Frontier AI is becoming a commodity, and the timing is brutal. In the past eight days, Meta, SpaceX, and China's Moonshot have each released models that compete with Anthropic's Opus 4.8 at a fraction of the cost. SpaceX's Grok 4.5 is less than half the price of Opus 4.8 and competitive on coding benchmarks. Moonshot's Kimi K3 is a 2.8-trillion-parameter model preparing to go open-weight. This is happening exactly as Anthropic and OpenAI are eyeing IPO windows built on the premise that owning the frontier generates premium margins.
Atreides Management's Gavin Baker put the structural threat plainly: a world with only 2 to 3 dominant frontier labs running 90% inference margins is net negative for every other layer of the stack. The original bet was exclusivity. The emerging reality is six or more competitive frontier developers, which compresses margins across the board and shifts value creation away from the model layer entirely.
The original piece is worth reading for what it reveals about the internal pivot already underway. Anthropic's Boris Cherny told the author in May that products now play a 'much bigger role' than they did a year ago, with Claude Code leading a shift away from pure API revenue. The irony the article traces is sharp: Anthropic and OpenAI spent billions to win the model race, and winning may force them to compete on product against everyone else, while losing money on the models that got them there.
[READ ORIGINAL →]