According to a report citing information from the tender documents, Chinese military entities, state-run artificial intelligence research institutes, and universities have over the past year purchased small quantities of Nvidia semiconductors prohibited by the U.S. from sale to China.
The transactions by mainly unidentified Chinese sources show how difficult it will be for Washington, even with its sanctions, to fully cut off China's access to cutting-edge American chips that could support AI advances and powerful military computers.
It is permissible for Chinese citizens to purchase or sell expensive American chips, and according to publicly accessible tender records, numerous Chinese companies have purchased and received Nvidia semiconductors since the restrictions were put in place.
These include the slower A800 and H800 chips Nvidia later produced for the Chinese market, which were also banned in October, as well as its more potent A100 and H100 chip, whose exports to China and Hong Kong were prohibited in September 2022.
Because they can process the massive volumes of data required for machine-learning jobs more efficiently than competitors' products, Nvidia's chip type, graphic processing units, is commonly regarded as being considerably superior to them for AI work.
Even if competing goods from Huawei and other companies are only getting started, Chinese companies still lack effective alternatives, as seen by the ongoing demand for and access to restricted Nvidia chips.
Nvidia held a 90% market share in China for AI chips prior to the sanctions.
Among the purchasers were prestigious universities and two organisations that are restricted from export by the United States: the University of Electronic Science and Technology of China and the Harbin Institute of Technology. These organisations have been charged with engaging in military affairs or having ties to a military organisation that goes against the interests of the United States as a nation.
Six Nvidia A100 chips were purchased by the former in May in order to develop a deep learning model. December 2022 saw the latter buy one A100. Its goal remained unknown.
There were no comments on the issue from none of the purchasers mentioned in the article.
The review report discovered that Nvidia and shops with company approval were not included in the list of suppliers. How the providers obtained their Nvidia chips was unclear.
However, an underground market for these chips has emerged in China following U.S. bans. Chinese suppliers have previously claimed that they either import through locally established businesses in locations like Singapore, Taiwan, and India, or they scoop up extra stock that ends up on the market after Nvidia sends huge volumes to major U.S. organisations.
According to Nvidia, all relevant export control requirements are complied with, and it mandates that its clients follow suit.
"If we learn that a customer has made an unlawful resale to third parties, we'll take immediate and appropriate action," a company spokesperson said.
Commenting was rejected by the US Department of Commerce. U.S. authorities have pledged to seal gaps in the export prohibitions and have taken steps to restrict the chips' availability to subsidiaries of Chinese businesses based abroad.
Professor Chris Miller of Tufts University and author of "Chip War: The Fight for the World's Most Critical Technology" stated that it was impractical to expect that export regulations from the United States could be completely impenetrable, considering chips' small size and ease of smuggling.
He noted that the primary goal is "to throw sand in the gears of China's AI development" by making it challenging to assemble sizable clusters of cutting-edge chips that can be used to train AI systems.
The examination included over 100 contracts where state-affiliated companies have purchased A100 chips, and some tenders from the October ban indicate A800 acquisitions.
Additionally, according to tenders released last month, Tsinghua University purchased two H100 chips, while a Ministry of Industry and Information Technology laboratory purchased one.
Tenders from a military database show that among the purchasers is one unidentified People's Liberation Army organisation with its headquarters in Wuxi, Jiangsu province. It searched for one H100 chip this month and three A100 chips in October.
Because Chinese military tenders are sometimes extensively censored, the report said that it was unable to find out who won the bids or why the purchase was made.
Most tenders demonstrate that AI is being employed with the chips. However, the majority of transactions are made in extremely tiny volumes, significantly smaller than what is required to start from scratch and create a massive, complex AI language model.
More than 30,000 Nvidia A100 cards would be needed for a model like OpenAI's GPT, according to research firm TrendForce. However, a small number are able to do intricate machine learning tasks and improve on current AI models.
For instance, last month, Shandong Chengxiang Electronic Technology was given a contract worth 290,000 yuan ($40,500) by the Shandong Artificial Intelligence Institute for 5 A100 chips.
A lot of the tenders require vendors to pay for the products only after they have been delivered and installed. The majority of colleges also released notifications indicating the transaction's completion.
Dubbed the Massachusetts Institute of Technology of China, Tsinghua University has bought about 80 A100 chips since the 2022 prohibition and is a frequent tender issuer.
Chongqing University released a call for bids in December for a single A100 chip, making it clear that the chip had to be "brand new" and could not be used or dismantled. This message indicated that the delivery had finished this month.
(Source:www.reuters.com)
The transactions by mainly unidentified Chinese sources show how difficult it will be for Washington, even with its sanctions, to fully cut off China's access to cutting-edge American chips that could support AI advances and powerful military computers.
It is permissible for Chinese citizens to purchase or sell expensive American chips, and according to publicly accessible tender records, numerous Chinese companies have purchased and received Nvidia semiconductors since the restrictions were put in place.
These include the slower A800 and H800 chips Nvidia later produced for the Chinese market, which were also banned in October, as well as its more potent A100 and H100 chip, whose exports to China and Hong Kong were prohibited in September 2022.
Because they can process the massive volumes of data required for machine-learning jobs more efficiently than competitors' products, Nvidia's chip type, graphic processing units, is commonly regarded as being considerably superior to them for AI work.
Even if competing goods from Huawei and other companies are only getting started, Chinese companies still lack effective alternatives, as seen by the ongoing demand for and access to restricted Nvidia chips.
Nvidia held a 90% market share in China for AI chips prior to the sanctions.
Among the purchasers were prestigious universities and two organisations that are restricted from export by the United States: the University of Electronic Science and Technology of China and the Harbin Institute of Technology. These organisations have been charged with engaging in military affairs or having ties to a military organisation that goes against the interests of the United States as a nation.
Six Nvidia A100 chips were purchased by the former in May in order to develop a deep learning model. December 2022 saw the latter buy one A100. Its goal remained unknown.
There were no comments on the issue from none of the purchasers mentioned in the article.
The review report discovered that Nvidia and shops with company approval were not included in the list of suppliers. How the providers obtained their Nvidia chips was unclear.
However, an underground market for these chips has emerged in China following U.S. bans. Chinese suppliers have previously claimed that they either import through locally established businesses in locations like Singapore, Taiwan, and India, or they scoop up extra stock that ends up on the market after Nvidia sends huge volumes to major U.S. organisations.
According to Nvidia, all relevant export control requirements are complied with, and it mandates that its clients follow suit.
"If we learn that a customer has made an unlawful resale to third parties, we'll take immediate and appropriate action," a company spokesperson said.
Commenting was rejected by the US Department of Commerce. U.S. authorities have pledged to seal gaps in the export prohibitions and have taken steps to restrict the chips' availability to subsidiaries of Chinese businesses based abroad.
Professor Chris Miller of Tufts University and author of "Chip War: The Fight for the World's Most Critical Technology" stated that it was impractical to expect that export regulations from the United States could be completely impenetrable, considering chips' small size and ease of smuggling.
He noted that the primary goal is "to throw sand in the gears of China's AI development" by making it challenging to assemble sizable clusters of cutting-edge chips that can be used to train AI systems.
The examination included over 100 contracts where state-affiliated companies have purchased A100 chips, and some tenders from the October ban indicate A800 acquisitions.
Additionally, according to tenders released last month, Tsinghua University purchased two H100 chips, while a Ministry of Industry and Information Technology laboratory purchased one.
Tenders from a military database show that among the purchasers is one unidentified People's Liberation Army organisation with its headquarters in Wuxi, Jiangsu province. It searched for one H100 chip this month and three A100 chips in October.
Because Chinese military tenders are sometimes extensively censored, the report said that it was unable to find out who won the bids or why the purchase was made.
Most tenders demonstrate that AI is being employed with the chips. However, the majority of transactions are made in extremely tiny volumes, significantly smaller than what is required to start from scratch and create a massive, complex AI language model.
More than 30,000 Nvidia A100 cards would be needed for a model like OpenAI's GPT, according to research firm TrendForce. However, a small number are able to do intricate machine learning tasks and improve on current AI models.
For instance, last month, Shandong Chengxiang Electronic Technology was given a contract worth 290,000 yuan ($40,500) by the Shandong Artificial Intelligence Institute for 5 A100 chips.
A lot of the tenders require vendors to pay for the products only after they have been delivered and installed. The majority of colleges also released notifications indicating the transaction's completion.
Dubbed the Massachusetts Institute of Technology of China, Tsinghua University has bought about 80 A100 chips since the 2022 prohibition and is a frequent tender issuer.
Chongqing University released a call for bids in December for a single A100 chip, making it clear that the chip had to be "brand new" and could not be used or dismantled. This message indicated that the delivery had finished this month.
(Source:www.reuters.com)