The Trump administration has made a surprising about-face on Nvidia’s ability to sell artificial intelligence chips to China, but there’s a strategic catch. Commerce Secretary Howard Lutnick revealed Tuesday that the reversal comes with an important caveat: Nvidia won’t be handing over its cutting-edge technology.
Speaking with CNBC’s Brian Sullivan, Lutnick explained that Nvidia wants to sell China its “fourth best” chip, significantly slower than the top-tier processors used by American companies. “We don’t sell them our best stuff, not our second best stuff, not even our third best,” Lutnick said, outlining the administration’s hierarchy of technology sharing.
US Allows Nvidia to Sell H20 Chips to China, Aiming for “Tech Stack” Addiction
The chipmaker announced Monday night that it would soon resume sales of its H20 chip to China after receiving signals that the Trump administration would grant the necessary export licenses. This marks a major victory for Nvidia, which had previously estimated it could have sold $8 billion worth of H20 chips in the current quarter before sales were halted.
Lutnick revealed that the decision isn’t just about immediate sales—it’s about maintaining long-term technological influence. The administration wants Chinese companies to remain dependent on American technology infrastructure, creating what he called a “tech stack” addiction.
“You want to sell the Chinese enough that their developers get addicted to the American technology stack,” Lutnick explained. “That’s the thinking.” This strategy aims to keep China reliant on American semiconductor technology while ensuring the U.S. maintains a competitive edge.
The Commerce Secretary emphasized that the approach prevents China from developing completely independent alternatives. “The idea is the Chinese are more than capable of building their own,” he said. “You want to keep one step ahead of what they can build, so they keep buying our chips.”
Nvidia’s China Chip: Slower, But Still Powering AI
The H20 chip at the center of this controversy was originally introduced in 2022 as a response to the Biden administration’s export controls. Based on the same underlying technology as Nvidia’s Hopper-generation chips sold in the U.S., the H20 was deliberately modified for the Chinese market.
Nvidia removed several key features to comply with export restrictions, including fewer graphics processing unit cores and lower bandwidth connecting different parts of the chip. Despite these limitations, the recent success of China’s DeepSeek R1 AI model demonstrated that many Chinese companies were perfectly capable of working with these slower processors.
The reversal came after President Donald Trump met with Nvidia CEO Jensen Huang in Washington last week. Huang had previously stated that the administration’s licensing requirements “effectively closed” the Chinese market for the company.
Nvidia Strategies to Balance Sales, Competition, and National Security
Nvidia CEO Jensen Huang has consistently argued that continued chip sales to China serve American interests. He believes that if the U.S. completely cuts off access, Chinese companies will simply invest more heavily in developing their own semiconductor infrastructure, potentially creating stronger competition in the long run.
Huang has also pointed out that the Chinese military wouldn’t use Nvidia chips anyway, and has acknowledged that China’s Huawei represents legitimate competition in the semiconductor space.
Lutnick connected the chip sales decision to broader trade considerations, mentioning that the renewed H20 sales were linked to a rare earths magnet deal. This suggests the administration is taking a more comprehensive approach to U.S.-China trade relationships.
The chip hierarchy Lutnick described places the H20 behind Nvidia’s current top performers: the Blackwell chips, H100, and H200 processors. The company releases new AI chips annually, with serious developers typically seeking the latest versions due to rapid technological improvements.
Currently, the best widely available AI chips are the Blackwell series, including the GB200 chip with paired central processing units, plus B100 and B200 versions. Nvidia is now installing “Blackwell Ultra” chips in data centers, with volume production expected over the next year. The company plans to release “Vera Rubin” chips in 2027.
This strategic approach to technology sharing represents a significant shift in how the U.S. manages its technological relationship with China, balancing economic opportunities with national security concerns while maintaining competitive advantages.
