Two days after the Trump administration limited his business sales to China, the CEO of Nvidia, Jensen Huang, was in Beijing on April 17 to save what he called a very important market.
Huang, whose visit deserves a reception of the Chinese Deputy Prime Minister, He Lifeng, in the Greater People’s Hall, also met Ren Hongbin, President of the Chinese Council for the Promotion of International Trade (CCPIT), and Liang Wenfeng, founder and CEO of Deepseek.
He, who is also a member of the Politburo, said: “We welcome more American companies, notably Nvidia, to deepen their presence on the Chinese market and take advantage of their forces here to gain an advantage in global competition.” In response, Huang said: “We are impatient to deepen our presence in China and support the progress of the local technological ecosystem.”
During a separate meeting organized by CCPIT, Huang told Ren: “We hope to continue to cooperate with China.” The Chinese media also quoted it saying that “Nvidia will continue to do everything possible to optimize its range of products in accordance with regulatory requirements and will firmly serve the Chinese market.”
At a meeting with Liang, Huang would have explained how Nvidia could provide Deepseek with AI processors who meet both the needs of the company and the regulatory requirements.
Nvidia has published a statement saying: “We regularly meet government leaders to discuss our business products and technology”, but these are not ordinary meetings. The company is now at the center of commercial and technological conflicts of American China increasingly acrimonious.
It was an eventful week for Nvidia. On the evening of April 15, Nvidia revealed that exports from its AI H20 processors and devices similar to China and other worrying countries now require a license from the US government, an order that “addresses the risk that covered products can be used or diverted to, a supercomputer in China”.
In the discussion after opening hours, Nvidia’s share price dropped by 6.3% to US $ 105.10. At the end of Thursday (Friday was a holiday), it fell to $ 101.42, bringing its drop in the year to 26.7%.
Without a license likely to be granted, Nvidia said that its results for the current quarter finished on April 27 “should include up to around 5.5 billion dollars in costs associated with H20 products for inventory, purchase commitments and related reserves.”
AMD, whose accelerators of the MI308 AI are subject to the same new restriction, fell 7.1% in the event of negotiation after working hours on Tuesday and ended the week down 27.5% since the start of the year. AMD expects to display special costs approaching $ 800 million.
The Intel Gaudi 3 processor is also affected. For this reason and for others, Intel’s share has dropped from 27% from month to Thursday.
Like the H20 of NVIDIA, the AMD MI308 was designed specifically with reduced performance to meet the requirements of the previous American government’s restrictions on exports to China. The redesigned versions were therefore redesigned by Intel Gaudi 2 and Gaudi 3 AI accelerators.
It is the third time since October 2022 that the Office of the US Trade Department and Safety (BIS) has put a ceiling on the performance of AI processors which can be exported to China, then, after NVIDIA, AMD and Intel designed new less effective versions of their chips, have lowered the ceiling. In this regard, President Trump follows the same policy as President Joe Biden.
For Nvidia, the restriction calendar works as such:
- October 2022: Administration Biden blocked the exports of the A100 and H100 GPUs of Nvidia, then the most advanced IA processors of the company.
- November 2022: NVIDIA launched the stupid A800, which met the requirements of bis in China.
- March 2023: NVIDIA launches the H800, a low performance version of the H100.
- November 2023: Bis Blocked Exports from GPU A800 and H800.
- March 2024: NVIDIA launched the H20, which met the new, more strict requirements.
- April 2025: The Trump administration blocked GPU H20 exports.
The reasons for this chain of events are, first of all, that Chinese demand has remained strong even if the performance of the available fleas has decreased and, secondly, that the capacities of the Chinese AI continued to advance despite the restrictions.
In addition to revealing incompetence in the curve in the analysis and response of the American government, he demonstrates that, in the case of semiconductors, all that the Chinese want to buy, the United States will refuse to sell, while complaining about its trade deficit with China.
The surprising success of the Chinese AI model Deepseek, which was formed using Nvidia H800 chips, sparked another wave of paranoia of McCarthyite among American politicians. As the New York Times pointed out in January, Deepseek “built a cheaper and competitive chatbot with fewer high -end computer chips than American giants like Google and Openai, showing the limits of flea export control”.
On April 16, President John Moolenaar (R-MI) and the Raja Krishnamoorthhi (I-I) ranking of the selective committee of the Chinese Communist Party published a report entitled “Deepseek Unmasked: Exhibition of the latest CCP for espionage, theft and subverter of American export control restrictions.” The committee calls Deepseek “a serious national security threat to the United States”.
President Moolenaar said:
Deepseek is not only another AI application – it is a weapon in the arsenal of the Chinese Communist Party, designed to spy on the Americans, steal our technology and overturn American law. We now know that this tool has exploited us models of AI and would have used advanced Nvidia fleas which should never have been in the hands of the CCP. This is why we send a letter to Nvidia to demand answers. American innovation should never be the engine of the ambitions of our opponents.
However, only two days earlier, on April 14, NVIDIA announced its intention to produce up to $ 500 billion in superordinators and other AI infrastructure in the United States over the next four years.
To do this, Nvidia works with manufacturers of Taiwanese contracts Foxconn (Hon Hai Precision) and Wistron to build factories in the American state of Texas. The Blackwell AI processors at the heart of AI supercomputers are provided by the Taiwanese foundry of TSMC semiconductors of its factories in Arizona.
The chips will be assembled, wrapped and tested by Amkor and Spil (Silicon Precision Industries Co., Ltd.), also in Arizona. Amkor, as the name suggests, is Korean-American; Spil is Taiwanese.
CEO Huang said: “IA global infrastructure engines are under construction in the United States for the first time. The addition of American manufacturing helps us to better meet the incredible and increasing demand for fleas and superordinators of AI, strengthens our supply chain and increases our resilience. ”
It was exactly what Trump wanted to hear. “This is the Trump effect in action,” read a press release from the White House.
The construction of AI supercomputer in the United States will not be cheap and could take more time than Nvidia hopes, but with so many first-rate companies on board, it should eventually be done.
At the end of March, the Chinese server manufacturer H3C said it was missing from Nvidia H20 processors, which seem almost exhausted in China. Alibaba, Tencent, Bytedance and other Nvidia customers will also be assigned, which indicates that the new bis restrictions should disturb the IT industry of Chinese AI as expected.
The restrictions could also cost Nvidia up to $ 15 billion in annual sales in addition to 5.5 billion dollars of special expenses, with this loss translating into gains for designers of Chinese ia Huawei and Cambcon, and H3C, which develops its own solution.
Deepseek already uses the new Huawei AI Ascen 910C processor, the most apparent Chinese alternative to Nvidia. Cambcon, which was founded in 2016, is much smaller than Huawei but has become a darling of the Chinese stock market, increasing about five times in the past year. Huawei is not listed on the stock market.
Deepseek has already been deployed through China to consumers, companies, finance and other companies, the city government, health care and other social services, and support for the Popular Liberation Army. Rather than having been designed “to spy on the Americans”, as the Moolenaar Congress member claims, he aims to provide AI solutions to practical problems in Chinese society.
Meanwhile, Dylan Patel and his colleagues from the highly respected of Newsletter Semi-Aléanalysis in technology write that the new Accelerator of Huawei AI, which is based on Ascend 910c, “competing directly” with the more advanced rail solution of Nvidia GB200.
“Rack scale solution” refers to the complete system of the data center, including GPUs, servers, networking, storage, power management and cooling.
In the opinion of semianalysis, “the advantage of engineering is at the level of the system not only at the level of the chips, with innovation in the accelerator, networking, optics and software layers … Huawei is a generation behind the chips, but its scaling solution is undoubtedly a generation before the current products of Nvidia and AMD on the market.”
Huawei’s solutions have used more electricity, but semiianalysis concludes that “power deficiencies are relevant but not a limiting factor in China”.
Again, American sanctions seem to be too little, too late and more likely to promote rather than prevent the progress of Chinese technology. Unless Trump has changed an agreement with Beijing, Nvidia and its small American rivals are likely to be more and more marginalized on the booming Chinese market for AI processors.
Follow this writer on X: @ scottfo83517667
