OpenAI-Anthropic Split Fuels Parallel Scaling Races

The Atlantic account centers the personal feud, yet the structural driver is the 2021 reorganization that converted OpenAI’s capped-profit arm into an uncapped vehicle while Anthropic adopted a similar structure after Amazon’s $4 billion infusion. Both entities abandoned their original nonprofit charters once frontier-model training runs exceeded $100 million, shifting incentives from alignment research to revenue growth and talent retention.

Public statements on safety function as market positioning rather than technical divergence. Anthropic’s constitutional AI papers and OpenAI’s Superalignment team dissolution both occurred within six months of each other; neither firm has published head-to-head red-teaming results on the models they actually ship. The pattern matches earlier platform rivalries where stated values track capital requirements more closely than internal research priorities.

The next phase hinges on sovereign compute access. Whoever secures the larger tranche of subsidized chips from the US, UAE, or Saudi funds will set de-facto safety standards through usage volume, not through published benchmarks. Regulatory proposals currently circulating in Brussels and Washington track the lobbying filings of the two firms rather than independent safety data.