Frontier price-cut debate: open-weight share or OpenAI–Anthropic competition? The data can't yet tell

Why it matters

【Key points】

Gavin Baker (X bio: "Managing Partner & CIO, @atreidesmgmt") and Ramp's lead economist Ara Kharazian ("Lead Economist @tryramp") clashed on X on Oct 9 (Beijing evening) over a Ramp chart Kharazian made: token volume at Ramp businesses keeps rising while spend sits below its peak. Baker (https://x.com/GavinSBaker/status/2108526656902091103) argued open-weight models squeezing model-layer margins are "net positive for AI infra demand"; Kharazian (https://x.com/arakharazian/status/2108560737215869226) replied that the decline reflects frontier price cuts from OpenAI–Anthropic competition, and thinner margins "will depress their ability to reinvest in infra."

【Verifiable】

Baker's post (462,155 views on X at writing): "An open-weight token consumes *roughly* the same amount of compute as a frontier token for a similar size model." Replying, he asked if labs cut prices because "they are losing share to open-weight models." Kharazian called the OpenAI–Anthropic rivalry "sufficient to explain the decline in AI prices"; cuts help only if met with more spend, "but we don't see that today." Ramp (Oct 6): the rate of firms switching model providers hit a record 8% (3-month average) in September, after OpenAI's June price cuts and launches. Ramp (Oct 7): cheaper models push AI spend down.

【Our read】

1. Different layers: Baker counts compute per token, Kharazian the dollars labs reinvest. The chart fits both: more tokens, fewer dollars.

2. Neither side's data separates the causes: spend over volume is an effective price blending list-price cuts, moves to cheaper models or tiers, and open-weight share. As we argued before, a matched list price is not a matched bill.

3. Test: if the open-weight story holds, Ramp's spend-by-maker data should show open-weight makers gaining on OpenAI and Anthropic; if their share stays flat as spend per token falls, Kharazian gains. Spend lagging rising volume would show a model-layer squeeze, not by itself weaker infra demand.