Anthropic Overtakes OpenAI in Revenue — Enterprise Bet Pays Off
In what may be the most significant power shift in the AI industry this year, Anthropic has surpassed OpenAI in annualized revenue, hitting a $47 billion run rate by May 2026 — nearly double OpenAI's reported $25–33 billion pace. The milestone validates Anthropic's enterprise-first approach: while OpenAI built mass-market appeal with ChatGPT, Anthropic quietly won procurement contracts, expanded its API footprint, and positioned Claude as the go-to model for business workflows.
The gap widened faster than analysts predicted. Epoch AI had forecast parity around mid-2026, but Anthropic crossed the $30 billion threshold in April alone. This week also saw Anthropic sign a $19 billion data center lease with TeraWulf, adding to over 12 US data center commitments totaling more than 1 gigawatt of capacity. The company also announced Claude Corps, a paid 12-month fellowship to train AI professionals within nonprofit organizations.
UN Holds First Multilateral AI Governance Summit
The Global Dialogue on AI Governance convened in Geneva on July 6–7, marking the first time all UN member states gathered specifically to discuss artificial intelligence regulation. The summit's opening message was stark: science currently cannot guarantee that increasingly capable AI systems will not cause catastrophic harm.
UN Secretary-General António Guterres struck a balanced tone, saying AI "could compress decades of development into years" while urging thorough safety testing and clear legal responsibility. Key outcomes included a call for nations to adopt an AI Child Safety Pledge, commitments to ensure developing countries have access to AI technologies, and a target for all AI data centers to run on renewable energy by 2030. A second session is scheduled for New York in May 2027.
Separately, the Five Eyes intelligence alliance issued a warning that frontier AI models will "fundamentally transform" offensive and defensive cyber capabilities within months, urging businesses to prepare identity checks, safety audits, and default watermarking on AI media.
Mistral Open-Sources Leanstral 1.5 for Formal Software Verification
Mistral AI released Leanstral 1.5 on July 2, an open-weights model under the Apache 2.0 license built for Lean 4, the proof assistant used in formal software verification. Unlike typical code generation models, Leanstral provides mathematical proof that software behaves as intended.
The 119B-parameter model (MoE architecture, 128 experts, 6.5B active parameters per token, 256k context window) delivers impressive benchmark results: it saturates miniF2F, solves 587 of 672 PutnamBench problems, and achieves state-of-the-art on FATE-H (87%) and FATE-X (34%). More importantly, it uncovered 5 previously unknown bugs across 57 repositories, including a sign-function overflow in the Rust library varinteger that caused crashes in debug mode and silent data corruption in release builds.
The model is free during beta and represents a significant step toward making formal verification accessible to everyday developers working on critical systems.
Meta Launches Muse Image — Its First In-House AI Image Generator
Meta unveiled Muse Image on July 7, the first image generation model from Meta Superintelligence Labs. The system is now available across the Meta AI app, Instagram Stories (US), and WhatsApp (select countries), with Facebook support coming soon.
What sets Muse Image apart is its agentic architecture: it invokes search and coding tools to improve accuracy, self-refines its own generations, and improves through scaling test-time compute. Since Meta AI maintains full conversation context, users can iteratively refine images without starting from scratch. Meta claims Muse Image generally surpasses Google's Nano Banana 2, trailing only ChatGPT's image generator in benchmarks.
The launch comes as Meta continues its AI-focused restructuring, having laid off approximately 8,000 employees while reassigning 7,000 to AI teams.
Google Launches Africa's First Applied AI Lab in Accra
Google announced the Africa Applied AI Lab, based at the Accra AI Community Centre in Ghana, to support African founders and researchers building AI solutions for uniquely African challenges. Applications opened July 1 and close August 31.
Selected startups will receive early access to Google DeepMind models (Gemini, Gemma, Veo), technical mentorship, and potential funding from Google AI Futures Fund and partner VCs including 4DX Ventures and Norrsken22. The initial cohort will run from mid-September to early December, culminating in a product demo day. The program focuses on five areas: the future of work, knowledge, software development, creativity, and entertainment.