part is that in May, the same company was worth one point three billion. Three months. Multiplied by more than five. There are people who signed a check in May and are waking up with a capital gain they wouldn't dare tell anyone about.

The thing in question doesn't make any AI. That's the whole point.

What is OpenRouter, for those who've never heard of it

Imagine you want to integrate artificial intelligence into your application. A chatbot, a tool that summarizes documents, whatever you want.

Generally, you choose a provider. OpenAI, for example. You create an account and retrieve a key, a long string of characters that proves the call really comes from you and, above all, tells them where to send the bill. Then you adapt your code to the provider's method and pay them directly. The day you want to try a competing model because it costs half as much, you start all over again. Another account, another key, another method, another bill.

OpenRouter has slipped right into that gap. You use a single key, a single bill, and a single way of writing your code. Behind the scenes, you get access to more than four hundred models: those from Anthropic, OpenAI, Google, xAI, DeepSeek, and a long list of others. Want to change brains? You modify one line. Not an account, not a contract, one line.

One socket, four hundred devices. The dream of anyone who's ever traveled with a charger.

Eight million people use it. Around one hundred trillion tokens pass through it every month. A token is a small piece of a word used by these models to measure consumption and bill for it. Roughly speaking, that's three-quarters of a French word. Six months earlier, the volume was five times lower. That's the curve that put seven billion on the table.

Why it's worth seven billion, and not seven hundred million

So let's talk about the commission, because that's where everything happens—and it's funny.

OpenRouter doesn't add a margin to the price of the models. The price displayed by OpenAI is exactly what you pay. However, when you top up your account, the house takes 5.5% along the way. You put in one hundred dollars, and you have 94.50 left to spend on AI.

Now look at what Stripe takes in its core business. For a European bank card, the price displayed on its own website is 1.5%, plus twenty-five cents.

Same action, same second, same money passing through. Not the same share.

You read that right. Stripe has just paid seven billion to acquire a business whose commission is roughly three and a half times higher than its own. In a market that didn't exist in 2022. Seen that way, this is no longer an acquisition. It's a move. Stripe is leaving the room where it takes 1.5% to move into the one where they take 5.5%.

And the irony of it all is that Stripe was already processing OpenRouter's payments. So they had been watching the volumes go by for months, live, on their own dashboard. It's hard to find a better market study: “Hey, this customer is growing strangely fast, maybe we should buy it.”

Three points on a graph. The third arrived this week.

The real news is that nobody won

What interests me is what this amount tells us. For the past three years, we've been sold the model war as a final: there will be one winner, just one, and the rest of the world will go to work for them. Except the company that just sold for seven billion doesn't make any models. It sells the exact opposite: the freedom not to choose.

Its CEO, Alex Atallah (yes, the OpenSea guy; he changed careers and did well), summed up the situation in May in one sentence: the multi-model future is already here. The market has just proven him right with a ten-digit check.

In other words, what is worth the most in AI today isn't having the best model. It's being the door through which people can leave for somewhere else.

And personally, I think that's excellent. An easy way out is the only thing preventing a provider from raising its prices once it becomes indispensable. Ask anyone who tried to leave their mobile carrier in 2005.

We should also admit that what Stripe sees is the future (as young people say: the “future”). Stripe, which already offers Web merchants multi-card, multi-currency payment services, is adding AI-agent payments without human intervention to its tools (and that's already underway), and now, on top of that, we'll be able to choose from several AIs so that everything runs continuously and without outages at lower cost. Because if I want to create a simple transaction and ask an AI to verify that transaction, I don't need to pay Anthropic's Claude (one of the most expensive).

What does this actually change for you?

If you don't code, you may think all of this is happening very far from your kitchen. Not quite.

First, you're already using it without knowing it. A good number of the small AI apps you come across don't talk directly to OpenAI: your accounting software's assistant, the thing that summarizes articles in your browser, the one that corrects your emails. They all go through a gateway of this kind. You use this plumbing every day, just as you use the electrical grid without ever thinking about the transformer at the end of the street.

On the left, when your app is tied to a single provider. On the right, when it can get a divorce in one line.

Second, and this is the real benefit, this competition drives prices down. If the app you pay eleven euros a month can switch engines in one line because a competitor costs half as much, it will. It has no reason to remain loyal. The savings can therefore make it all the way to your bill instead of ending up in someone else's margin.

Third, it makes smaller publishers less vulnerable. If a provider raises its prices, removes a feature, or goes down for six hours, the app switches elsewhere in the meantime. You won't even notice the difference. That's new. Two years ago, an outage at a major provider brought thousands of services to their knees all at once.

The deadline, honestly? It’s already in place. This isn’t a laboratory promise: it runs, it’s measured, and it’s being bought for seven billion.

The downside, because there is one

The neutral referee now belongs to someone.

The service that tells you “here are the four hundred models, choose whichever one you want” now belongs to a company that makes its living by taking a percentage of the flows. As long as the rankings remain honest, everything is fine. But if one model is ever favored for the wrong reasons, no one outside the company will be able to prove it.

I’m not saying that will happen. Stripe has a fairly good reputation in this area, and ruining that reputation would cost it more than it would bring in. I’m just saying that the question didn’t exist last week, and now it does.

And one important clarification: as I write these lines, no one has officially confirmed anything. The information comes from Bloomberg, which cites sources close to the matter, and it was subsequently repeated everywhere. When asked, Stripe replied that it does not comment on rumors. The final price may still change.

What I think

There’s something I like about this story, and it’s not the check.

It’s that seven billion is being paid to guys who simply created a simple service, a button, a plug, an adaptation layer. Something that creates nothing, thinks nothing, and simply translates. All of computing is made up of these kinds of things, and they’ve never received the slightest bit of glory. The printer driver, the plug adapter, the format converter. No one writes articles about them.

Except that when an entire industry starts changing providers every six weeks, the adapter becomes the most strategic place in the room. This isn’t the first time. The same scenario repeats whenever a market has too many providers and no standard: the guy selling the universal extension cord makes more than those manufacturing the devices.

If the $7 billion transaction goes through, hats off to the people behind OpenRouter, and safe travels to the Maldives.