Nvidia CEO says US export controls on AI chips are a 'failure'
Jensen Huang says 'formidable' Chinese rival Huawei is innovating fast
TAIPEI -- Nvidia CEO Jensen Huang says U.S. export controls on AI chips to China are a "failure," as they have only motivated the country to develop its own solutions and allowed Chinese players to secure a bigger share of their massive home market.
The head of the American chip giant told reporters in Taipei on Wednesday that Nvidia had lost a lot of market share in China in recent years.
"Four years ago, at the beginning of the Biden administration, Nvidia's market share in China was nearly 95%. Today it is only 50%. The rest of it is China's [own] technology," Huang said. And because Nvidia has had to sell lower-spec chips to China, the company's average selling price in the country has also gone down. "We lost a lot of revenue," Huang said.
The export controls also did not change the fact that 50% of AI researchers are in China and still doing AI research there, he said, adding that if they are not able to get Nvidia's chips, they will just use the "second-best."
"Local [Chinese] companies are very talented and the export controls gave them the spirit and energy and government support to accelerate their development. I think the export control was a failure," he said.
He added that competition in China is intense and developers there are "incredibly good" at software, with capabilities that can rival those of any country.
"Huawei has the most formidable technology and they are innovating fast," Huang said. "I really do hope the U.S. government will recognize the ban on H20 is not effective and give us a chance to go back [to China]."
The CEO, who is in Taipei for the Computex trade show, said the latest ban on Nvidia's H20 resulted in the company writing off multiple billions of dollars, as much as some smaller chip developers' revenue for a full year. Nvidia said it wrote down $5.5 billion in the first quarter after the U.S. in April cut off its shipments of H20 -- a further downgrade from the H800 -- to China.
Huang said if the U.S. wants to continue to lead, the nation needs to maximize the sale of U.S. chips and infrastructure to other nations, not limit them. The CEO also touted President Donald Trump's decision to overturn the Biden administration's rule to regulate AI chip sales to many Middle East and Southeast Asian nations, where the demand for AI infrastructure is booming.
He said the Chinese computing market is the world's second largest and is estimated to reach some $50 billion next year and the business opportunities for Nvidia are significant.
"DeepSeek works incredibly well everywhere and Qwen [Alibaba's AI model] is excellent. ... They are all built on Nvidia," he said.
"It will be a shame not to be able to enjoy that opportunity, bring home tax revenues to the United States, create jobs and sustain the industry," Huang said, adding that Nvidia will comply with the U.S. regulations and is still trying its best to serve the Chinese market, which is "complicated."
Meanwhile, Huang said it's not "necessary" and not "possible" to build everything in the U.S., but acknowledged ongoing efforts to onshore some critical chip production for national security and supply chain resilience.
"It is impossible to do all the manufacturing, all onshore. It's also unnecessary. But, we should do as much as we can for security, while having resilience and equal redundancy all around the world," he said, adding that the "rebalancing" of manufacturing is happening at a "very good time" to introduce AI technologies into new factories.
Nvidia announced it is planning to build up to $500 billion worth of AI servers in the U.S. over the next four years. The company's major chip production partner, Taiwan Semiconductor Manufacturing Co., is investing $165 billion to ramp up cutting-edge chip production in Arizona, while Nvidia server builders Foxconn, Quanta and others are expanding capacity on American soil.
Nvidia is still the market leader in AI chips. According to Morgan Stanley's estimate, it is the biggest consumer of leading-edge wafers for AI-related semiconductors, with an estimated 66% share in 2025, followed by Google at 22% and AWS with 6% for 2025.
Nvidia also consumes the largest global chunk of high-bandwidth memory (HBM) chips, an estimated 74% of the global total for this year.