Major open-weight model releases dominated AI discussions overnight. Moonshot AI unveiled Kimi K3, a 2.8-trillion-parameter multimodal model with 1M context window that is already topping coding and frontend benchmarks. The model is positioned as a high-performance, lower-cost alternative to closed systems, with open weights expected later this month.

Thinking Machines Lab followed with Inkling, a 975B-parameter MoE multimodal model supporting text, image and audio. The model targets enterprise autonomy use cases and also highlighted open-weight plans.

Infrastructure and Market Moves

TSMC reported record quarterly profit of approximately $22 billion, up 77% year-over-year, driven by AI chip demand. The company announced an additional $100 billion expansion in Arizona, bringing total planned U.S. investment to $265 billion.

NVIDIA is supplying 27,500 next-generation Rubin GPUs to a major Japanese “physical AI” robotics data center project backed by SoftBank, Sony and Honda. The initiative aims to build sovereign compute capacity for robotics, factories and autonomous systems.

Regulatory Developments

The European Commission is requiring Google, under the Digital Markets Act, to allow third-party AI assistants on Android and share certain search data by 2027. Separately, a Meta Oversight Board study found leading models tend to refuse or become more cautious when asked to criticize repressive governments, raising questions about opaque safety policies.

Outlook

Open-weight models continue to close the performance gap with closed frontier systems at a rapid pace. National-scale compute investments for robotics and physical AI, alongside regulatory pressure on platform control, remain the dominant themes heading into the weekend.

השורה התחתונה

The AI sector is seeing accelerated competition from open models and massive sovereign infrastructure bets, particularly in Asia.