Most AI startups are at peak value right now. They just don't know it.
Elad Gil, one of the more consistently accurate technology investors, has been direct about this. Most AI startups exist because foundation model companies haven't yet expanded into their specific vertical or use case. That gap is real. It's also temporary.
The Window and Why It Closes
Gil's framework from a recent TechCrunch piece is worth understanding precisely. When GPT-5 or Claude 4 natively handles what your AI startup does, your competitive moat becomes a feature request. The companies that built on top of foundation models without developing proprietary data, distribution, or user relationships are most exposed to this.
His recommendation: schedule dedicated board meetings once or twice a year with a single agenda item. Is this our moment? Are the next six to twelve months when we will be worth the most we will ever be worth? Most boards never have this conversation explicitly. They have the fundraising conversation instead, which is a different question entirely.
The historical examples Gil cites are instructive. Lotus, AOL, Mark Cuban's Broadcast.com. Each was a market leader at the exact moment when the infrastructure beneath their competitive advantage was being rebuilt. The founders who recognized the moment and acted on it preserved value. The ones who kept building toward a larger outcome ended up on the wrong side of a technology transition.
What This Looks Like From the Inside
I've watched this pattern in the agency and SaaS world long enough to recognize it at the early stages now.
A startup that builds vertical AI for legal document review isn't primarily competing with other legal tech companies. It's competing with a clock. The clock says: six to twelve months until a foundation model ships a legal review workflow natively. Maybe longer if the vertical is complex. Maybe shorter if there's been visible investment.
The defensible positions are: proprietary data that the foundation model doesn't have access to, distribution relationships the foundation model can't replicate, and user trust built through demonstrable track record in a specific domain. If you have at least two of those three, your window extends meaningfully beyond 12 months. If you have none, that framing might be generous.
The Cursor fundraise this week is a signal worth reading. A coding assistant raising at $50 billion implies the market believes developer workflow tools have distribution and trust that foundation models can't easily replicate. That may be right. Or it may be a bet placed just before the window closes. Both are possible.
The Founder Psychology Problem
The hardest part of this isn't analytical. It's psychological.
Founders are optimists by occupational requirement. The natural instinct when things are going well is to raise again, hire more, and build toward a larger outcome. The contrarian move is to recognize peak value and act on it.
The companies I've seen exit well in multiple tech cycles share one trait: the founders knew exactly why they were worth what they were worth, and they didn't confuse a cyclical peak with a long-term structural moat.
The infrastructure bets being made by Amazon and Microsoft, billions into Anthropic and OpenAI, are a signal that the foundation layer is being locked in for the next decade. What sits on top of that foundation will be competed away over time. Understanding which side of that line you're on is the most important strategic question most AI startup founders aren't asking directly enough.
FAQ
What is the "12-month window" for AI startups?
It refers to the idea that many AI startups are currently at their maximum competitive value because foundation model companies haven't yet built native versions of their product. As foundation models expand into vertical categories, specialized startups lose their differentiation. The window is the period when valuation reflects potential that hasn't been competed away yet.
How can AI startup founders know if they're at peak value?
The test is defensibility. Ask: if OpenAI or Anthropic ships what we build natively in six months, what do we still have? If the answer is proprietary data, deep distribution, or domain-specific user trust, you may have something durable. If the answer is "we do it better," that's a feature advantage, not a structural moat.
What should AI founders do if they believe they're in the window?
Start having the conversation you've been deferring. That means scheduled board discussions about exit timing, not just growth metrics. It means understanding who your strategic acquirers are and what they would pay. It doesn't mean selling immediately, but it means having the information to make a deliberate decision rather than defaulting to the next funding round.