The first wave of AI startups tried to build horizontal platforms. Everyone wanted to be the picks-and-shovels layer. Most of them got commoditized by OpenAI, Anthropic, or Google within 18 months.
The second wave is different. The companies gaining traction aren’t selling AI — they’re selling outcomes in specific industries. Legal, healthcare, construction, logistics. The AI is invisible. The workflow transformation isn’t.
Why vertical wins now
Three things have changed:
Models are good enough. You no longer need a research team to build a useful product. You need deep domain knowledge and distribution.
Enterprise buyers are ready. The conversation has shifted from “should we use AI?” to “which vendor do we trust with our workflow?” That’s a buying question, not a research question.
Data moats are forming. A vertical AI company that processes thousands of contracts for law firms has training data and customer feedback loops that a horizontal player can never replicate cheaply.
What I’m watching
The most interesting bets are in industries that are data-rich but software-poor — places where the incumbent software is a decade old and the AI wedge is obvious once you’re inside.
The pattern I keep seeing: founders who spent 5+ years inside an industry, saw the inefficiency up close, and are now building the thing they always wished existed.
That’s the unfair advantage no foundation model can buy.