Competition Is for Losers

“We have no competitors” is the one sentence on a pitch deck that will never cease to make me laugh. This week I played the first ten minutes of Peter Thiel’s 2014 lecture, Competition Is for Losers, to my entrepreneurship class at UT Dallas. I picked it because it explains the mistake better than I can, even though its title is how a lot of founders learned to make it in the first place.

Thiel’s point is that every founder lies about their market, and the lies run in opposite directions (transcript). Companies that actually have a monopoly describe their market as enormous so nobody notices they own it. Google, in his example, would rather call itself a small player in global advertising than the dominant player in search, because one of those descriptions invites regulators and the other doesn’t. Companies without a monopoly do the reverse and shrink the market until they’re the only thing in it. His example is a restaurant that pitches itself as the only British food place in Palo Alto. That’s technically true and completely useless, because the restaurant still competes with every other restaurant in town for the same dinner budget.

The Lie Everyone Tells

Then Thiel gets to the startup version, which is the part I wanted the room to hear. The worst pitches take a pile of buzzwords… sharing, mobile, social… combine them, and wrap a narrative around the intersection. Whether or not there’s a real business underneath, he treats the pattern itself as a bad sign. His line for it is that the something of somewhere is mostly the nothing of nowhere, like being the Stanford of North Dakota. You’re one of a kind, and you’re still not Stanford.

In 2014, that buzzword stack at least described something the founder built. “Mobile” meant you shipped an app, and “social” meant you had a graph of users talking to each other. The adjectives were marketing, but they were marketing for work the team actually did, and a customer couldn’t go buy “mobile” somewhere else and assemble the product at home. The lie was about how big the market was, while the product underneath it was still something the team had made.

The 2026 stack is AI-powered, agentic, orchestration, infrastructure, and the “layer” for whatever the deck happens to be about. The pattern works the same way Thiel described, with newer adjectives. You take a small market, bolt the stack onto it, and end up with something like the only AI tool that previews how an aftermarket part will look on your specific car. I see some version of that sentence in a lot of the decks I read, and it’s the British restaurant with a new menu.

The Adjective Is Retail

The difference between 2014 and 2026 is where the adjective comes from. Most of the value in an AI-powered product comes from a model the startup rents. The customer can rent the same model. Actually, the customer doesn’t even have to rent it, because the lab gives a version away for free in a chat window that’s already on their phone. So the adjective that makes the pitch sound unique is also the name of the customer’s cheapest alternative.

Every AI-powered pitch now has a free competitor that runs on the same technology and is sold by the startup’s own supplier. That’s an awkward thing to share with a competitor, because the supplier sets your costs, and it also decides what the free version does next. The incentives aren’t aligned in your favor there. The lab makes money whether the customer pays you or pays them directly, and it has every reason to make the direct option better.

I teach this with Juicero. Juicero raised about $120 million to sell a Wi-Fi connected juice press, and then Bloomberg reporters squeezed the juice packs by hand and got roughly the same juice (NPR, summarizing Bloomberg). The competitor was your hands. At least two investors were reportedly surprised to learn the packs could be squeezed at all, which tells you nobody on the diligence side had tried. For a lot of AI-powered startups, the hands are a prompt box. If a founder hasn’t typed their own pitch into the prompt box and compared the output to their product, they’re sitting where the Juicero investors sat, and I’d rather find that out before the wire than after.

So the 2026 pitch tells the same small-market lie Thiel described, and then adds a second one. It says the competitor list is empty when the most obvious competitor is the ingredient printed on the label.

Where Thiel and I Part Ways

I disagree with Thiel on the part that made the lecture famous. His argument is that competition destroys profit, so a founder should aim for monopoly and avoid competitors entirely. As strategy for building a company, there’s a lot to that… nobody wants to open the hundredth restaurant in Palo Alto. As guidance for how a founder talks about their market, it’s done real damage. A generation of founders heard “competition is for losers” and translated it into “we have no competitors,” which brings us back to the sentence that makes me laugh. It tells me either nobody has the problem or you haven’t looked, and I can’t fund either one.

Every customer is already doing something about the problem. Sometimes it’s another company. More often it’s a workaround, like a spreadsheet, a group text, or emailing files to yourself. Sometimes the customer does nothing and lives with the cost, and that cost is the ceiling on what anyone will ever pay you. The competitor founders forget to list is almost always free. I think the fair reading of Thiel is that the losers are the founders who take his title literally and leave those rows off the slide.

In 2026 the competition table needs one more row. It’s the row where the customer skips the startup and uses the model directly.

A free alternative doesn’t automatically win, though, and Dropbox is the proof. When Dropbox launched in 2007, the top comment on Hacker News explained that you could build the same thing yourself with FTP and version control, and pointed out that most people just email files to themselves. Both of those alternatives were free. Dropbox won anyway because the free options weren’t actually cheap. FTP cost an evening of setup, and emailing yourself cost you twelve copies of final_v3. Free only wins when it’s also easy for the customer, which was true of Juicero’s hands and wasn’t true of Dropbox’s FTP.

To be fair to the chat window, FTP was retail too. Anyone could set it up. What’s different now is that the free alternative runs on the same technology as the product, and the supplier keeps shipping the thing startups were charging for. I wrote about this a few weeks ago. Healthtech startups spent two years wiring Epic into language models, and on September 1 OpenAI and Epic shipped that as a feature. Dropbox’s FTP competitor never got a partnership announcement. The chat window gets one every few months, and the company making the announcement is the one you’re paying for tokens.

What Survives the Chat Window

The question I ask an AI-powered company, then, is what makes typing the prompt yourself expensive for the customer. The good answers tend to share a structure, which is that the product lives somewhere the chat window can’t reach.

A team in the class is building a tool that texts the best-matched customer when a salon appointment cancels. A model could write that text in about a second. It can’t see the salon’s booking calendar, though, and it doesn’t know which regulars want Tuesday afternoons, and it isn’t going to send the message at 9:02 when the cancellation lands at 9:01. That team’s real competitors are the waitlist inside the booking software the salon already pays for, the owner texting regulars by hand, and the empty chair. None of those rows is ChatGPT. The pitch comes down to the price of the empty chair, meaning one cancelled slot at the shop’s average ticket times cancellations per week, against a monthly subscription. They can put real numbers on that, and it’s a fight a small company can win.

This is also what I mean when I say Mimosa invests in infrastructure. Real infrastructure has Thiel’s economics. It costs a lot to build and very little to serve the next customer, and the position gets harder to copy as it grows. Integration alone doesn’t get you there… the companies that died were integrated, they just didn’t own either end of the pipe. What holds up is owning something the lab would have to rebuild or take responsibility for, like the system of record itself, regulated permissions, liability someone has to carry, or distribution the lab doesn’t have. A company that calls itself infrastructure because it sits between a model and a user is the British restaurant describing itself as a food logistics platform.

Thiel’s British restaurant was one of a kind, and it still competed with every other place serving dinner in Palo Alto. The 2026 version is the only AI-powered British restaurant in Palo Alto, and it also competes with whatever the customer can cook from a recipe the chat window handed them for free.