China is reportedly accessing banned US-made AI chips via cloud services from Amazon and other providers, despite US efforts to restrict technology.
The United States is ensuring that China is unable to catch up by utilizing US-made technology, as the race to develop and deploy the most potent artificial intelligence models has rapidly escalated over the past couple of years.
Nevertheless, a recent report by Reuters disclosed that state-affiliated Chinese entities have been utilizing cloud services from Amazon or its competitors to obtain access to these prohibited advanced semiconductor processors and AI capabilities.
Bypassing Constraints
Reuters reported that an examination of more than 50 public tender documents revealed that at least 11 Chinese entities have attempted to gain access to restricted US technologies or cloud services.
Four of the entities that were examined explicitly referenced Amazon Web Services (AWS) as a cloud service provider. Nevertheless, the entities obtained the services through intermediaries, specifically Chinese companies, rather than directly from AWS.
The investigation discovered in a tender document from April 2024 that Zhejiang Lab, a Chinese research institute that is currently working on its own LLM, GeoGPT, intends to allocate 184,000 yuan ($25,783 USD) for the acquisition of AWS cloud computing services. It stated that its AI model is presently incapable of acquiring an adequate amount of computing power by relying solely on Alibaba, which is a Chinese company.
An additional document disclosed that Shenzhen University expended 200,000 yuan ($27,996 USD) on an AWS account to access cloud servers that were fueled by Nvidia A100 and H100 chips for an unspecified task.
The Chinese semiconductor market has been actively pursuing methods to circumvent its restrictions over the past year, despite the imposition of US embargoes. This entails utilizing new products from the world’s foremost chip manufacturer, Nvidia, that were capable of being shipped to China, as well as collaborating with local chip manufacturers.
In September 2022, the United States prohibited the use of Nvidia’s A100 and H100 chips, which are among the most potent and restricted AI products.
Initially, the A800 and H800 processors were designed for the Chinese market, but they were also prohibited as of October 2023. The United States emphasized the necessity of restricting the AI capabilities of the Chinese military.
However, in April 2024, tender documents were discovered by Reuters, which indicated that China had obtained access to the processors through Supermicro Dell servers.
Tightening of Restrictions
Nevertheless, the most recent discoveries regarding China’s access to the United States’ sophisticated processors and AI models do not constitute a violation of US regulations. The regulations are specifically designed to address the export or transmission of commodities, software, or the technology itself.
Reuters was informed by an AWS spokesperson that:
“AWS complies with all applicable U.S. laws, including trade laws, regarding the provision of AWS services inside and outside of China.”
In an effort to counteract China’s efforts to circumvent its regulations, the United States is attempting to further tighten them to encompass cloud-based access.
The loophole has been a source of concern for Michael McCaul, the chair of the US House of Representatives Foreign Affairs Committee, for years, and he believes that it is now time to resolve it.
Many Chinese entities, such as Sichuan University, have also targeted Microsoft cloud servers. In another tender document, the university announced that it is constructing a generative AI platform and purchasing 40 million Microsoft Azure OpenAI tokens.
Historically, China has implemented export controls on metals that are primarily employed in the production of semiconductors in response to the United States’ restrictions.
