Nvidia just confirmed it's taking orders from China again, and Jensen Huang betting on OpenClaw tells you everything about where the AI arms race is actually happening.

The Summary

  • Nvidia is now receiving orders from China and ramping production of H200 chips, signaling a thaw in export restrictions or clever workarounds
  • Jensen Huang called OpenClaw "the next ChatGPT," immediately pumping Chinese AI stocks and validating China's homegrown AI ecosystem
  • The timing matters: this isn't just about chips, it's about who controls the infrastructure layer for autonomous agents

The Signal

For two years, the U.S. tried to kneecap China's AI ambitions by blocking advanced chip exports. Nvidia's confirmation that Chinese orders are flowing again means one of three things happened: Washington blinked, Nvidia found compliant chip designs that thread the regulatory needle, or Chinese buyers are routing orders through third countries. None of these options are comforting if you believed export controls would create a durable moat.

The H200 ramp is the technical story, but Huang's OpenClaw endorsement is the strategic one. OpenClaw is China's answer to ChatGPT, built on Chinese infrastructure, trained on Chinese data, optimized for Chinese language models. When the CEO of the world's most valuable AI company calls it "the next ChatGPT," he's not making a product review. He's reading the geopolitical map and deciding Nvidia can't afford to sit out the Chinese market while homegrown competitors like Huawei's Ascend chips gain traction.

This is the agent economy splitting into parallel stacks. Western enterprises will run agentic workflows on OpenAI, Anthropic, and Google models. Chinese enterprises will run them on OpenClaw, Baidu, and Alibaba models. Both need chips. Both need inference at scale. Nvidia just decided it's playing both sides, and the stock market reaction to Chinese AI companies suggests investors think that's the right call.

The real question isn't whether Nvidia can sell to China. It's whether the agent economy fragments into incompatible regional blocs, or whether the economics of training and inference force a common substrate. Right now, bet on fragmentation.

The Implication

If you're building AI products, assume your Chinese competitors just got a serious compute boost. If you're investing in AI infrastructure, watch how quickly Chinese labs close the capability gap now that H200s are in play. And if you're in policy circles still pretending export controls are working, this is your wake-up call. The global agent economy isn't waiting for permission. It's routing around restrictions and building parallel realities.

Source: Bloomberg Tech