Despite tightening US export restrictions, China's artificial intelligence infrastructure continues to be built around advanced Nvidia graphics processors. An analysis of more than 150 documents — from equipment procurement tenders to data center construction estimates and investment agreements — identified computing infrastructure projects in China with a combined value of over 23 billion yuan ($3.2 billion) that were initially designed to use chips from the American manufacturer.
The scale of dependence is impressive
This is not about individual purchases, but about the systemic architecture of entire computing clusters. The documentation covers a wide range of initiatives — from government programs to private data centers focused on training and inference of large language models. In all cases, the equipment configuration directly or indirectly points to Nvidia products as the preferred or only option for achieving target performance.
This picture exposes a fundamental contradiction: Beijing declares a course toward technological self-sufficiency and promotes its own accelerators, yet the actual engineering practice of leading AI projects still relies on the CUDA ecosystem and Nvidia architecture. Replacing these solutions requires not only hardware alternatives but also rewriting software stacks — a process measured in years, not months.
What this means for the market
The identified projects demonstrate that demand for advanced chips in China remains robust even under sanctions pressure. This creates fertile ground for gray imports, re-export schemes, and parallel supply channels — factors that are already having a noticeable impact on global semiconductor logistics. At the same time, the incentive for accelerated development of domestic analogues is growing, shaping long-term competition in the AI accelerator market.
Cryptalist analysis: The discovered dependence of Chinese AI projects on Nvidia confirms that the technological gap in AI hardware remains significant, and sanctions restrictions do not eliminate demand but merely transform the channels through which it is met. For investors, this is a signal of continued robust demand for computing power and of potential volatility in the semiconductor segment amid geopolitical risks.