Gavin Baker
Technology and growth investor; managing partner and chief investment officer of Atreides Management, a hedge fund he founded after more than a decade running Fidelity’s flagship OTC growth fund. Known for reading the AI cycle through its physical constraints — power, wafers, and compute — and for a historian’s habit of testing today’s boom against past technology bubbles.
Baker built his reputation as a public-markets growth investor with unusually deep supply-chain fluency: he tracks the semiconductor stack from TSMC’s leading-edge nodes down to DRAM pricing, optical interconnect, and the memory-bandwidth economics of GPU inference. He reads markets through cost-curve logic borrowed from commodity investing, through the S-curve dynamics of technology adoption, and through the bubble histories of Carlota Perez — railroads, canals, and dark fibre as precedents for the AI build-out.
Core positions
- The AI build-out has two hard inputs — ‘watts and wafers’. Power is a problem capitalism solves within a few years (and eventually via orbital compute); leading-edge wafer supply, controlled by TSMC, is the true and lasting bottleneck.
- TSMC’s supply discipline may single-handedly prevent a bubble. If it built everything Nvidia wanted, GPU supply would overshoot demand and crash; its capacity decisions are the single variable that signals whether the cycle is overbuilding.
- Economic value at the model layer concentrates almost entirely at the frontier, and access is being rationed through usage-based pricing — a shift ‘from all you can eat to pay by the drink’ that makes flat-rate plans deliver a starved model.
- The largest tail risk to the whole trade is a violation of the Bitter Lesson: human algorithmic ingenuity displacing brute compute. Baker weights it lower than most, reasoning that proximity to ASI may itself bend the rule temporarily.
- For chip and application start-ups, the only durable strategy is ‘different and hard to do’ — wafer-scale computing, prefill/decode disaggregation, or a defensible position in the ‘token path’ — because anything obvious is crushed before it reaches scale.
In the wiki
| Episode | Source | Date |
|---|---|---|
| Gavin Baker on AI Infrastructure, Power Constraints, and Semiconductor Investing | Invest Like the Best | 20 May 2026 |
See also
- Bitter Lesson — the compute-versus-ingenuity question Baker names as the trade’s biggest risk.
- Token Economics — frontier-token value capture and the token path.
- Private Credit — GPU financing and the useful-life debate.
- Patrick O'Shaughnessy — host of Invest Like the Best.