Open-weight AI's 29% token share pressures OpenAI and Anthropic valuations

Open-weight AI is quietly capturing production volume without needing to beat frontier models. Vercel's July AI Gateway Production Index found open-weight models handled 29% of gateway tokens in June, up from 11% in April, but accounted for under 4% of spend. Anthropic still captured 61% of spending on 32% of tokens, and at least 72% in high-stakes use cases like coding and back-office agents. Meanwhile, OpenAI has already cut prices—slashing GPT-5.6 Luna prices by 80% and Terra by 20% on July 30, framing them as efficiency gains passed to customers. The risk to frontier labs is not being out-performed, but being under-priced: if cheap models absorb routine workloads or force broad price cuts, revenue per task and gross margins face pressure.

That pressure hits trillion-dollar expectations head-on. Anthropic’s latest private valuation is $965 billion with annualized revenue above $47 billion, and investors are modeling an October IPO at $2 trillion or more. OpenAI, valued at $852 billion in an employee tender, could seek up to $1 trillion when it lists next year, with annualized revenue around $40 billion. Both filed confidential IPO paperwork in June. Defensively, Anthropic is reportedly pursuing a $6 billion acquisition of Decart to make AI infrastructure more efficient, while products like Claude Code tie spending to higher-value workflows. For enterprises and investors, the key signals are gross margin after inference costs, revenue per completed task, retention, and the share of revenue from agentic products. If those hold while open-weight volume rises, frontier labs can absorb the competition; if premium workloads move down-market, valuation assumptions become far harder to defend.