Chinese AI Labs Dominate Open-Weight Models; Kimi K3's 2.8T Parameters Trigger July 2026 Policy Debate

Chinese AI laboratories have emerged as leaders in open-weight large language models, with Alibaba's Qwen 3.6 exceeding one billion cumulative downloads and spawning over 180,000 derivative models, according to industry analysis published in July 2026. Moonshot AI's Kimi K3, unveiled mid-July with 2.8 trillion parameters and slated for full open release, prompted Treasury Secretary Scott Bessent to float potential sanctions against overseas AI models on intellectual-property grounds.

On July 24, Nvidia, Microsoft, and Meta led roughly two dozen firms including AMD, Cisco, and Hugging Face in publishing the "Open Weights and American AI Leadership" letter, urging policymakers to expand compute access and avoid restrictions on open models. Notably absent were Alphabet, Anthropic, and OpenAI. The coalition argued that concentrating advanced capabilities behind closed systems created systemic risk, while open weights enabled external scrutiny proprietary models denied.