US, China eye AI model controls; DeepSeek releases inference stack

The geopolitical landscape of open-weight AI is intensifying with new regulatory actions from both the US and China. China's Ministry of Commerce proposed pre-approval for sharing frontier AI models internationally, though open-source technologies may receive automatic approvals. Concurrently, the US Department of Homeland Security has restricted DeepSeek as a vendor for federal civilian procurement, raising questions about its models' use in other AI systems. In response, DeepSeek has released its internal inference stack, DeepInfer, aiming for greater efficiency and potentially impacting model pricing across the ecosystem. AI

IMPACT New regulatory frameworks in the US and China could shape the development and distribution of open-weight AI models globally.

RANK_REASON The item discusses regulatory proposals and actions impacting AI models, rather than a direct release from a frontier lab.

Read on Interconnects (Nathan Lambert) →

• Commerce/BIS
• DeepSeek
• General Language Model
• MiMo V2.6 Pro RL
• Nemotron
• Nvidia
• People's Republic of China
• SGLang
• United States Department of Homeland Security
• vLLM
• XiaomiMiMo
• Zhipu AI

• policy
• model release
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →