A Reuters investigation has found that computing projects worth more than US$3 billion (S$3.86 billion) in China use Nvidia processors subject to US export controls.
The report said many projects in China still use Nvidia technology, despite the company’s earlier assertion that US export controls had effectively shut it out of the Chinese market.
The investigation found that the projects stretch from technology hubs along China’s economically developed coast to remote western provinces, underscoring the continued importance of Nvidia technology to the development of China’s artificial intelligence (AI) industry.
In Xianyou county in the south-eastern coastal province of Fujian, a planned AI computing centre with an investment of 6.2 billion yuan (about S$1.2 billion) is seeking commercial partners. Project plans show that the centre will use Nvidia H100 chips to support large-scale AI development.
In the north-western province of Shaanxi, state-owned China National Machinery Industry Corporation said one of its subsidiaries had won a contract to build a computing centre. The project, with a total investment of 3 billion yuan, plans to use Nvidia H100 and H200 processors.
A government investment prospectus released in April in the north-eastern province of Jilin disclosed that another computing project, worth 2.26 billion yuan, plans to use both Nvidia H100 processors and Chinese-developed chips. The developer expects to raise about 80% of the investment through bank loans.
A computing project in Inner Mongolia, with an investment of 1.5 billion yuan, plans to install 625 servers equipped with Nvidia H200 chips. Chinese media reported in June that construction had begun on two buildings in the project’s first phase.
The investigation concluded that these projects reflect China’s continued reliance on US processors to build its computing infrastructure. Further tightening of US export controls could delay construction, raise costs or even force developers to switch to domestically produced chips and redesign the facilities.
Some computing facilities already operating in China also use Nvidia hardware subject to US export controls.
The government of Wuxi in Jiangsu province said in March last year that a local computing cluster using Nvidia H800 processors had begun operating.
In Fujian, Shanghai-listed Greattown Holdings disclosed in June that a computing business in which it was involved had built a cluster using H800 processors. The company had previously signed a computing services agreement originally valued at 759 million yuan, but the contract was later terminated early.
Some smaller purchases show that Chinese research institutions also have demand for the chips. Reuters previously reported that Shenzhen University bought three H100 processors in September, while Northwestern Polytechnical University, which is on a US blacklist, purchased two H200 chips the following month.
