Kimi K3 Fully Released, Nvidia and Microsoft Form a Security Alliance Without OpenAI/Google/Anthropic, and Satya Nadella Warns: Companies That Trust a Single AI May Not Survive

🔴 Kimi K3 fully released (July 27) — the biggest open-weight model launched this year, if not ever — Chinese startup Moonshot AI released the complete Kimi K3 weights, positioned directly at the U.S. market. The company claims the model rivals OpenAI and Anthropic's flagship models at a much lower price. There's a catch: the license becomes more restrictive once the model leaves internal organizational use — a critical point for any business considering it. Meanwhile, the U.S. Department of Commerce (OSTP, represented by Michael Kratsios) is examining whether Moonshot improperly used Anthropic's Fable model and restricted Nvidia GB300 servers routed through Thailand. The U.S. Treasury is considering sanctions.

🔵 Nvidia and Microsoft form a new AI security alliance — without OpenAI, Google or Anthropic — The alliance comes shortly after a disclosed security incident involving a rogue AI agent, aiming to coordinate defenses against autonomous AI agent risks. The unusual absence of three of the biggest AI companies from the alliance raises real questions about how unified the industry actually is on security.

🟢 Microsoft launches Project Perception — an AI system for cyber-defense — Powered by the new MAI-Cyber-1-Flash model, the system scans an organization's data, files and workflows to identify and patch security gaps before they're exploited — at roughly half the cost of comparable models. Particularly relevant for small and medium businesses looking for affordable cybersecurity protection.

⚪ Meanwhile: Claude Opus 5 is already breaking subscription records on Claude Max — The model, which launched July 24 (priced the same as Opus 4.8, at $5/$25 per million tokens), continues to strengthen Anthropic's app appeal for consumers and businesses.

⚪ Gemini — a Google milestone in new users — Google reported 950 million monthly active users on Gemini (up from 750 million the previous quarter), alongside a 24% jump in Alphabet's stock over the past two quarters.

Why this matters: This was one of the busiest days yet for major announcements — Kimi K3's release, the security-alliance split, and Nadella's warning all point to the same conclusion: betting on a single AI model is a real business risk, not just a technical footnote.