What Changes Now in Building Markets

The founder’s slide deck was immaculate. In a conference room with borrowed air and rented chairs, he pointed to a box labeled "Proprietary Data & Custom Models." This, he explained to the venture capitalists, was the moat. The deep, unassailable competitive advantage that would protect his startup from copycats and giants alike. The VCs nodded, their faces reflecting the cool blue of the projector. They’d seen a dozen slides just like it that week.

That box on the slide is the most expensive fiction in technology right now. For a generation of startups, the moat is not a fortress wall built on a bedrock of unique technology. It is a tollbooth on a highway owned by someone else.

The business model for countless new "AI companies" is a simple arbitrage play. They take an API call that costs them a fraction of a cent from OpenAI or Anthropic, wrap it in a slick user interface, and sell the result for a few cents more. They are translation apps, content generators, coding assistants, and PDF summarizers. Their entire operational existence is defined by a third-party’s terms of service and pricing page.

This is not a moat. It is a dependency with a monthly bill. The ground beneath these companies is perpetually unsteady. What happens when the platform owner, the one running the multi-billion-dollar models, decides to raise its API prices? The tollbooth operator has two choices: absorb the cost and watch margins evaporate, or pass it on to customers and watch them flee.

What happens when the next version of the foundational model bakes in your startup’s core feature for free? Last year’s killer app for summarizing legal documents becomes this year’s native function, a checkbox in the API parameters. Your entire product is suddenly a commodity feature of the platform you rely on. The highway owner just built a free bypass around your tollbooth, and the traffic is gone overnight.

Competition in this environment is a knife fight in a phone booth. When everyone is building with the same Lego bricks, differentiation is a mirage. You can claim your prompt engineering is superior, your UI is cleaner, your brand is stronger. But you are fundamentally selling the same engine as everyone else. The "secret sauce" is a text file of instructions whispered to a model that also listens to all your competitors.

There are exceptions, of course. A true moat can still be dredged. It is filled with genuinely proprietary data—the kind of messy, specific, hard-won information that can’t be scraped from the public web. It is reinforced by deep integration into complex, regulated workflows where the AI is an enhancement, not the whole show. Think medical diagnostics or industrial robotics, not a better way to write marketing emails. A handful of companies are building their own models from scratch, a brutally expensive path reserved for the heavily capitalized or the truly insane.

For everyone else, the current gold rush is a dangerous illusion. They believe they are building castles, but they are merely operating franchises. They are utterly beholden to the whims of their landlord, a landlord who has every incentive to eventually build its own stores on the very plots it leased out. The venture capitalists will learn this lesson. The founders will learn it. The engineers, tasked with patching the roof on a rented shack, are already starting to understand. They aren't building a permanent structure; they are just collecting fares until the road gets rerouted.