AI Trends

Nvidia Tightens Customer Access, Restricts Chip Sales to Customers Renting More Than Half of Their GPUs to China

cnBetaUpdated Nov 17 20:57

AI summary

In communications with customers, Nvidia mentioned the Remote Access Security Act (RASA) being advanced by the U.S. Congress and raised its own sales threshold, stating that it will not sell chips to customers that rent more than half of their GPU resources to Chinese enterprises. It was previously reported that Nvidia had verbally informed some data center customers that future cooperation agreements should ensure at least 51% of service capacity is directed at non-Chinese customers. The U.S. Bureau of Industry and Security recently also published new interim rules aimed at restricting Chinese enterprises' remote access to advanced U.S. AI chips and supercomputing resources.

Why read this

AI score 66

Nvidia has incorporated the proportion of customers' GPU rentals to China into its sales threshold, revealing a path by which export controls extend from chips to cloud computing services.

Topics

Tags

Body · AI translation

Currently, U.S. export control policy already restricts the direct export of high-performance AI chips to China. However, the U.S. government is trying to further close regulatory loopholes by expanding oversight from physical chips to overseas cloud computing services and remote computing power access channels. Regulators worry that even if Chinese companies cannot directly purchase advanced chips, they may still be able to rent computing resources through overseas data centers and thus continue to use the most advanced U.S. AI hardware.

It is understood that in communications with customers, Nvidia specifically mentioned the Remote Access Security Act (RASA) being advanced in the U.S. Congress. The bill aims to give the U.S. government greater authority not only to restrict exports of advanced chips but also to regulate remote access to these chips through cloud services. If the bill ultimately takes effect, relevant companies will face stricter compliance requirements when providing AI computing power services to Chinese customers.

At the same time, Nvidia has also raised its own sales threshold. Sources said the company has stated that it will not sell chips to customers that rent more than half of their GPU resources to Chinese companies. This means that data centers and cloud service providers will face more restrictions in structuring their customer base in the future.

In fact, similar signs had already appeared earlier. Reports said Nvidia had verbally told some data center customers that in future cooperation agreements they should ensure that at least 51% of service capacity is directed to non-Chinese customers, in order to reduce potential regulatory risk and meet policy requirements that may be introduced in the future.

U.S. regulators have in recent years been concerned that renting computing power through overseas cloud platforms has become an important way for some companies to bypass physical chip export restrictions. Because the relevant services take place in overseas data centers, the United States previously lacked sufficient regulatory means for this remote access model, so promoting a new regulatory framework has become a policy priority.

Earlier this year, the Remote Access Security Act passed the U.S. House of Representatives but has not yet completed Senate review. At the same time, the Trump administration continues to push for stronger export controls in the AI field, hoping to maintain the United States' leading position in global AI competition by restricting the flow of advanced computing power to China.

This month, U.S. media reported that Chinese AI company Moonshot AI had attempted to obtain Nvidia's advanced chip resources through European data centers. Previously, the U.S. government had also accused the company of obtaining relevant computing power support through data centers in Asia. These cases have further intensified U.S. regulators' attention to the remote access model.

Recently, the U.S. Department of Commerce's Bureau of Industry and Security also published a new interim rule aimed precisely at restricting Chinese companies' remote access to advanced U.S. AI chips and supercomputing resources. Analysts believe this shows that the focus of U.S. regulation is gradually shifting from traditional hardware export controls to regulation of computing power services.

Regarding issues of outside concern, Nvidia and the Bureau of Industry and Security have so far made no public comment.

In addition to regulatory pressure, Nvidia has recently also faced additional scrutiny from U.S. political circles. U.S. Treasury Secretary Scott Bessent recently publicly called for "major reform" at Nvidia and questioned the company's internal culture regarding export control compliance. He said that relevant U.S. authorities need to further review the responsibility of Nvidia's management on related issues.

Facing sustained pressure, Nvidia previously said in filings submitted to regulators that the company has always strictly complied with all export control regulations, but after products are sold, companies cannot exercise actual control over the hardware, so they need to rely on the compliance management systems of customers and partners themselves to ensure that relevant regulations are implemented.

Industry generally believes that the United States is building a full-chain regulatory system covering chip production, hardware sales, cloud services and remote computing power access. In the future, global data center operators, cloud computing companies and AI service providers may all face stricter review requirements, and international competition over control of advanced AI computing power may further escalate.

Source

cnBeta