Startups and the new compute production function
We argue that AI is a cognitive revolution on par with, and likely greater than, the Industrial Revolution, driven by a specialization imperative that compresses decades of progress into years. The $10 trillion US services market is the near-term arena, where AI can both gain share and expand the total pie as cloud did for software. We outline the current investment trends and the next 12–18 month themes we’re backing to build enduring, standalone AI-native companies.
Key elements
- We frame AI as a transformation as big or bigger than the Industrial Revolution, with the “specialization imperative” turning general tools into specialized systems at scale.
- History’s pace-setters: 67 years from the steam engine to the first factory and 144 more to the assembly line—our aim is to compress that cognitive timeline dramatically.
- Tech milestones: the GeForce 256 (1999) as the first GPU and 2016 as the “first AI factory” moment that assembled the core stack to produce AI tokens.
- Market sizing: a $10 trillion US services TAM with roughly $20 billion automated by AI today—a 10^13 opportunity to grow both share and the market itself.
- Cloud analogy: software spend was ~$350 billion with only ~$6 billion SaaS at the dawn of cloud; the market expanded to $650+ billion, signaling AI can similarly expand categories.
- Market structure: the S&P 500 is dominated by a few giants—Nvidia sits near $4 trillion—leaving room for AI-first services companies to become large, standalone publics.
- Portfolio exposure: we’re backing verticals like nursing (Open Evidence, Freed), software development (Factory, Reflection), and law (Harvey, Crosby, Finch).
- Trendline shift: work is moving to 100%+ leverage via AI agents with more outcome uncertainty (e.g., Rocks), while excellence is proven in the real world (Expo topping HackerOne).
- RL and hardware: reinforcement learning is central (Reflection in coding), AI is building and assuring physical systems (Nominal), and compute/FLOPs per knowledge worker are set to rise by at least 10x—potentially 1,000–10,000x.
- Investment themes: persistent memory, seamless communication protocols (beyond MCP, à la TCP/IP’s “starting gun”), AI voice now, end‑to‑end AI security with many agents per user, and a critical push to keep open source competitive.
To remember
If you’re not a specialist, congratulations—you’re exactly who AI wants to supercharge. Start by piloting agents where “mostly right, very fast” beats “perfect, very slow,” measure results in the real world (not just in pretty dashboards), and budget for compute like you’d budget for coffee—because you’ll need a lot of it. Lock down your security (yes, even that mysterious terminal window), embrace voice workflows, and keep an eye on open source—otherwise you’ll be renting your future from the biggest kids on the playground.
Sources
Quiz on the vidéo: 5 questions
