ENTRY 121 has been downgraded to "the wallet is starting to move too": the revenue ceiling of the low-price layer still sits far below what token volume implies, and the gap is narrowing. On Vercel's gateway in June, open-weight models ran 29% of tokens for under 4% of spend; the author's self-computed broad-China spend share averaged 16.1% over thirty days.
The mainstream story says Silicon Valley is switching its base models and reads token share as market share. The diary counts volume and money separately: open weights took the workload, and the wallet has only started to move. Investors ask for spend share because valuation is set on revenue, not on call volume.
Vercel AI Gateway Production Index (June): open-weight models took 29% of tokens and under 4% of spend, while Anthropic took 32% of tokens and 61% of the money. From 8/15 to 9/13, the spend share of 12 broad-China labs averaged 16.1% over thirty days, with a lowest day of 10.4% (self-computed by the author, export and script published), crossing the 10% line. The a16z "80%" figure, after Casado's clarification, is actually about 16-24%.
For founders: stop pricing product value by token volume and charge by outcome or by seat; expanding capacity is fine, but building a three-year revenue model on wallet share is not. For investors: when a deck gives only token share, ask for spend share on the spot, and where the two differ by an order of magnitude, value the company on revenue, not call volume.