Despite the U.S. government’s efforts to prevent advanced AI chips from falling into the hands of Chinese companies, some American corporations are finding ways to circumvent these restrictions. Oracle in particular has reportedly helped China’s TikTok by “renting” AI chips to the communist social media company.
Engadget reports that in 2022, the United States banned companies like Nvidia from selling their most advanced AI chips to China, citing concerns over potential military, surveillance, and economic implications. However, a recent report by the Information has revealed that U.S.-based cloud computing company Oracle is allowing TikTok owner ByteDance to “rent” Nvidia’s cutting-edge H100 chips to train AI models on US soil.
Shou Zi Chew, chief executive officer of TikTok Inc. Bryan van der Beek/Bloomberg via Getty Images
Larry Ellison, Chairman of Oracle ( Phillip Faraone /Getty)
ByteDance, a Chinese company that has direct ties to the Chinese government, is reportedly taking advantage of this loophole to access the coveted chips. While the practice runs against the spirit of the U.S. government’s chip regulations, it is technically allowed because Oracle is merely renting out the chips on American soil, not selling them directly to companies in China.
Former ByteDance employees have raised concerns about the company’s Project Texas initiative, which claims to separate TikTok’s U.S. operations from its Chinese leadership. They describe the project as “largely cosmetic,” alleging that ByteDance’s U.S. wing regularly works closely with its Beijing-based leadership.
ByteDance is not the only Chinese company seeking to exploit these loopholes. Alibaba and Tencent are reportedly discussing similar arrangements to gain access to the sought-after chips. These deals could be more difficult to prevent, as these companies have their own U.S.-based data centers and would not need to rent servers from American companies.
The U.S. Commerce Department, responsible for closing such loopholes, may already be aware of these practices. Earlier this year, the department proposed a rule requiring U.S. cloud providers to verify foreign customers’ identities and notify the U.S. if any of them were training AI models that “could be used in malicious cyber-enabled activity.” However, most cloud providers disapproved of the proposal, claiming that the additional requirements might outweigh the intended benefits, leaving the proposed rule in limbo.
Read more at Engadget here.
Lucas Nolan is a reporter for Breitbart News covering issues of free speech and online censorship.
