Frontier AI model prices are collapsing exactly as Anthropic and OpenAI approach their IPO windows. In the past eight days alone, Meta released Muse Spark 1.1 competitive with Anthropic's Opus 4.8 on key benchmarks, SpaceX released Grok 4.5 at less than half the price of Opus 4.8 with comparable coding performance, and Moonshot released Kimi K3, a 2.8-trillion-parameter model set to go open-weight. The premise that owning the frontier generates premium margins is cracking in real time.
Gavin Baker, managing partner at Atreides Management, put it plainly: a world with 2-3 dominant labs running 90% inference margins is bad for every other layer of the AI stack. That world may not arrive. The more likely outcome is six or more frontier labs, compressed margins, and a competitive product market that neither Anthropic nor OpenAI originally planned to fight on. Anthropic's Boris Cherny confirmed the shift in a May interview, noting that products now represent a substantially larger share of Anthropic's revenue mix than they did a year ago, when the business ran mostly on API.
The full piece is worth reading for the structural argument underneath the pricing news: the labs spent billions assuming model supremacy was the moat, and now they must win on products while still burning cash to maintain model leadership. The original covers how that dual burden lands at the exact moment investors are supposed to be rewarded, and what it means for the broader AI stack if commodity intelligence becomes the baseline.
[READ ORIGINAL →]