30 Years of Lessons From Silicon Valley’s Top AI Investor

He outlines four mindsets—traditional finance (markets/metrics as proxies), productive (technical value creates markets), frontier (founder-led “YOLO” into new worlds), and speculative (timing markets more than building businesses). He argues confusion arises when investors mix mindsets and then justify deals based on bias.

He says “network effects” is used inconsistently, and that people confuse the moat’s mechanism (e.g., network effects, brand, scale, marketplaces) with the moat’s shape (superlinear, linear, sublinear with scale). He suggests discussing shape first because defensibility depends on whether the moat strengthens disproportionately as scale increases.

He argues that in fast-growing markets, companies tend to avoid direct competition by moving into new white space, making “moat talk” less relevant; in slowing or shrinking markets, consolidation happens regardless. He uses AI as an example where early “not defensible” objections didn’t stop value accruing while the market grew.

He describes infrastructure as compute, network, storage, databases, and now models, and says AI is refactoring large parts of the stack. He highlights changes like databases becoming easier to use via agents, networking reawakening due to cluster scale/performance constraints, and shifting abstractions for running agents.

He says the key unknown is what happens if the model market slows or saturates: frontier models currently have strong pricing power and capture most spend, but model lifespans may be only “three to six months,” creating high risk if a lab fails to ship. He also distinguishes “task saturation” from “ability to downgrade to cheaper models,” arguing competitive pressure keeps buyers on the smartest model.