China faces an uphill battle to wean its technology firms off Nvidia chips, according to a Financial Times report that examines the gap between Beijing's semiconductor ambitions and the purchasing behaviour of the country's largest technology companies.

The FT's assessment frames the problem in blunt terms: Chinese technology firms continue to depend on Nvidia's processors, and the effort to replace them with domestic alternatives is proving far harder than policy announcements suggest.

Why does the dependency persist?

Nvidia's chips remain the workhorse hardware for artificial intelligence training and inference across China's cloud and internet sectors. The company's graphics processors have become the de facto standard for machine-learning workloads, and the FT reports that this installed base — plus the software ecosystems built around it — gives US silicon a grip that domestic suppliers have not yet broken.

For Chinese firms, switching away from Nvidia carries real technical risk. Models, toolchains and engineering workflows are calibrated to Nvidia's architecture. Domestic accelerators, while advancing, do not yet match the performance and reliability that large-scale AI development demands, the report indicates.

The dependency cuts both ways across the Pacific. Washington's export controls have restricted which Nvidia products Chinese customers may buy, and Beijing has pushed local firms toward homegrown chips. Yet the FT's reporting shows that Chinese technology companies still seek Nvidia hardware wherever the rules allow — a signal of how wide the performance gap remains.

What does this mean for Beijing's self-sufficiency drive?

China has made semiconductor self-sufficiency a stated national priority, with state funding and procurement pressure directed at local chip designers and manufacturers. The FT analysis concludes that this drive is running into a structural obstacle: the country's most advanced AI developers need the best available hardware, and that hardware is American.

The result is a tension between policy and practice. National directives point one way; procurement decisions by the firms doing China's most demanding AI work point the other.

For Nvidia, China remains a market it cannot fully serve and cannot fully abandon. For Chinese chipmakers, the Nvidia dependency defines the benchmark they must clear — not merely in raw specifications, but in the software maturity and deployment track record that determine whether a large AI laboratory will actually switch.

The FT report does not forecast a rapid substitution. Its framing — an "uphill battle" — signals sustained friction between geopolitical restriction and technical reality.

The takeaway

China's tech sector still runs on Nvidia silicon, and closing that gap is proving slower and harder than Beijing's self-sufficiency targets imply.