Nvidia wants to put generative AI within reach of every data center, but selling its most capable chips in China has become a complicated part of that ambition. U.S. export restrictions block sales of two key processors there, leaving Nvidia to offer a slower compliant chip while Chinese firms weigh their options.
Advanced chips are central to the dispute
During a two-hour keynote at Computex in Taipei, Nvidia co-founder and CEO Jensen Huang outlined a broad push into AI computing. His announcements included chip release dates, the DGX GH200 supercomputer and partnerships with major companies. The speech was his first public address in almost four years, according to the source article.
The market context is significant: Nvidia’s stock had surged over the prior year, and the company’s valuation was about $960 billion at the time of the article. Generative AI depends on substantial computing power and data, making access to advanced processors an important issue for companies building large language models.
The U.S. government restricted Nvidia from selling its A100 and H100 graphic processing units to China the previous year. The source says both chips are used to train large language models such as OpenAI’s GPT-4. The H100, based on Nvidia’s Hopper GPU computing architecture and equipped with a built-in Transformer Engine, was seeing particularly strong demand.
Restrictions change what Nvidia can sell
The performance gap helps explain why these processors matter. Compared with the A100, the H100 can deliver 9x faster AI training and up to 30x faster AI inference on large language models, according to the source. Those capabilities make the H100 attractive for both training models and generating responses from them.
With the A100 and H100 restricted, Nvidia turned to a slower chip for China that complies with U.S. export control rules. That gives the company a way to continue serving some demand, but it does not provide the same performance as the products that cannot be sold there.
The limits also affect a market Nvidia cannot easily ignore. The source estimates that the chip export ban would have cost Nvidia $400 million in potential sales in the third quarter of last year alone. That figure illustrates the near-term commercial stakes, while the availability of a compliant alternative shows how export rules can shape product choices.
China’s response could reshape the market
Chinese AI firms have reason to watch what Nvidia can offer, since generative AI requires more computing power and data than previous generations of AI. At the same time, continued restrictions could push China to seek more robust alternatives of its own.
The source frames the restrictions as a reminder for China to pursue self-reliance in key technology sectors. That is a longer-term consequence rather than an immediate replacement for Nvidia’s chips: the article does not identify a particular Chinese alternative or say when one might become available.
Huang made the possibility explicit in an interview with the Financial Times: “If [China] can’t buy from … the United States, they’ll just build it themselves. So the US has to be careful. China is a very important market for the technology industry.” His comment links the sales question to a wider strategic concern: restricting access can affect both current business and incentives to develop competing technology.
A market caught between demand and limits
Nvidia’s China position therefore sits between two forces. Demand for powerful AI computing makes the market commercially important, while export controls determine which chips the company can supply. Selling a slower, compliant processor preserves some access, but the restrictions could also make homegrown alternatives more attractive over time.
The keynote’s announcements placed Nvidia’s data center ambitions in the spotlight, yet the China issue shows that expanding AI infrastructure is not only a matter of launching faster hardware. Where companies can buy that hardware, and how they respond when they cannot, may shape the competitive landscape as generative AI develops.