Benedict's Newsletter: No. 663
AI Summary
Anthropic's 2025 financials show $4.6B revenue and an $8B operating loss, but annualized revenue hit $65B by July; OpenAI is nearing $70B annualized net revenue. Google released Gemini 4 to retake the lead, while OpenAI launched Dots agents and Meta expanded Muse. AMD acquired World Labs for $8.2B, and OpenAI disclosed security lapses affecting 100+ organizations.
Key Facts
Author Takes
Anthropic financials
Focusing on Anthropic's pre-agentic-coding numbers is deeply unserious; the real questions are inference profitability and future model costs.
AI competition
There's no sign of winner-takes-all effects in frontier labs, and pricing power is limited beyond compute shortages.
OpenAI Dots agents
Dots launch exposes the same blank-screen and jagged-model problems; agents may need to be horizontal/ambient or vertical/embedded to work.
Meta enterprise cloud
Building an enterprise infra business from zero is easier said than done, as Google Cloud's struggles show.
OpenAI security
OpenAI's unauthorized access issues are a failure of management and engineering, not alignment, and raise legal liability questions.
Related
Benedict's Newsletter: No. 664
This newsletter covers AI industry news including OpenAI's revenue misreporting, Meta's new agent protocol, Amazon blocking AI agents, open-source mod
Benedict's Newsletter: No. 662
Benedict Evans' newsletter covers the latest AI developments: Meta's Muse AI assistant gaining traction and being integrated into new hardware, OpenAI
Benedict's Newsletter: No. 660
This issue covers the latest AI safety debates, including Anthropic incidents and calls for slowdown, OpenAI's controversial AI-driven maths breakthro