We're in the middle of an accelerating AI cycle, where the dominant investment thesis has been: find PhDs spinning out of Stanford, MIT, or CMU, write them $50M checks, or $100M sign-on bonuses, and hope their transformer research turns into the next Google.
I take a different view: most researchers don't make great entrepreneurs. Breakout companies are rarely built by brilliant academics working in isolation. They're built by well-rounded operators — people with deep technical skills, strong product sense, and the ability to hire a great team that can build products customers actually want to pay for.
The current AI investment model resembles the old semiconductor playbook: back the scientist with the breakthrough and figure out commercialization later. In semiconductors, that worked — physics breakthroughs translated directly into durable moats through massive capital requirements and specialized manufacturing.
That's mostly been true in AI so far. Research-driven investments into foundation models have paid off — but the moat is shrinking. Open source is catching up fast, and "good enough" is already here for many mainstream use cases.
I'm not saying AI research labs don't matter — they absolutely do, and we're all thankful for the foundational work they enable. But just as most early tech companies didn't work on transistors, they accrued value from them. The same will happen in the age of AI, where most of the long-term value creation will come from leveraging AI, not building technology in search of a customer.
In this view, AI investing is just software investing, but the founders worth backing are AI-native founders — operators who treat GPT, Claude, or even AGI like AWS or Stripe: powerful infrastructure that lets them focus on the hard part of solving real problems for customers.
The next Salesforce or Facebook won't be a foundation model — but it will sure be built on top of one.
The future belongs to those who ship products, not just publish papers.
/k