Nvidia just took a $5.5 billion punch to the balance sheet—courtesy of the U.S. government’s latest move to tighten the leash on AI chip exports to China. The company’s most advanced processor available in the Chinese market, the H20, has now fallen under indefinite export restrictions, triggering a 6% slide in Nvidia shares in after-hours trading.
The decision, announced Tuesday, marks a major escalation in the U.S.-China tech standoff and underscores Washington’s growing concern over how AI hardware could fuel China’s supercomputing ambitions. The U.S. Commerce Department has now slapped licensing requirements not only on Nvidia’s H20, but also on AMD’s MI308 and similar chips. AMD shares dropped 7% after the news.
Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram
A Commerce Department spokesperson said the move reflects President Biden’s directive to safeguard U.S. national and economic security. Nvidia, meanwhile, confirmed the charges would cover unsold H20 inventory, outstanding purchase commitments, and related reserves.
A Workaround, Now Blocked
Nvidia had designed the H20 chip specifically to navigate around previous U.S. export limits—delivering toned-down performance but retaining high-speed interconnectivity. That design made the H20 attractive for AI inference tasks, an increasingly dominant segment of the market where models provide real-time answers rather than undergoing initial training.
Despite not being as powerful as Nvidia’s top-tier chips sold outside China, the H20 gained traction with major Chinese tech players including Tencent, Alibaba, and ByteDance. Reuters previously reported that demand surged after startups like DeepSeek ramped up development of low-cost AI models.
But that very design—optimized for high-bandwidth memory access and chip-to-chip connectivity—set off alarm bells in Washington. Analysts argue it still carries supercomputing potential, especially if deployed at scale.
“Likely In Violation”
A Washington, D.C.-based think tank, the Institute for Progress, didn’t mince words. In a statement Tuesday, it claimed that Tencent had already installed H20 chips in a facility likely used to train large AI models—potentially breaching U.S. export restrictions already in place. The group added that DeepSeek’s infrastructure, used for its latest V3 model, might also be in violation.
U.S. restrictions on chips used in supercomputing have been in effect since 2022. Now, the H20 is joining that list. Nvidia said it was formally notified on April 9 that the chip would require an export license—and on April 14, that the restriction would be indefinite. Whether the U.S. will issue any such licenses remains unclear.
A Fork In The Road
This latest move throws a wrench into Nvidia’s China strategy, just as demand in the region for generative AI tools is accelerating. It also highlights the growing friction between global innovation and geopolitical control—a tension Nvidia CEO Jensen Huang must now navigate carefully.
The setback comes one day after Nvidia unveiled plans to invest up to $500 billion into U.S.-based AI server infrastructure, working with partners like TSMC to align with American industrial policy.
Now, as Nvidia absorbs the financial blow and recalibrates, one thing is clear: the AI chip race isn’t just about performance anymore. It’s a front line in the broader battle over who controls the future of intelligent computing.
