Artificial intelligence is no longer a single application or a mere ‘chatbot.’ It is a broad technological transformation that is reshaping the structures of the economy, the processes of companies, and the sources of competitive advantage. In earlier letters we described this shift through two frameworks: (Q1/2025) the Gulf Stream metaphor — we focus on the deep currents running beneath the surface rather than the short-term waves — and (Q2/2025) the economics of scalability — in the spirit of Jevons’ paradox and Moore’s law, as costs fall, usage grows exponentially. These underlying forces drive demand for AI infrastructure and create winner-take-all markets.
In the Q2 letter we extended the framework to AI agents: a coordinated software ecosystem in which agents perform multi-step tasks, use tools, and operate where the data already lives. For investors, the conclusion was twofold: (i) the structural demand for infrastructure (compute, memory, cloud, security) strengthens, and (ii) organisations that can connect agents to their existing data will capture a disproportionate share of the value. We made this concrete with the Oracle case: massive, hard-to-migrate enterprise data, and on top of it vector search, RAG and ready-made models — all in the same database.
In this letter we present two new holdings — MP Materials and Duolingo — and explain why they fit precisely into the same core strategy: we invest in probable winners in markets whose dynamics are winner-take-all.