Chinese businesses access US AI technology

through cloud loopholes

As per a Reuters investigation based on public procurement documents, Chinese organizations are using cloud services from Amazon and its competitors to acquire powerful AI chips and capabilities from the United States that they cannot obtain otherwise.

In a comprehensive investigation, Reuters revealed how Chinese cloud access to US AI chips is facilitated through intermediaries. Over 50 tender documents posted in the past year revealed that at least 11 Chinese entities have sought access to restricted US technologies or cloud services. Four of these explicitly named Amazon Web Services (AWS) as a cloud service provider, though accessed through Chinese intermediaries rather than directly from AWS.

“AWS complies with all applicable US laws, including trade laws, regarding the provision of AWS services inside and outside of China,” an AWS spokesperson told Reuters.

The report highlights that while the US government has restricted the export of high-end AI chips to China, providing access to such chips or advanced AI models through the cloud is not a violation of US regulations. This loophole has raised concerns among US officials and lawmakers.
One example cited in the report involves Shenzhen University, which spent 200,000 yuan (£21,925) on an AWS account to access cloud servers powered by Nvidia A100 and H100 chips for an unspecified project. The university obtained this service via an intermediary, Yunda Technology Ltd Co. Neither Shenzhen University nor Yunda Technology responded to Reuters’ requests for comment.

The investigation also revealed that Zhejiang Lab, a research institute developing its own large language model called GeoGPT, stated in a tender document that it intended to spend 184,000 yuan to purchase AWS cloud computing services. The institute claimed that its AI model could not get enough computing power from homegrown Alibaba cloud services.
Michael McCaul, chair of the US House of Representatives Foreign Affairs Committee, told Reuters: “This loophole has been a concern of mine for years, and we are long overdue to address it.”

The US Commerce Department is enforcing stricter regulations in reaction to these worries. “Seeking additional resources to strengthen our existing controls that restrict PRC companies from accessing advanced AI chips through remote access to cloud computing capability,” a government spokesperson told Reuters.
A rule that would force US cloud computing companies to authenticate big AI model users and alert authorities when they use US cloud computing services to train big AI models capable of “malicious cyber-enabled activity” has also been suggested by the Commerce Department.

The report also discovered that Chinese businesses are attempting to use Microsoft’s cloud services. Sichuan University, for instance, disclosed in a contract submission that it was creating a generative artificial intelligence platform and that, in order to facilitate project delivery, it would buy 40 million Microsoft Azure OpenAI tokens.
According to a Reuters article, Amazon has given Chinese companies access to cutting-edge AI models like Anthropic’s Claude and contemporary AI chips that they would not have otherwise had. Public postings, tenders, and marketing materials that the news organization examined served as examples of this.

At a generative AI-themed conference in Shanghai in May, Chu Ruisong, President of AWS Greater China, said, “Bedrock provides a selection of leading LLMs, including prominent closed-source models such as Anthropic’s Claude 3.”
The overall focus of the paper is on how challenging it is to control who gets access to powerful computing resources in a global technology ecosystem that is becoming more interconnected. It focuses on the complex interactions that exist between cloud service providers, Chinese businesses seeking to advance their AI skills, and US export rules.

The scenario raises questions about the effectiveness of current export controls and the possible need for more comprehensive rules that cover cloud-based access to prohibited technologies as the US government attempts to close this gap.
The results of this study will probably contribute to current debates over national security, technological transfer, and the battle against artificial intelligence. These results may lead to new conversations about how to strike a balance between national security concerns and technical cooperation in an era of rapidly expanding AI as legislators and business executives examine them.