[CHINA PULSE] Weekly

Both labs set their flagship previews below cost, using price to compete where benchmarks alone can't close the gap.

The world's first profitable humanoid-at-scale is now in the final administrative steps before becoming a public company — setting a price benchmark for the entire embodied-AI sector.

Two fresh Wednesday statements from the Treasury and USTR desks harden Washington's rhetorical posture — without yet naming a specific model or company.

A Foreign Policy analysis published Wednesday reads Xi's conference address as a pointed contrast to US closed-model strategy — with five Southeast Asian states now institutionally aligned.

A new fundraising target — up two-thirds from June — arrives the same day Moonshot launches its desktop agentic product.

MOFCOM's consultation now explicitly targets offshore fabrication of Chinese chip architectures — a new front beyond model weights.

The Beijing GPU maker's densest system yet arrives as domestic cloud operators face pressure to swap foreign silicon.

The target's AI-driven live-ops engine — not its game IP — is what Tencent wants to scale globally.

The facility runs on at least 10,000 Chinese-made chips; XCore Sigma adds the compiler stack to run them efficiently.

Built on a fully domestic supply chain, the M100 focuses on cost-effective large-model inference — Baidu's first public showing of the chip.

With DeepSeek V4 tuned for Huawei hardware and ByteDance committing $5.6B to Ascend this year, China's inference stack is pivoting away from Nvidia faster than expected.

The CAC's companion-AI measures are now enforceable law — and the first compliance wave has already swept through China's two largest consumer AI platforms.