China learns to build without Nvidia

Hi, this is Evelyn, writing to you from Beijing. Welcome to the latest edition of The China Connection — a succinct snapshot of what I’m seeing and hearing from local businesses.

China’s tech self-sufficiency push is rapidly becoming a reality as companies focus on business questions that run deeper than geopolitics. What does that mean for Nvidia?

Robovan startup Zelostech plans to use multiple chip suppliers from China and elsewhere, over the next year or two, instead of relying only on Nvidia for its self-driving systems, the company told CNBC.

A major factor is cost, said Shi Yunjian, director of finance and investment. Using China-made chips, for example, would cost far less than the two Nvidia Orin chipsets currently used in each vehicle, he said.

That’s a big deal because scale is becoming a competitive advantage. The more autonomous vehicles can deploy, the more operating data they can collect and the easier it becomes to convince regulators that the technology is ready for wider use.

Zelostech claims it already has more than 25,000 vehicles operating in over 20 countries, with plans to expand rapidly. These don’t carry people, and many are smaller than a mail truck. Most operate in mainland China, mostly for logistics companies delivering packages.

By comparison, Alphabet-backed Waymo has just under 4,000 vehicles on the road, while Chinese rivals Baidu, WeRide and Pony.ai have yet to deploy fleets at a similar scale.

Waymo uses custom chips, while Chinese electric car giant BYD last week joined Nio and Xpeng in revealing their own semiconductors for driver-assist systems.

This year, Nio said it’s planning a fivefold increase in spending on computing power. When I asked whether that included Nvidia, CEO William Li said the company was no longer buying chips but renting compute power powered by a variety of processors.

A vehicle Xpeng co-developed with Volkswagen is also using the Chinese company’s “Turing chip,” while the German automaker has partnered with China’s Horizon Robotics to develop driver-assist systems in China — without Nvidia.

Nvidia’s driver-assist chips are not subject to the same U.S. export restrictions that apply to the more advanced semiconductors used to train and run AI models.

Yet even after Nvidia CEO Jensen Huang joined U.S. President Donald Trump on his trip to Beijing in May, it’s clear China is not eager to let more Nvidia chips in.

The shift extends beyond vehicles. Chinese AI developers have increasingly optimized their models to run on homegrown hardware, rather than Nvidia’s widely used CUDA ecosystem.

The latest MiniMax and Kimi models, along with DeepSeek’s V4, are compatible with local Chinese…