High Capacity China's New AI Wave: Agents, Open Source, OpenClaw and More with Tom Wang at Hugging Face
25 snips
Mar 19, 2026 Tom Wang, Head of APAC Ecosystems at Hugging Face and open-source AI advocate. He surveys the new wave of Chinese models and how open source shaped their rise. He explains why autonomous AI agents matter, dives into OpenClaw’s token and safety risks, and touches on multimodal, robotics progress, hardware constraints, and how communities sustain innovation.
AI Snips
Chapters
Transcript
Episode notes
China Drives Architectural Diversity Beyond Transformers
- Chinese researchers are actively exploring alternative architectures like linear attention and hybrid designs beyond Transformers.
- Tom highlights experiments (RWA, Minimax M1, Kimi) showing architectural diversity emerges in open source China.
Constraints On Chips Fuel Efficiency Innovation
- Efficiency optimization is a global priority but more visible in China because labs must run models on weaker domestic chips, which breeds innovation.
- Tom argues constraints like cheaper chips push work on linear attention and RAM/cost reductions.
A Parallel Ecosystem Grows Around Domestic Chips
- Chinese labs are adapting models to domestic chips (Huawei Ascend, Cambrian) and building parallel software ecosystems to lower adoption friction.
- Tom says co-evolving software and chip teams matters for practical deployment and training new engineers.