State of AI Report 2026 is out: OpenAI and Anthropic near a $105B combined run rate

Air Street Capital's ninth State of AI Report (Nathan Benaich, 244 pages) landed yesterday and it is the research find of the week. Read the whole thing on Lounge: lounge.ai/research/state-of-ai-report-2026-air-street-capital-october-2026-openai-and-anthropic-nea

The headlines:

• A three-lab frontier. Anthropic leads Artificial Analysis's index, OpenAI and Google tie just behind.
• $105B. The combined annualized run rate across OpenAI and Anthropic, with inference revenue growing fastest.
• China leads on citations. Chinese open-weight model families now draw more research citations than American ones.
• $138B in sovereign AI pledges, while data-center NIMBYism is starting to slow buildouts.

Two things for founders in particular. First, the harness now matters as much as the model: swapping the scaffolding around a fixed model produced gains of up to 6x on agent benchmarks, so your product team owns more of the performance curve than you think. Second, safety incidents are getting concrete (agents attacking a public code host during reduced-safeguard tests), which will shape the questions enterprise buyers ask you over the next year.

Which chart surprised you most? Bring it to Thursday's mastermind.