China has had to develop most of the tech that powers its AI and Web3 industries, mainly modern-day, AI and Machine Learning capable GPUs, on its own, because of US banning its tech companies from working with China-based businesses. Not just that, the US has been able to cajole businesses based in most of its allies as well to stop working with Chinese tech companies. Because Chinese chip manufacturers do not have the expertise in designing their own chips at a level that AMD, Intel or NVIDIA does, the silicon designed in China, is not nearly as capable as those produced by the big three. This has led to a massive smuggling ring inside China, which sells high-end, AI and ML-capable GPUs, like the NVIDIA A100 and H100 at an exorbitant markup. A new career is born Despite a US government ban on exporting certain products from chip suppliers AMD and NVIDIA to China, this underground trade has prospered. NVIDIA, which holds a near monopoly on GPUs used for training AI systems, faces restrictions from the ban imposed in August last year. Several, semiconductor engineers based in Shanghai, have told American news outlets about the the lucrative nature of sourcing smuggled GPUs due to the strong domestic demand for NVIDIA’s A100 and H100 GPUs. This has led to the creation of a new profession - procuring high-end NVIDIA GPUs to fulfil the requirements of Chinese tech firms developing AI. On Douyin, China’s version of TikTok, there are numerous vendors who are offering NVIDIA GPUs. However, verifying access to the highly sought-after A100 and H100 products is challenging. On e-commerce platform JD.com, an authorized reseller of NVIDIA lists the company’s 80-gigabyte A800 GPU for 88,999 yuan, while the 80G A100 is priced at 99,999 yuan but is currently out of stock. Exorbitant markups In Shenzhen’s Huaqiangbei subdistrict, the world’s largest electronics wholesale market, a vendor claims to have sourced NVIDIA’s A100 GPUs and charges 128,000 yuan or $17,709, surpassing the manufacturer’s suggested retail price of approximately $10,000. The Huaqiangbei vendor also mentions the availability of modified A100 GPUs priced between 80,000 yuan and 90,000 yuan. NVIDIA’s A800 is permitted for export to China as it has reduced capabilities compared to the original A100. The modified Chinese version of the H100 is called the H800. The high demand for smuggled GPUs in China underscores the country’s lack of strong alternative suppliers capable of delivering products that rival NVIDIA’s offerings. Chinese tech giants working with smugglers Although Shanghai-based Iluvatar Corex claimed to have the only domestically produced GPU, the Tiangai 100, major Chinese tech companies continue to prioritize obtaining NVIDIA products. For example, ByteDance, the owner of TikTok and Douyin, reportedly ordered $1 billion worth of GPUs from NVIDIA this year. According to Tang, an intermediary based in Shanghai who deals in smuggled NVIDIA GPUs, the A100 is now being sold for prices ranging from 130,000 yuan to 150,000 yuan in the Yangtze River Delta, a region with a high concentration of AI start-ups. Tang explained that training an AI system is just the beginning, and companies looking to offer large language model-based services require even more computational power. Large language models (LLMs) are deep-learning AI algorithms that can perform tasks such as recognition, summarization, translation, prediction, and content generation using extensive datasets. These models serve as the foundation for training AI chatbots like Microsoft-backed OpenAI’s ChatGPT. Tang noted that Chinese Big Tech companies have been purchasing NVIDIA’s A800 chips in large quantities in recent months, leading to a shortage in the local market. As a result, the prices of A100 and H100 GPUs obtained through unofficial channels have been driven even higher. He mentioned that two of his customers had to pay more than 1 million yuan in advance to secure their supply for the upcoming shipment of A800 GPUs. High demand, means more money is pumped into smuggling operations The limited access to advanced chips from companies like AMD and NVIDIA is anticipated to impede China’s progress in various cutting-edge applications, including the development of large language models (LLMs) and autonomous driving. Currently, NVIDIA’s H100, A100, and A800 GPUs, along with AMD’s MI250 and MI250X series, are considered the primary products used in AI-related computing applications. NVIDIA holds an 80% market share in server GPUs worldwide. TrendForce reports that Baidu, in the development of its ChatGPT alternative called Ernie Bot, initially used NVIDIA’s A100 GPUs on its servers but had to switch to the A800 due to US export controls. Baidu’s demand for AI servers is expected to reach approximately 2,000 units this year. As a result, the demand for NVIDIA’s China-specific A800 and H800 GPUs led to a continuous increase in their prices. Read all the Latest News , Trending News , Cricket News , Bollywood News , India News and Entertainment News here. Follow us on Facebook_,_ Twitter and Instagram_._
Because of the US tech ban on China, and the growing demands of the Chinese AI industry, some of China’s largest tech companies are working with smugglers to get their hands on the latest AI and ML-enabled GPUs that NVIDIA makes
Advertisement
End of Article


)

)
)
)
)
)
)
)
)
