[ad_1]
Written by Eduardo Baptista
BEIJING (Reuters) – Over the past year, Chinese military agencies, state-run artificial intelligence research institutes and universities have been purchasing small quantities of Nvidia semiconductors, which are banned from export to China by the United States, a Reuters bid revealed. This was revealed through document research.
Sales by little-known Chinese supplier completely cut off China’s access to advanced US-made chips that could power breakthroughs in AI and advanced computers for the military, despite ban This highlights the difficulties faced by the US government.
Buying and selling high-end U.S. chips is not illegal in China, and published bidding documents show that dozens of Chinese companies have bought and received Nvidia chips since the restrictions were imposed.
These include the company’s A100 chip and the more powerful H100 chip, which was banned from export to China and Hong Kong in September 2022, as well as the chip that Nvidia was developing for the Chinese market at the time but also banned in October last year. Contains banned and slow A800 and H800 chips.
The graphics processing unit (a type of chip) developed by Nvidia is widely considered to be far superior to competing products in AI work because it can more efficiently process the large amounts of data required for machine learning tasks. Masu.
Continued demand for and access to banned NVIDIA chips highlights the lack of good alternatives for Chinese companies, even as rivals such as Huawei are in early stages of development. There is. Before the ban, Nvidia had a 90% share of China’s AI chip market.
In addition to elite universities, the buyers include two entities subject to U.S. export restrictions, Harbin Institute of Technology and China Electronics, which have been accused of involvement in military affairs and ties to military institutions opposed to the United States. It also included the University of Science and Technology. National interest.
The former purchased six Nvidia A100 chips in May to train deep learning models. The latter purchased one A100 in December 2022, but its use has not been specified.
None of the buyers mentioned in this article responded to requests for comment.
A Reuters investigation found that none of the identified suppliers included Nvidia or its authorized retailers. It’s not clear how the supplier sourced the Nvidia chips.
But in response to U.S. regulations, an underground market for such chips has emerged in China. Chinese vendors have previously said Nvidia is taking surplus inventory that enters the market after being shipped in bulk to major U.S. companies or imported through local subsidiaries in India, Taiwan, Singapore and other countries.
Reuters contacted 10 suppliers named in the bidding documents, including those mentioned in this article, for comment, but none responded.
Nvidia said it complies with all applicable export control laws and expects its customers to do the same.
A company spokesperson said, “If we discover that a customer has illegally resold products to a third party, we will immediately take appropriate action.”
The U.S. Department of Commerce declined to comment. U.S. authorities have vowed to close loopholes in export controls and are moving to restrict access to chips by Chinese corporate units based outside China.
Chris Miller, a professor at Tufts University and author of “The Chip Wars: The Battle for the World’s Most Important Technology,” said given that chips are small and can be easily smuggled, he doubts U.S. export restrictions will be complete. He said it was unrealistic to think about it.
The main goal is to “throw sand in the wheels of China’s AI development” by making it harder to build large clusters of advanced chips that can train AI systems, he added.
military, AI buyers
The investigation includes more than 100 bids in which state agencies procured A100 chips, as well as dozens of bids indicating purchases of A800 since the October ban.
The tender, published last month, also showed Tsinghua University procuring two H100 chips and a research institute run by the Ministry of Industry and Information Technology procuring one.
According to military database bids, buyers include an anonymous People’s Liberation Army organization based in Wuxi, Jiangsu province. In October I asked for 3 A100 chips and this month I asked for 1 H100 chip.
Chinese military bids are often heavily redacted, making it impossible for Reuters to learn who won the bids or why they were purchased.
Most bids indicate that the chip is being used for AI. However, most purchases are very small amounts, far from the amounts needed to build sophisticated AI large-scale language models from scratch.
According to research firm TrendForce, a model similar to OpenAI’s GPT would require more than 30,000 Nvidia A100 cards. However, only a handful can perform complex machine learning tasks and enhance existing AI models.
As an example, the Shandong Provincial Artificial Intelligence Research Institute signed a 290,000 yuan ($40,500) contract for five A100 chips to Shandong Chengxiang Electronic Technology last month.
Many tenders stipulate that the supplier must deliver and install the product before receiving payment. Most universities also issued notices indicating that the transaction was completed.
Tsinghua University, known as the Massachusetts Institute of Technology of China, has issued numerous bids and has purchased around 80 A100 chips since the 2022 ban.
In December, Chongqing University announced a bid for one A100 chip, specifying that it cannot be used or disassembled, but must be “new.” I received a notification that the delivery was completed this month.
(1 dollar = 7.1592 Chinese yuan)
(Reporting by Eduardo Baptista; Additional reporting by Josh Ye in Hong Kong and Brenda Goh in Shanghai; Editing by Edwina Gibbs)
[ad_2]
Source link
