Frontier AI labs may stop selling their best models. The core argument: when five, six, or seven labs reach roughly the same capability frontier, selling model access by the token becomes a race to the bottom. OpenAI and Anthropic no longer hold a commanding lead. Open-weights competition from China, compute-rich pressure from Google, Meta, and SpaceX have crowded the top tier. A narrow lead over competitors changes the optimal strategy entirely.
The math is direct. One or two labs with a long lead: sell the model at markup, bank margins. Several labs essentially tied: sell the product built on top of the model, not the model itself. Anthropic has already moved in this direction with Claude Code and Claude Design. OpenAI is pushing toward a ChatGPT superapp. The threat the software industry feared, AI labs competing directly with their own API customers, is now a business necessity, not just a risk.
The original piece is worth reading for its framing of the 'pull up the ladder' scenario: a lab with a narrow but real lead using its frontier model to build products competitors cannot replicate with last-generation tech, before that window closes. That strategic moment is happening now. The question of who owns the compute beneath all of it is where the argument gets coldest and most consequential.
[READ ORIGINAL →]