行业新闻 美盛纪事
Nvidia To Resume H20 Shipments To China

Following the Trump administration’s decision to revoke unpopular export control rules, Nvidia is to resume AI GPU shipments to China, specifically, a China-specific Hopper-generation SKU, the H20. Shipments had been restricted while the US government reworks earlier rules put in place by the Biden administration, which were widely seen in the industry as unnecessarily strict.

Nvidia indicated that AI chips, such as its GPUs, will still require a license from the Bureau of Industry and Security (BIS) for shipments to China, but the company stated that it has assurances from the government that these licenses will be granted.

A blog on the company’s website said that Nvidia is filing applications to sell the H20 to Chinese customers again. “The U.S. government has assured Nvidia that licenses will be granted, and Nvidia hopes to start deliveries soon,” the blog said.

Nvidia CEO Jensen Huang visited Washington D.C., last week and is in Beijing this week, where he has already discussed “how researchers worldwide can advance safe and secure AI for the benefit of all,” with Chinese government and industry officials, according to the blog.

The Chinese market represents potentially billions of dollars in revenues for Nvidia. Huang told the Stratechery podcast in May that the company had had to walk away from $15 billion in sales as a result of rule changes. Nvidia had also written off $5.5 billion in H20 inventory, which is presumably now back in play.

Jensen Huang (Source: Nvidia)

“Anybody who thought that one chess move to somehow ban China from H20s would somehow cut off their ability to do AI is deeply uninformed,” Huang told Stratechery.

AI Diffusion

GPU (and all AI chip) shipments to China had been put on hold since the Trump administration cancelled a planned update to the AI Diffusion rules, implemented by the previous administration to tighten shipments to countries where onward distribution to China was thought to be a possibility.

The Trump administration reportedly indicated at the time that it would replace the rules with something simpler. It remains unclear what rules are in place and what the rules will be going forward.

This change in tactics followed a visit by Huang to the White House in May, during which Huang said Nvidia would manufacture at least some of its forthcoming generation of AI hardware systems in the United States. This move was seen as a way of hedging against potential tariffs on chips manufactured outside the U.S.

The H20 is a significantly cut-down version of Nvidia’s H100, its flagship previous-generation GPU. The H20 was designed to comply with the AI Diffusion rules in effect at the time, which limited the performance and bandwidth of any AI accelerator chips shipped to customers in certain countries, including China. Restarting shipments may signal that the Trump administration’s rules have the same, or looser, parameters than the Biden administration’s version.

Points put forward by Huang at GTC, and by GPU competitor AMD’s CEO Lisa Su at a recent Senate hearing, indicate that the US semiconductor giants are worried that, without access to US chips, Chinese chip competitors would surge forward.  

“The United States leads today, but what I would like to say is: it is a race,” Su told the Senate hearing. “Leadership is absolutely not guaranteed. It’s a global race that will shape the outcome of national security and economic prosperity for many decades to come.”

While the industry understands the importance of national security, Su said, export controls need careful consideration.

“We lead today because we have the best technology,” she said. “However, if we’re not able to fully have our technology adopted in the rest of the world, there will be other technologies that will come to play. They may not be as good as we are today, but frankly, usage really spurs innovation.”

At GTC, Huang praised Chinese engineering talent, noting that 50% of the world’s AI researchers are originally from China, “the single largest population [of AI researchers] by far,” as he put it. “So it stands to reason that there will be a great contribution of AI research coming from China, simply because every single AI lab in America has many excellent Chinese researchers, without exception.”

“However China is producing so many fine computer scientists, however you’re doing that, please keep doing it,” he added.

From TTimes

返回概述
清微智能成功融资20亿元,瞄准非GPU架构AI芯片市场
清华系AI芯片企业北京清微智能科技有限公司于2025年12月2日宣布成功完成超过20亿元人民币的C轮融资。本轮融资由北京市属国企京能集团领投,参与的投资方还包括北创投、京...
更多信息
日本Rapidus计划2029年生产1.4纳米芯片
日本半导体制造商 Rapidus 近日宣布,计划于 2027 财年在北海道建设第二座晶圆厂,预计最快 2029 年开始生产先进的 1.4 纳米芯片。此举旨在缩小与全球芯片制造巨头台积电的...
更多信息
长鑫存储发布DDR5内存新品 最高速率达8000Mbps
11月23日,长鑫存储正式发布其最新DDR5产品系列:最高速率达8000Mbps,最高颗粒容量24Gb,并同步推出覆盖服务器、工作站及个人电脑全领域的七大模组产品。长鑫存储现场...
更多信息
0.1065s