A recent Reuters investigation has uncovered that Chinese organizations are accessing advanced U.S. AI technology by using cloud services from companies like Amazon, despite export restrictions on these technologies. 

How Do Chinese Firms Access U.S. AI?

---

• Chinese firms use intermediaries to access U.S. AI technologies through cloud services like AWS, despite export restrictions.
• U.S. export controls are bypassed through cloud services, raising concerns among officials.
• The U.S. Commerce Department is tightening regulations to better control access to advanced AI technologies.

---

This access is facilitated through intermediaries, as revealed by an analysis of public tender documents.

How Cloud Services Circumvent Export Restrictions

While the U.S. government restricts the direct export of high-end AI chips to China, using cloud services to access these chips does not violate U.S. laws. This loophole has prompted concerns among U.S. officials about the effectiveness of current trade restrictions.

According to the Reuters investigation, at least 11 Chinese entities have sought access to restricted U.S. technologies or cloud services in the past year. 

Four of these specifically named Amazon Web Services (AWS) as a provider, though they accessed these services through intermediaries rather than directly from AWS. An AWS spokesperson confirmed that the company adheres to all relevant U.S. trade laws regarding its services in China.

Instances of Cloud Access Revealed

The report provides several examples of how Chinese institutions leverage intermediaries to obtain U.S. AI technology. Shenzhen University, for instance, spent approximately 200,000 yuan (£21,925) on an AWS account to use cloud servers with Nvidia A100 and H100 chips for an unspecified project. 

This access was arranged through Yunda Technology Ltd Co., a Chinese intermediary. Neither Shenzhen University nor Yunda Technology responded to Reuters’ inquiries.

Another case involves Zhejiang Lab, a research institute developing a large language model called GeoGPT. The institute indicated in a tender document that it planned to spend 184,000 yuan on AWS cloud services, citing insufficient computing power from domestic alternatives like Alibaba’s cloud services.

U.S. Government’s Response and New Regulations

Michael McCaul, chair of the U.S. House of Representatives Foreign Affairs Committee, expressed his concerns about this loophole, stating, “This loophole has been a concern of mine for years, and we are long overdue to address it.”

In response, the U.S. Commerce Department is tightening regulations. A department spokeswoman revealed that efforts are underway to enhance controls that prevent Chinese companies from accessing advanced AI chips through cloud computing. 

Additionally, a proposed rule would require U.S. cloud firms to verify users of large AI models and report any potential misuse of these models for malicious activities.

Microsoft and Other Cloud Providers Involved

The investigation also found that Chinese companies are seeking access to Microsoft’s cloud services. Sichuan University, for example, has announced plans to purchase 40 million Microsoft Azure OpenAI tokens to support its generative AI projects.

The report further revealed that Amazon has provided Chinese businesses with access to advanced AI models, such as Anthropic’s Claude, which they would not have otherwise obtained. 

Chu Ruisong, President of AWS Greater China, highlighted at a recent conference that AWS’s Bedrock platform offers access to leading large language models, including Claude 3.

Regulatory Challenges and Future Implications

The findings underscore the challenges of regulating advanced technology access in a globalized tech landscape. They reveal the complex interplay between U.S. export laws, cloud service providers, and Chinese firms seeking to enhance their AI capabilities. 

As U.S. authorities work to address these issues, questions about the adequacy of current export controls and the need for comprehensive regulations covering cloud-based technology access are emerging.

This report will drive further discussions about technology transfer, national security, and the global AI competition. As policymakers and industry leaders assess these findings, new debates will likely arise about balancing technological advancement with national security in an era of rapid AI development.