Models: cloud, not Windows

The most rational take on AI you’ll hear this year

Topic 9 of 18

Published May 31, 2026 · Clip 38:13–40:41 (2:28)

Benedict restates the deterministic thesis: models don't seem to have network effects, so there's no winner-take-all dynamic and competition persists; without radical product differentiation there's no pricing power. Since thousands of different applications can't all be built by the model companies — 'just as they didn't all get built by Microsoft' — the industry should look more like cloud than Windows. He adds epistemic humility (almost everyone predicting mobile in 2000 would have missed Apple and Google) and recalls a senior banker's verdict on a dot-com component reseller — 'It's a low margin reseller, one-time sales' — concluding model labs look like undifferentiated commodity infrastructure providers, like flat-panel screens with Nobel-prize science but commodity margins, while noting he'd enjoy being proven wrong.