Two Chinese nationals are on trial in California for allegedly exporting hundreds of Nvidia H100 AI chips to China without proper licenses, as the company pushes back against political calls for hardware kill switches.
According to the US Department of Justice, Chuan Geng and Shiwei Yang are accused of sending millions of dollars' worth of Nvidia AI chips to China between October 2022 and July 2025, bypassing US export controls. The chips are considered critical for AI infrastructure and are commonly used to train large language models. Some shipments reportedly passed through transit routes in Singapore and Malaysia.
Reuters reports that Geng and Yang operated a company called ALX Solutions, which was founded shortly after the US imposed new export restrictions in 2022. Over 20 shipments were officially sent to freight forwarders in Southeast Asia, but payments came directly from companies in China and Hong Kong, suggesting a deliberate effort to conceal the chips' final destination. In one case, a shipment worth $28.4 million was listed as destined for Singapore, but US officials could not locate the supposed recipient.
Nvidia said in a statement that it only works with known partners and all sales comply with US export laws. Diverted products receive no support, no updates, and no services, a spokesperson said. Super Micro Computer, the server maker that reportedly supplied ALX with more than 200 H100 chips, stated it strictly follows export regulations and is cooperating with authorities.
Nvidia rejects kill switches for AI chips
As legal pressure mounts, politicians are calling for AI hardware to include "kill switches" or tracking features to prevent misuse. Nvidia has come out firmly against these proposals in a new blog post.
David Reber Jr., Nvidia's Chief Security Officer, argues that adding such features would create serious security risks. Kill switches and tracking functions could become targets for attackers and introduce major vulnerabilities. Instead, Nvidia advocates for established security strategies like "defense in depth" - layering protections without relying on centralized control. The company notes that its chips are used in sensitive fields like healthcare, aviation, and autonomous vehicles, where a centrally controlled kill switch would be especially risky.
Reber points to the US government's failed Clipper chip initiative in the 1990s as a cautionary tale. The project was abandoned due to critical security flaws after attempting to add government access to encrypted communications.
Nvidia also dismisses comparisons to smartphone features like "Find my phone," stressing that these are optional software tools, not permanent hardware controls.
The company says it relies on proven security concepts like "defense in depth" and transparent software tools for monitoring, diagnostics, and patch management - always under user control. Nvidia categorically rejects any form of hardware backdoor.
