Select Language:
The United States has given conditional approval for Nvidia to sell its H200 chips, some of its most advanced AI training and processing products, to China. However, this approval comes with the stipulation that 25% of each sale proceeds must be remitted to the U.S. government.
Nvidia is authorized to sell its H200 chips to “approved customers” in China and other regions, as announced by the U.S. President on December 8. He also stated that the Department of Commerce is working out the final details, and that similar rules will soon apply to other major American tech companies like AMD and Intel.
The H200 chip, launched by Nvidia in November 2023, began deliveries last year. It is specifically engineered for training and deploying generative AI models, conducting large-scale scientific computing, and analyzing enormous datasets. Its performance surpasses that of the H20 chip, which was tailored for the Chinese market.
Nvidia expressed pride in its decision to sell H200 chips to commercial clients, viewing it as a positive step.
In 2023, Nvidia introduced the H20 chip following a U.S. government ban on selling more advanced chips to China. That ban was further expanded this April amid escalating trade tensions. However, by July, Nvidia announced it could resume chip sales to China, after receiving assurances that export licenses would be approved.
In an interview with China Central Television in July, Jensen Huang emphasized the uniqueness of this market, stating, “The industry is full of innovation and energy. Despite ongoing trade tensions, China remains an indispensable market.” He added, “If we’re not involved, Chinese tech companies will develop their own solutions. Many cloud providers are already working on their own chips, and China’s AI industry will continue to grow regardless of Nvidia’s involvement.”
While Nvidia was largely absent from the Chinese market during this period, domestic AI chip producers like Huawei Technologies, Moore Threads, MetaX, Cambricon Technologies, and Hygon Information Technology made significant progress.
However, Wu Wenchang, an analyst at a financial news outlet, pointed out that although Chinese firms have succeeded in manufacturing chips, the quality still falls short. Many major manufacturers faced lower-than-expected capital expenditures in the first half of the year, largely due to difficulties in sourcing buyers for their chips.
Tencent reported that its graphics processing units are sufficient for their needs, with adequate resources. Additionally, a senior executive from Baidu recently mentioned that most inference tasks are now being run on their Kunlun P800 chips. Furthermore, Zhou Hongyi, founder of 360 Security Technology, indicated that his company has shifted to using domestically produced chips.
ChatGPT
Perplexity
Gemini AI
Grok AI
