The Conversation is happening right now — in every startup office, at every VC dinner, in the hallways of every conference. It comes down to one question: will the AI labs steamroll our company?
The short answer is yes.
OpenAI, Anthropic, and Google hold enormous advantages — capital, distribution, talent, and unlimited access to the models themselves. But the advantage founders should worry about most isn't any of those. It's that, unlike the incumbents of past technology shifts, the labs are wide awake.
Why This Time Is Different
In previous platform shifts, the incumbents were constrained by their own success. They had org charts to defend, business models to protect, and a deep institutional reluctance to disrupt themselves. That hesitation is what left room for startups.
This time it's different. At OpenAI, any curious person can start a new initiative. At Anthropic, they're spinning up teams to go after entire industries one at a time. At Google, they're disrupting themselves on purpose — adding AI to every product they own and aggressively partnering with startups rather than falling behind. If your idea is big and obvious, one or more of these incumbents will go after it.
So what should you do about it?
All's not lost — but you need to be strategic. The move is to choose your ground deliberately — to pick problems where the labs' advantages don't apply, or where their size and incentives work against them.
Choose a Small or Contrarian Idea
Choose a market that looks deceptively small today but is poised to grow quickly — or an idea that most people, including the labs, write off as wrong. The labs triage opportunities by size, so they will dismiss a small market as too small to matter and a contrarian idea as a mistake.
That dismissal is the gift. It buys you years of air cover to build, find product-market fit, and compound with no serious competition — while the market grows underneath you into something worth winning. By the time it is obviously big, you are the incumbent. Being early and alone is uncomfortable, but it beats being right alongside three of the best-funded companies in the world.
Enter a Market They Can't
Some markets are effectively off-limits to incumbents. A few are blocked by brand risk: a company at the scale of OpenAI or Google cannot ship a product that occasionally fails in catastrophic, headline-making ways.
Others are blocked by the innovator's dilemma: serving the market would cannibalize a business the lab already depends on, or demand a go-to-market motion that conflicts with how they sell today.
Build in the Real World
The labs are software companies to their core — they ship bits. Anything that depends on operations, physical logistics, real-world supply, licensed professionals, or hard-won offline relationships sits outside their natural strengths. Hardware, deep tech, defense, and robotics sit there too — capital-intensive, physically demanding categories that a better model does not flatten.
The messy, unglamorous, operationally heavy parts of a business are a moat precisely because the labs do not want to build them. Here you'll be competing with typical incumbents — often being sold to by the labs, but at least not AI-native themselves.
Build a Moat
Some companies survive the steamroller not by hiding from it but by being too solid to flatten. The defense here is a moat — an advantage that compounds while you operate and that a better model cannot reproduce on its own.
The labs can match your features but they cannot match a proprietary dataset, a real switching cost, or a decade of trust on the day they decide to enter. If you are building in a market the labs may eventually want, make sure that by the time they look, you already have a data, network or brand advantage they cannot replicate.
Be Extremely Ambitious & Focused
Pick an extremely ambitious idea, not an incremental one. A genuinely ambitious idea takes years of focused, single-minded effort, and a lab spread across a dozen fronts will struggle to match that focus.
Ambition is also what attracts the talent and capital you will need to win. The labs concede very little. The only way to beat them on a problem is to want it more — and to organize your entire company around wanting it.
Stay Out of Their Path
Above all, choose ideas that are not sitting in the obvious path of the labs. Some categories are squarely in that path: personal AI assistants, AI employees, and AI versions of the products people already use every day — like email, documents, spreadsheets, or the browser. If your pitch is a better version of something the labs themselves use every day you are volunteering to be steamrolled.
This is also where tar pit ideas cluster — ideas that sound obvious to everyone you pitch and still bury a long list of well-funded teams. It's worth reading tarpitideas.com before you commit to a consensus idea.
Final Thoughts
None of this means a startup cannot succeed doing something on the labs' path. Some will — but the degree of difficulty is much higher.
The labs' awareness, their resources, and their breadth all create a particular shape of opportunity. They will go wide. They will chase what is big and obvious now. That leaves the small-but-growing, the off-limits, the contrarian, the operationally hard, and the wildly ambitious — for you.
So before you write the first line of code: have you chosen an idea the labs will leave alone long enough for you to win?