Chinese AI groups use cloud services to circumvent US chip export controls

Chinese artificial intelligence groups are bypassing export controls to access high-end US chips through intermediaries, exposing potential loopholes in Washington’s blockade of cutting-edge tech to the country.

AI surveillance groups hit by US sanctions have found ways to obtain restricted technology by using cloud providers and third-party lease agreements, and buying the chips through subsidiaries in China.

iFlytek, a state-backed speech recognition company blacklisted by Washington in 2019, has leased access to Nvidia’s A100 chips, which are in the race to develop groundbreaking AI applications and services, according to two employees with direct knowledge of the matter are of vital importance.

Facial recognition group SenseTime, which was sanctioned at the same time as iFlytek, has used intermediaries to buy banned components from the US, according to three senior officials familiar with the situation.

Privately controlled cloud computing companies also provide access to high-end US chips. AI-Galaxy, a Shanghai-based cloud computing company founded by former Nvidia and AliCloud employees, charges $10 for one-hour access to eight of its A100 Nvidia chips.

The ability of Chinese AI groups to continue accessing Nvidia’s key high-end chips and other cutting-edge technologies underscores the challenge the US faces in enforcing its trade restrictions on Chinese companies.

Last October, Washington imposed unilateral export controls that prevented US companies from selling advanced chipmaking equipment and high-end semiconductors, including the A100, to Chinese corporations, and extended its trade restrictions only to certain blacklisted companies.

“iFlytek can’t buy the Nvidia chips, but that’s not a problem because it can rent them and train our datasets on other companies’ computer clusters,” said an executive familiar with the AI ​​firm’s operations.

READ :  NVIDIA GTC Highlights Memory And Storage Importance For The Future

“It’s like a rental car system. You cannot take the chips out of the facility. It’s a huge building with a computer cluster, and you’re buying time on CPUs [central processing unit] or GPUs to train the models,” the person said.

While iFlytek can’t fully own the chips under US export controls, two employees said the rental system is a good, albeit more expensive, alternative. An engineer at iFlytek said the company “rents the chips and devices for the long term, which effectively amounts to owning them.”

iFlytek was banned from buying these semiconductors directly after Washington blacklisted it for its alleged role in providing technology for state surveillance of Uyghur Muslims in Xinjiang.

In some cases, SenseTime bought advanced chips directly through its own subsidiaries, which aren’t on Washington’s “company list,” according to three senior executives familiar with the situation.

SenseTime said it “strictly adheres to various domestic and foreign trade-related laws and regulations” and the group has developed a program to ensure it “meets trade compliance standards.”

iFlytek did not respond to a request for comment.

A spokesman for the US Department of Commerce said the Bureau of Industry and Security is “vigorously investigating potential export control violations but is not commenting on specific allegations.”

An export control expert in Washington said US export regulations would not cover cloud providers even if restricted chips were used. They added that the only violation is when the technology is used to build weapons of mass destruction.

Local governments in China rushed to help the industry stockpile high-end chips before Washington imposed tough export controls last October.

READ :  Mercedes-Benz partners Microsoft to amp up sustainability in car manufacturing

A person close to iFlytek said US export restrictions have sparked the proliferation of state-backed computer clusters that hoard Nvidia chips and lease access to the technology to blacklisted companies.

In addition, a government official in Zhejiang noted that “several places [local governments and companies] have or are building AI data centers that offer enterprise cloud rental services. This is part of the new infrastructure supported by national policy.”

“All AI companies use and store advanced chips like Nvidia A100, including us,” said an engineer at SenseTime, adding, “We do our compliance and legal work very thoroughly and diligently.”

“China’s AI industry faces huge risks due to US sanctions,” a person close to SenseTime said. “We expect the US to impose further sanctions on Chinese AI firms, their suppliers and customers. We need to make sure our supply chain and sales remain stable,” the person added.

Private cloud providers are seeing increasing demand for Nvidia’s A100 chips, which are also used in generative AI models like OpenAI’s ChatGPT.

An executive at a US tech giant in China said they continue to offer local companies access to cloud computing running on A100 chips, and the offer has attracted a large number of startups trying to clone ChatGPT. The company’s US legal team was initially wary of continuing to offer cloud services running on A100s chips, but eventually decided it didn’t violate export controls.

Even if US export controls do not explicitly prohibit cloud service providers from accessing the chips, some companies still take measures to disguise their identities.

READ :  Electromobility retraining of automotive engineers needed

An executive at a cloud services start-up in Shenzhen said they had seen a surge in demand for the A100 chips from “strange” companies that cloaked their true identities behind shell companies.

“Our colleagues told us that we should never seriously find out who was behind them,” said the executive.

Nvidia said, “While we cannot control every future use or downstream sale of our products, we require our distributors to comply with all US export regulations and only sell to eligible commercial, consumer and academic customers who support our products for charities.” use purposes.” .

Additional reporting from Ryan McMorrow and Sun Yu in Beijing and Tim Bradshaw in London