Essay

Distribution is the new moat: every company is an AI wrapper now

Nobody is paying for the model anymore.

Ask an AI startup what makes it defensible and you used to get an answer about the model. Ask now and the honest ones don’t even try. The model is an API call away from every competitor they have, and the call gets cheaper every quarter.

The clearest price on that shift was set in December. Lovable, a company built on foundation models it does not own, raised a $330 million Series B led by CapitalG and Menlo Ventures at a $6.6 billion valuation, against a reported $200 million ARR. Roughly 33 times revenue. It got there at a speed nobody has matched: $100 million ARR within eight months of launch, doubled four months after that.

Whatever that money was buying, it was not a proprietary model. There isn’t one to buy.

I learned this the expensive way, and it cost me a company. My first startup in 2014 wrote good code and made no money, because I knew nothing about go-to-market. Writing the code was never the hard part. Distribution was. What has changed since is not the lesson, it’s the stakes. When anyone can stand up a credible product in a weekend, the distance between the companies that win and the companies that fold is a go-to-market distance.

GTMfund will tell you the same thing: distribution, not product, is the final moat. That is a go-to-market fund announcing that go-to-market is the moat, so price it like the advertisement it is. Take it as a hypothesis, then go look at what people with no such book to talk are paying in cash.

Here is where I am supposed to defend the multiple, and I am not going to.

33x might be insane. Anyone who lived through 2021 knows that venture capital misprices for a living. Some days it looks less like a failure mode than a job requirement. And this round has the shape of it: Lovable was worth $1.8 billion in July and $6.6 billion five months later, a triple, which is the signature of FOMO at least as much as diligence. I cannot tell you they keep these customers, either. Nobody outside the company can. They have never published a churn number, and the failure mode of the whole category is the user who ships one app and cancels.

But the price can be wrong and the priced thing still be revealing. Cut the multiple in half. Cut it in half again. The composition of what is being bought does not move, and at 33x or at 8x, none of it is for the model, because the model is for rent to anyone with a credit card.

What is left is two layers, and I want to be fair to the first one. There is real product underneath Lovable: deployment, auth, databases, integrations, the unglamorous plumbing that turns a demo into something a company will actually run payroll on. That is hard work and it is not nothing. It is also the layer with the shortest half-life, because it is precisely what a well-funded competitor ships next quarter. Then there is the second layer: distribution, brand, and the millions of people who typed the URL. You can dispute the number. What you cannot dispute is which of those two layers most of it is for.

And the competitor holding the same API key was never the real threat. The one that should keep a wrapper awake is the company that issues the key. Lovable runs primarily on Claude, and Anthropic’s own website says so, in a glowing customer case study. Then in April, screenshots surfaced of an unreleased app builder inside Claude: prompt to full-stack app, live preview, one-click deploy, which is a fair description of Lovable. Anthropic never confirmed it. It did not have to. Lovable’s staff reportedly worked through the night to ship an update.

Sit with the shape of that. Your supplier publishes a testimonial about you, and may be building your product on the side, priced at zero, funded by the API bill it charges you.

Lovable’s defense is that it is not married to any one lab. It routes across providers and can swap engines if the terms turn hostile, which is a real hedge and the right one to hold. But look hard at what that hedge actually covers. Swapping providers protects your inputs: your margin, your supply, your leverage in a renegotiation. It does exactly nothing about a supplier who decides to ship your product to the customers you were counting on keeping. You can change the engine. You cannot change who owns the demand.

That is the strongest form of the bear case, and it does not break the argument. It is the argument, in its harshest available form. Product scaffolding and switching costs are a moat of sorts; they are simply a shallow one, and a screenshot from your own supplier is enough to let everyone see the bottom. If the platform can credibly threaten to clone the product layer, then nothing on the product side was ever a durable moat, and the only question still standing is whether those millions of people came for an app builder or came for Lovable. That is a distribution question. It was always a distribution question. The labs are simply the ones who get to ask it first, and loudest.

And demand comes in two shapes, which is the part the wrapper argument keeps missing. There are the millions who typed the URL and came back the next weekend. Then there are the Klarna and Uber and Zendesk contracts, where someone still had to sit in a room and be believed. A lab can clone the builder. It cannot inherit either one of those.

Every company is an AI wrapper now. That stopped being an insult the moment it stopped being a distinction.

You can rent the intelligence. You cannot rent the smile or the handshake.

Sources: