Two days after the Trump administration restricted his company’s sales to China, Nvidia CEO Jensen Huang was in Beijing on April 17 to salvage what he referred to as a very important market.
Huang, whose visit merited a reception by Chinese Vice Premier He Lifeng at the Great Hall of the People, also met with Ren Hongbin, chairman of the China 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 US firms, including Nvidia, to deepen their presence in the Chinese market and leverage their strengths here to gain an edge in global competition.” In reply, Huang said, “We look forward to deepening our presence in China and supporting the advancement of the local tech ecosystem.”
In a separate meeting arranged by CCPIT, Huang told Ren, “We hope to continue to cooperate with China.” Chinese media also quoted him as saying that “Nvidia will continue to make every effort to optimize its product lineup in compliance with regulatory requirements and will steadfastly serve the Chinese market.”
In a meeting with Liang, Huang reportedly talked about how Nvidia might provide DeepSeek with AI processors that meet both the company’s needs and regulatory requirements.
Nvidia issued a statement saying, “We regularly meet with government leaders to discuss our company’s products and technology,” but these were not ordinary meetings. The company is now at the center of the increasingly acrimonious US-China trade and tech disputes.
It was an eventful week for Nvidia. In the evening of April 15, Nvidia revealed that exports of its H20 AI processors and similar devices to China and other countries of concern now require a license from the US government, an order that “addresses the risk that the covered products may be used in, or diverted to, a supercomputer in China.”
In after-hours trading, Nvidia’s share price dropped 6.3% to US$105.10. By Thursday’s close (Friday was a holiday), it was down to $101.42, bringing its year-to-date decline to 26.7%.
With no license likely to be granted, Nvidia stated that its results for the current fiscal quarter ending on April 27 “are expected to include up to approximately $5.5 billion of charges associated with H20 products for inventory, purchase commitments, and related reserves.”
AMD, whose MI308 AI accelerators are subject to the same new restriction, dropped 7.1% in after-hours trading on Tuesday and finished the week down 27.5% since the beginning of the year. AMD expects to post special charges approaching $800 million.
Intel’s Gaudi 3 processor is also affected. For this and other reasons, Intel’s share price was down 27% in the month to Thursday.
Like Nvidia’s H20, AMD’s MI308 was designed specifically with reduced performance to meet the requirements of previous US government restrictions on exports to China. So were redesigned versions of Intel’s Gaudi 2 and Gaudi 3 AI accelerators.
This is the third time since October 2022 that the US Commerce Department’s Bureau of Industry and Security (BIS) has put a ceiling on the performance of AI processors that can be exported to China and then, after Nvidia, AMD and Intel designed new, less efficient versions of their chips, lowered the ceiling. In this regard, President Trump is following the same policy as President Joe Biden.
For Nvidia, the restriction timeline runs as such:
- October 2022: Biden administration blocked exports of Nvidia’s A100 and H100 GPUs, then the company’s most advanced AI processors.
- November 2022: Nvidia launched the dumbed-down A800, which met BIS requirements for sale in China.
- March 2023: Nvidia launched the H800, a low-performance version of the H100.
- November 2023: BIS blocked exports of A800 and H800 GPUs.
- March 2024: Nvidia launched the H20, which met the new, tighter BIS requirements.
- April 2025: Trump administration blocked exports of H20 GPUs.
The reasons for this chain of events are, first, that Chinese demand remained strong even as the performance of the available chips declined and, second, that Chinese AI capabilities continued to advance despite the restrictions.
In addition to revealing the behind-the-curve incompetence of US government analysis and response, it demonstrates that, in the case of semiconductors, anything the Chinese want to buy, the US will refuse to sell, all the while complaining about its trade deficit with China.
The stunning success of Chinese AI model DeepSeek, which was trained using Nvidia H800 chips, set off another wave of McCarthyite paranoia among US politicians. As The New York Times pointed out in January, DeepSeek “built a cheaper, competitive chatbot with fewer high-end computer chips than US behemoths like Google and OpenAI, showing the limits of chip export control.”
On April 16, Chairman John Moolenaar (R-MI) and Ranking Member Raja Krishnamoorthi (D-IL) of the House Select Committee on the Chinese Communist Party released a report entitled “DeepSeek Unmasked: Exposing the CCP’s Latest Tool for Spying, Stealing, and Subverting US Export Control Restrictions.” The committee calls DeepSeek “a serious national security threat to the United States.”
Chairman Moolenaar stated that:
DeepSeek isn’t just another AI app — it’s a weapon in the Chinese Communist Party’s arsenal, designed to spy on Americans, steal our technology, and subvert US law. We now know this tool exploited US AI models and reportedly used advanced Nvidia chips that should never have ended up in CCP hands. That’s why we’re sending a letter to Nvidia to demand answers. American innovation should never be the engine of our adversaries’ ambitions.
However, only two days earlier, on April 14, Nvidia announced plans to produce up to $500 billion worth of supercomputers and other AI infrastructure in the US in the next four years.
To do so, Nvidia is working with Taiwanese contract manufacturers Foxconn (Hon Hai Precision) and Wistron to build factories in the US state of Texas. The Blackwell AI processors at the heart of the AI supercomputers made there will be supplied by Taiwanese semiconductor foundry TSMC from its factories in Arizona.
The chips will be assembled, packaged 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, “The engines of the world’s AI infrastructure are being built in the United States for the first time. Adding American manufacturing helps us better meet the incredible and growing demand for AI chips and supercomputers, strengthens our supply chain and boosts our resiliency.”
That was exactly what Trump wanted to hear. “It’s the Trump Effect in action,” read a press release from the White House.
Building AI supercomputers in the US won’t be cheap and might take longer than Nvidia hopes, but with so many top-notch companies on board, it should eventually get done.
At the end of March, Chinese server maker H3C reported that it was running short of Nvidia H20 processors, which appear to be nearly sold out in China. Alibaba, Tencent, ByteDance and other Nvidia customers will also be affected, indicating that the new BIS restrictions are likely to disrupt China’s AI computing industry as intended.
The restrictions could also cost Nvidia up to $15 billion in annual sales in addition to the $5.5 billion in anticipated special charges, with that loss translating into gains for Chinese AI chip designers Huawei and Cambricon, and H3C, which is developing its own solution.
DeepSeek is already using Huawei’s new Ascend 910C AI processor, the apparent most advanced Chinese alternative to Nvidia. Cambricon, which was founded in 2016, is much smaller than Huawei but has become a darling of the Chinese stock market, rising by about five times over the past year. Huawei is not publicly traded.
DeepSeek has already been rolled out across China to consumers, corporations, finance and other businesses, city government, healthcare and other social services, and support services for the People’s Liberation Army. Rather than having been designed “to spy on Americans”, as Congressman Moolenaar claims, it aims to provide AI solutions to practical problems across Chinese society.
Meanwhile, Dylan Patel and his colleagues at the highly respected technology newsletter SemiAnalysis write that Huawei’s new CloudMatrix 394 AI accelerator, which is based on the Ascend 910C, “competes directly” with Nvidia’s top-end GB200 Grace Blackwell Superchip and “in some metrics is more advanced than Nvidia’s rack scale solution.”
“Rack scale solution” refers to the complete data center system, including GPUs, servers, networking, storage, power management and cooling.
In the view of SemiAnalysis, “The engineering advantage is at the system level not just at the chip level, with innovation at the accelerator, networking, optics, and software layers… Huawei is a generation behind in chips, but its scale-up solution is arguably a generation ahead of Nvidia and AMD’s current products on the market.”
Huawei’s solutions used more electricity, but SemiAnalysis concludes that “The deficiencies in power are relevant but not a limiting factor in China.”
Once again, US sanctions appear to be too little, too late, and more likely to promote rather than prevent the advance of Chinese technology. Unless Trump changes tack in a deal with Beijing, Nvidia and its smaller American rivals are likely to find themselves increasingly marginalized in the booming Chinese market for AI processors.
Follow this writer on X: @ScottFo83517667
