NVIDIA AI Chip Prices Double in China Despite Export Controls
Export controls are working — but they haven't killed demand. Street prices for NVIDIA AI chips inside China have doubled, while China Telecom purchased 40,000 servers worth roughly $1.7B, mostly from the Huawei ecosystem. The market split is now unmistakable.
One Year In — the Market Has Split Between Switchers and Premium Buyers
More than a year has passed since the U.S. export controls on AI chips to China kicked into high gear. Early expectations that cutting off supply would also suppress demand have not played out. The Chinese market has split clearly: companies switching to domestic alternatives on one side, and companies willing to pay a steep premium for NVIDIA on the other.
Huawei's Ascend chip line is the primary recipient of the switchers. Its performance and ecosystem are maturing, and large telecoms and state-owned enterprises are adopting it at scale. Meanwhile, many companies working on frontier model development or research cannot afford to give up NVIDIA's performance edge — and continue to acquire chips through gray-market channels regardless of price.
China's AI Chip Market Under Controls — By the Numbers
Demand Won't Stop — Huawei's Self-Sufficient Ecosystem Is Getting Closer
The most important takeaway from this data is that controls are limiting supply, but not demand. The fact that companies will pay double for NVIDIA chips means that demand-side pull is outweighing the deterrent effect of controls in certain segments.
China Telecom's 40,000-server, $1.7B procurement signals that Huawei-based hardware is being embedded into national-level AI infrastructure at scale. As Huawei consistently absorbs server procurement demand, the completion of a self-sufficient AI ecosystem inches closer — raising the long-term risk of full technological decoupling.
For companies outside China, the implications cut two ways. Semiconductor equipment and materials suppliers face a signal of medium-term demand risk as the domestic Chinese ecosystem matures. AI service companies may soon face a choice about which ecosystem to commit to — and that choice may not be as reversible as it looks today.