There are growing allegations that Chinese developers are using U.S.-based cloud services to circumvent measures that block access to advanced chips and other technologies to speed up AI development.
The White House began imposing export restrictions on advanced chips to China nearly two years ago and extended them last year, effectively cutting off high-performance GPUs and accelerators from the Chinese market.
The reason for these sanctions, at least according to the US government, is to prevent Beijing's military from accessing high-performance computing capacity that could be used to develop and weaponize AI.
As The Register reported earlier this year, some Chinese entities were found to have purchased or attempted to purchase export-controlled chips, such as high-performance GPUs from Nvidia, that were running in U.S. data centers, cleverly skirting the restrictions without explicitly violating export sanctions.
Reuters reported that public bidding documents posted in a Chinese database over the past year showed several China-based organisations seeking access to restricted US technology.
The company claims that some of the documents clearly identify Amazon Web Services (AWS) as the cloud service provider from which it is purchasing services, but it appears the access may have been through a Chinese intermediary rather than directly from AWS.
Amazon has denied any wrongdoing. “AWS complies with applicable US laws, including trade laws, regarding the provision of AWS services inside and outside of China,” a spokesperson told The Register.
According to the documents, organizations attempting to access resources for AI processing include the state-run Zhejiang Research Institute, government-funded research universities Shenzhen University, Sichuan University, and the Suzhou Advanced Research Institute at the University of Science and Technology of China (USTC).
Some of these organizations are said to be interested in accessing OpenAI tools via Microsoft's Azure cloud platform. We have reached out to the company for comment.
Shenzhen University spent 200,000 yuan ($27,996) on an AWS account to access cloud servers equipped with Nvidia's A100 and H100 accelerator chips, without disclosing details about the project, according to Reuters. Reuters said the findings show the lengths some in China will go to to access the cutting-edge resources needed for generative AI models.
The U.S. government has been steadily tightening regulations to close loopholes, but The Register reported in March that some Chinese companies that the U.S. has blacklisted from acquiring certain technologies are still able to rent them, such as Nvidia's A100 GPUs.
We have asked the U.S. Department of Commerce whether it is aware of these latest activities and whether it is considering additional measures. We will update if we receive a response.
But some in the semiconductor industry have criticized the White House's approach to China on the tech front. In July, the former president of chip-making equipment giant ASML said the U.S.-China “semiconductor war” was primarily ideological in nature. His former employer is caught in the crossfire due to U.S. restrictions on exports of ASML's products to China, one of its largest markets.®
