Nvidia cuts over half of its Asian AI chip customers in China export crackdown
Nvidia has cut its approved distributor list in Asia in half and is implementing strict new vetting requirements to keep its AI chips out of China.
In Japan, Singapore, and Malaysia, the firm has been working on new compliance standards over the past few months, according to sources.
The tougher vetting process has now locked out more than 50% of Nvidia’s former buyers, a source said, but rejected firms can reapply after adjustments are made. So far, the crackdown has mostly affected “neoclouds,” or custom-built cloud platforms designed for AI computing.
The tougher screening is in keeping with Washington’s growing concern that advanced AI chips are still reaching China through intermediaries despite years of export restrictions.
Nvidia’s GPUs are one of the world’s most popular and sought-after processors for training and incorporating large language models and are a key target for companies looking to sidestep US controls. Nvidia is trying to strengthen its compliance by reducing the number of approved distributors and conducting more physical inspections to keep access to other international markets open.
Nvidia moves to conduct on-site visits for compliance checks
US pressure forced Nvidia to act, sources say, because the government intends to dismantle the broker networks that sell chips on the black market. In the past, Nvidia has always checked buyers for US export compliance, but the current vetting process has really doubled down on both paperwork requirements and real-world field audits. The new process involves auditing data centers, verifying client contracts, and interviewing users, with the US Department of Commerce providing official oversight and support.
The Trump administration has been looking to slow China’s access to advanced US technology for the last few months. In March, a Supermicro co-founder was even indicted for orchestrating a $2.5 billion scheme to evade supply chain controls and ship Nvidia chips to China via a Southeast Asian intermediary. The smuggling ring concealed high-tech hardware by removing it from their original Supermicro boxes and using unlabelled packaging to transport it, investigators said. Supermicro’s servers use Nvidia chips.
Moreover, later in May, the US Commerce Department issued strict guidelines to prevent Nvidia’s high-end Blackwell processors from being funneled to Chinese-based operations in places like Malaysia. Chinese firms had previously taken advantage of a US export ban loophole by buying the restricted hardware from their Singapore or Malaysia branches. Under the new guidance, the Bureau of Industry and Security (BIS) now mandates export licenses for top-tier AI processors whenever the parent company is based in China, regardless of where the hardware is physically shipped.
Beijing still bars the entry of Nvidia’s H200 chips
While Washington has tightened export controls on China, Beijing has also barred Nvidia from selling its H200 series to strengthen its domestic semiconductor industry.
At the moment, however, consumer demand still far outstrips local supply. Tech companies are actively lobbying Beijing to allow Nvidia’s H200, but without official clearance, large-scale shipments are effectively off the table for the foreseeable future.
The Chinese government is still counting on domestic chipmakers to expand production capacity fast enough to meet the shortfall. Authorities anticipate that overall local output will increase threefold by 2026’s close; however, manufacturing still lags behind most rivals’ capacity.
One tech executive even commented, “All domestic suppliers are sold out. Even those lower-grade chips that no one wanted before are sold out as long as they can be put to work somehow.”
Nonetheless, there’s some speculation that Beijing will allow some companies to buy the H200 series. According to insiders, authorities told Alibaba Group Holding (BABA), ByteDance, and DeepSeek that they would soon be granted access to the chips. However, the sources claimed that the government will restrict Nvidia’s chips to AI training, leaving inference workloads to domestic chips. The chips would also be limited to public data, not sensitive customer details. Then only about 200,000 of those chips will be allowed in, less than 50% of what firms had asked for earlier this year.
Nevertheless, Nvidia does not seem to be interested in leading the Chinese market. In May, the firm’s CEO Jensen Huang had noted, “Huawei is very, very strong. They had a record year; they’ll likely, very likely, have an extraordinary year coming up, and their local ecosystem of chip companies is doing quite well, because we’ve evacuated that market. We’ve really largely conceded that market to them.”
Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.
บทความแนะนำ
