An
informal market circumvents U.S. export controls through supply-chain blind spots—and
in one case, a student’s luggage
SINGAPORE—A 26-year-old Chinese student in Singapore packed suitcases last fall to return home for vacation. Besides his clothes and shoes, his luggage included six of Nvidia’s NVDA -1.31% advanced artificial intelligence chips.
A connection from college asked him to bring the chips because the U.S. restricted their export to China. Each chip was roughly the size of a Nintendo Switch game console, and the student didn’t flag any suspicions at the airport.
Upon arrival, the student said he was paid $100 for each chip he carried, a fraction of the underground market worth.
The student is part of a barely concealed network of buyers, sellers, and couriers bypassing the Biden administration’s restrictions to deny China access to Nvidia’s advanced AI chips, The Wall Street Journal has found. Nvidia’s chips are highly coveted for their ability to handle the massive computations needed to train AI systems critical to the China-U.S. tech rivalry.
More than 70 distributors are openly advertising online what they purport to be Nvidia’s restricted chips, and the Journal got in direct contact with 25 of them. Many of the verified sellers said they have supplies amounting to dozens of high-end Nvidia chips each month.
The flow of Nvidia chips is so steady that most of those sellers take preorders and promise delivery in weeks, the Journal found. Some also sold entire servers—costing upward of roughly $300,000—with each containing eight high-end Nvidia chips.
These merchants don’t sell enough of the powerful Nvidia processors to satiate a single tech giant’s demands. But for AI startups or research institutions with more modest needs, procurement can be done. Every Nvidia chip matters to China, which wants to stay competitive with the U.S. in an AI race seen as increasingly crucial to tech sovereignty and national security.
The Chinese sellers could be advertising Nvidia chips they don’t actually have or offering refurbished ones from older processors. But counterfeits would be nearly impossible to make, both in physical or performance terms, given the uniqueness of Nvidia’s top-end products.
The Journal’s review includes verifying purchases with Chinese buyers who used the underground channels, as well as access to transaction records, customs filings, and photos of the Nvidia chips up for sale. The Commerce Department, which oversees enforcement of the U.S. restrictions, didn’t respond to requests for comment.
Tracking troubles
Nvidia globally doesn’t sell its powerful data center chips individually or provide them directly to its AI customers. Instead, it ships them to third parties, such as Dell Technologies and Super Micro Computer SMCI 3.06%, which delivers fully-built AI servers or systems to those customers. These equipment providers often order more Nvidia chips than they need, in case demand unexpectedly surges or they run into manufacturing snags, according to industry participants. The equipment providers’ visibility would also be limited if the end buyer chooses to route the servers—and the Nvidia chips inside them—to elsewhere, the participants added.
Dell and Super Micro said they comply with U.S. export controls and act if illicit behavior is discovered.
Nvidia says it doesn’t sell its restricted advanced chips to China, in accordance with the U.S. export controls, and it works primarily with well-known partners to comply with the rules. “We apply the same standard to all transactions, large or small, and expect our partners to do the same,” an Nvidia spokesperson said.
Enforcement of the Biden administration’s export curbs on Nvidia chips largely rests with the Commerce Department and the myriad companies along the semiconductor supply chain. Many foreign governments and jurisdictions aren’t legally required to impose U.S. controls, and such chip sales to China aren’t generally considered a criminal offense in those places, according to international trade lawyers.
For instance, the student transporting the Nvidia chips didn’t break any Singaporean laws, according to the lawyers, as the tech components aren’t subject to any local export restrictions.
“Whether these transactions occur through distributors or intermediaries is very difficult to track,” said Frank Kung, an analyst at TrendForce who focuses on semiconductors and cloud data centers.
The free flow of Nvidia’s top-end chips in China began to dry up in late 2022 after Washington imposed a first round of export restrictions. Nvidia offered scaled-down versions for Chinese buyers, but Washington further limited access to such chips in October. That move led to the cancellation of hundreds of thousands of Nvidia orders, worth at least $5 billion.
Some Chinese companies have considered temporary solutions that can help them survive the U.S. restrictions until local chip makers such as Huawei Technologies are capable of making better alternatives, people familiar with the matter say.
The precise scale of the informal market for Nvidia’s advanced chips couldn’t be learned but is believed to be relatively small compared with the overall market. One estimate put the median number of AI chips smuggled annually at 12,500, according to an analysis by the Center for a New American Security, a Washington, D.C.-based think tank. However, Nvidia sold worldwide last year an estimated more than 2.6 million A100 and H100 chips, along with their scaled-down versions, according to tech research firm Omdia.
The power of ‘Brother Jiang’
The chips that the Chinese student carried to China were passed down from a mysterious broker in Singapore known as “Brother Jiang,” who is well-known among chip distributors and buyers in the region. In interviews with the Journal, he said he taps contacts at distribution channels and system integrators in Southeast Asia to help Chinese customers get chips and servers.
He said his clients include AI companies, research institutions, and chip resellers, some of which have used entities set up in Singapore, Malaysia, Vietnam, and Taiwan to circumvent U.S. restrictions.
After placing orders, he also helped his buyers arrange logistics. That process could mean an individual ferrying in the product or a more traditional delivery where Jiang helped prepare documents for customs declaration and contact with shipping companies.
“We don’t do big orders. That would be too conspicuous,” said the broker, who said he has worked in the cloud-computing and chip industry for more than a decade.
The system relies, in part, on incomplete paperwork that can avoid triggering authorities, some of the sellers say. In two March transactions, a Shenzhen, China-based merchant received shipments of 20 Nvidia GPUs from Singapore and an additional 40 from a Taiwan-based exporter—without specifying what the chip models were, according to customs filings seen by the Journal.
The seller told the Journal that the chips were one of the most off-limits: the high-end A100 processors.
There is always a way
The distributors that the Journal found have circulated product information in industry-group chats, online yellow pages of electronics suppliers, and e-commerce sites such as Alibaba’s Taobao and Idle Fish. Many run brick-and-mortar outlets in China’s leading AI-research hubs such as Shenzhen and Beijing.
These resellers typically asked $22,500 for an A100 chip and $32,400 for the higher-end H100—a premium versus the typical costs of roughly $10,000 and $25,000 for the pair of processors, respectively. The underground market prices have already more than halved from last summer, as supply has become more stable while panic buying has abated, the sellers said.
Some merchants said they have up to dozens of chips in stock at one time, and preorders of bigger quantities can be delivered in one to two weeks. The chips come in their original wholesale packaging, they said.
“It does become very hard, but don’t be silly, there is always a way,” said a Beijing-based distributor when asked about how they managed to get the chips. He added he has received a shipment every month, dozens of chips each, over the past few months.
Some sellers offer a three-year quality assurance plan, including replacing defective products and repairing malfunctions—but without involving Nvidia. Buyers said they were concerned about the uncertainty of unofficial aftersale services.
After Washington tightened export restrictions late last year, Chinese research institutes and universities continued to buy Nvidia’s high-end AI chips in limited quantities from resellers in China, according to official procurement documents reviewed by the Journal. The purchases include H100, a processor that Nvidia never officially shipped to China as the export restrictions took effect before the chips went on sale globally.
China’s elite Tsinghua University and state research powerhouse Chinese Academy of Science have been among the main buyers of Nvidia’s advanced AI chips, according to the documents.
Chinese tech companies including U.S.-sanctioned Huawei are attempting to make chips as capable as Nvidia’s but are faced with capacity and technology bottlenecks, people familiar with the matter say. Training large AI models with domestic chips is prone to system crashes, a top AI research institution told Chinese Premier Li Qiang in March, according to a presentation reviewed by the Journal.
Until viable domestically-made chips emerge in China, the market for Nvidia’s high-end products—and the supply chain that has sprouted around them—will likely remain robust and adaptive. The Chinese student who brought Nvidia processors in his suitcase said he is willing to transport the tech components again.
“I’m glad I was able to do something for my country—and make a little extra money,” the student said. “So, why not?”
0 Comments