When President Trump announced late last year that Nvidia might sell one of its most powerful chips to China, the deal seemed like a rare win-win in a fraying geopolitical relationship. That would provide a major boost to China’s artificial intelligence ambitions, while handing a victory to America’s leading chipmaker.
Cutting-edge AI systems run on impressive amounts of computing power, and Nvidia’s chips are considered the gold standard around the world. Chinese competitors have yet to build anything that rivals Nvidia’s best, and Mr. Trump’s decision undermined years of U.S. policy aimed at keeping such chips out of China’s reach.
Former officials in the Biden and Trump administrations have warned that the move could squander the lead that U.S. AI companies held over their Chinese rivals, helping China narrow the gap until its own chipmakers can catch up with Nvidia.
But six months later, Beijing has not allowed any of its companies to buy a single one.
This impasse lays bare the depth of distrust between the world’s tech superpowers. For decades, American and Chinese companies worked side by side to create products like the iPhone that shook up industries. But relations have deteriorated over the past decade as both governments have come to view technology as the linchpin of economic supremacy.
Nvidia is stuck in the middle. The chipmaker became the world’s most valuable company by making its semiconductors essential to running AI systems. But Washington and Beijing increasingly view the technology as a matter of national security, especially after observing how it has been used to coordinate attacks in Gaza, Ukraine, Venezuela and Iran.
Rather than turning to Nvidia, Chinese authorities have pushed domestic companies toward local alternatives from chipmakers like Huawei and Cambricon. After meeting last week in Beijing with Xi Jinping, the Chinese leader, Mr. Trump said that China’s lack of interest in the chip he had approved, known as the H200, was due in part to this desire for industrial self-sufficiency.
“They chose not to do it. They want to try to develop their own,” he said.
Chinese AI companies are hungry for more computing power. Kevin Xu, founder of Interconnected Capital, a hedge fund that invests in artificial intelligence technologies, recently spent nine days in China meeting with leading AI start-ups. Each of them, he said, pointed to lack of computing power as the main factor holding them back.
But Chinese companies are now starting to build their AI systems around these constraints rather than waiting for them to relax. And Beijing wants them to continue doing just that.
“It’s a balancing act,” Mr. Xu said. The government knows its companies need more computing power, but it also wants to incentivize domestic chipmakers to improve and improve faster.
Jensen Huang, Nvidia’s chief executive, who visited Beijing last week as part of Mr. Trump’s delegation, said he had not discussed the H200 chip with Chinese officials. But he remains optimistic about the market opening, given China’s huge appetite for AI chips.
“The Chinese government has to decide: How much of its local market does it want to protect and how much of its local market does it want to develop with more AI capability,” Mr. Huang said in an appearance on Bloomberg TV on Monday. “My feeling is that over time the market will open up. »
In March, Xi outlined an ambitious plan to deepen China’s decade-long effort to rely on its own technology. Over the next five years, he wants the country to continue its breakthroughs in areas such as AI, quantum computing and fusion energy.
The quest for self-sufficiency is becoming increasingly feasible as domestic chipmakers rapidly improve their offerings. Huawei, the Chinese tech giant, and Cambricon, a start-up, and others are now producing chips with performance comparable to the H200.
DeepSeek, the startup that symbolizes China’s growing AI capabilities, has also become a sign of the country’s abandonment of Nvidia.
When it released its latest artificial intelligence model last month, the company said for the first time that its new system had been optimized to run on chips made by Chinese tech giant Huawei – a small but significant step in China’s long-standing efforts to develop advanced technologies at home.
Although Washington and Beijing have effectively blocked Nvidia’s most advanced products from accessing the world’s largest chip market, the restrictions have done little to slow the chipmaker’s business. On Wednesday, Nvidia reported quarterly profit of $58.3 billion. The company said it hasn’t seen any sales of its flagship AI chips in China and doesn’t expect that to change anytime soon.
Even though Chinese tech companies now use domestic chips for some tasks, they still rely on Nvidia chips for training, the demanding process of teaching an AI model how to operate. Part of the reason is that Chinese chipmakers have struggled to make advanced chips in sufficient volumes to meet demand.
Many Chinese AI companies use a workaround to avoid purchasing their own Nvidia chips. They rent remote access to chips hosted in data centers operated by other companies, often outside the country. When two AI startups, MiniMax and Zhipu AI, went public in Hong Kong earlier this year, their disclosures showed that the companies were spending several times their revenue to train their models using “cloud services.”
But there are drawbacks to this approach, said Jiang Tianjiao, an associate professor at Fudan University in Shanghai. This means slower processing times and a higher risk of data leaks. This makes Chinese companies dependent on third-party data center operators, which could disrupt them at any time. They also run the risk that the United States will adopt regulations to completely shut down remote access to Nvidia chips, a move championed by some Washington officials.
Despite this, Chinese companies overall spend much less on AI than their American rivals. They are expected to spend $123 billion this year on AI chips and data centers, according to Bernstein Research. In contrast, U.S. tech companies are expected to spend around $1 trillion.
The reduction in spending reflects both ambition and means. Chinese companies have been slower than their American counterparts to bet their future on AI, and they have less revenue to spend on it. That restraint could ease pressure on Beijing to clear the way for Nvidia purchases, said Chris McGuire, a senior fellow at the Council on Foreign Relations who worked on chip export controls under President Joseph R. Biden, Jr.
Beijing may also be reluctant to authorize Nvidia purchases due to national security concerns. Last July, the Internet regulator, the Cyberspace Administration of China, summoned Nvidia to explain the security risks associated with chips it had developed for the Chinese market.
Nvidia said its products do not contain any backdoors that would allow anyone to gain remote access. But Chinese suspicions made it difficult to resume sales.
Mr. Huang has not abandoned China. After a last-minute invitation to last week’s summit, he spent time in Beijing posing for photos with people on the street and sipping noodles on the sidewalk, part of an effort to maintain ties in the hope that Nvidia might one day return.
His shuttles between Washington and Beijing underscore the value he places on the Chinese market and the profits Nvidia could reap there, said Wendy Chang, a senior analyst at the Mercator Institute for China Studies, a think tank. But that eagerness, she added, could encourage Beijing to demand greater concessions.
At Washington’s request, Nvidia has withheld its most powerful chips from China, limiting sales to the United States and its allies. This chip, which Nvidia calls Blackwell, outperforms anything currently available on the Chinese market.
“By playing hard to get, Beijing may be hoping to gain more leverage to access better chips,” Ms Chang said.
Xin Yun Wu contributed reporting from Taipei.
