The US authorizes the export of Nvidia’s H200 AI chip to China and levies a 25% fee
The US has approved the export of Nvidia’s H200 chip to China, imposing a 25% fee, which has raised security concerns even as it garners support from the industry.
President Donald Trump announced Monday that the United States will permit the export of Nvidia’s H200 processors, the company’s second-most advanced artificial intelligence chips, to China, while imposing a 25% fee on the sales.
The decision concludes extensive discussions in Washington regarding whether American chip manufacturers should maintain their global dominance by continuing sales to China or impose export restrictions to curb Beijing’s military and technological advancements. The impact of the decision on new demand is still uncertain, particularly since China has recommended that domestic companies steer clear of US technology.
Nvidia’s shares increased by 2% in after-hours trading after Trump’s statement on Truth Social, having already risen by 3% earlier in the day due to a Semafor report. Trump stated that he communicated the decision to Chinese President Xi Jinping and received a response that was described as “positive.”
The US Commerce Department is in the process of finalizing the conditions for exports, which will also be applicable to other chipmakers such as AMD and Intel. A White House official has confirmed that a 25% fee will be imposed as an import tax on chips entering the US from Taiwan, intended for security screening prior to their re-export to China.
Trump highlighted that the agreement will “protect national security, create American jobs, and maintain America’s leadership in AI,” pointing out that Nvidia’s latest Blackwell and forthcoming Rubin chips are excluded from the arrangement.
Administration officials view the decision as a balanced approach: withholding US chips from China could enhance Huawei’s domestic AI chip goals, while the option of supplying the most advanced Blackwell chips has been dismissed. Nvidia expressed appreciation for the decision, describing it as a “thoughtful balance.”
Critics in Washington, however, caution that the chips may bolster China’s military capabilities. A number of Democratic senators characterized the approval as a “colossal national security failure,” whereas Republican Congressman John Moolenaar expressed concerns that China would “rip off” Nvidia’s technology and mass-produce it.
A report from the Institute for Progress indicates that Nvidia’s H200 chip boasts nearly six times the power of the H20, which is the most advanced AI processor currently permitted for export to China. Nvidia’s Blackwell chips, currently utilized by leading AI laboratories in the US, offer enhanced speed.
Even with the approval, China has been ramping up efforts to reduce dependence on Nvidia. Regulators have recently accused Nvidia of potential security backdoors in its downgraded H20 chips, a claim that the company denies. They have also cautioned local firms against depending on US semiconductors.
“Chinese firms desire H200s, yet the Chinese state is motivated by a sense of paranoia and pride,” stated Craig Singleton of the Foundation for Defense of Democracies. “Washington might grant approval for the chips, yet Beijing still needs to permit their entry.”
The announcement coincided with the US Justice Department’s disclosure of having dismantled a smuggling ring linked to China, which was accused of unlawfully exporting at least $160 million worth of Nvidia’s restricted H100 and H200 chips during late 2024 and early 2025.
According to experts, Chinese companies might continue to purchase H200s due to their superior performance compared to any domestically produced chips. China’s local AI chipmakers consist of Huawei, Cambricon, and Moore Threads; however, none currently reach the performance levels of Nvidia.
Chinese semiconductor indices experienced a decline at the market’s opening on Tuesday but subsequently recovered much of their losses as investor uncertainty regarding Beijing’s position on the US approval persisted.
