Nvidia has denied claims from Chinese authorities that its data center GPUs for AI contain hardware capable of remotely disabling the chips, often referred to as a “kill switch.”

The US-based chipmaker responded after the Cyberspace Administration of China requested documentation about potential security vulnerabilities, specifically mentioning “backdoor” risks in the H20 GPU designed for China.

Nvidia’s chief security officer David Reber wrote in a company blog post that its GPUs do not have kill switches or backdoors, and argued that such features would pose significant security risks.

🔗 Source: CNBC

🧠 Food for thought

1️⃣ Hardware backdoor concerns have proven justified in similar cases

China’s scrutiny of Nvidia chips reflects legitimate security concerns that have materialized in other networking hardware.

Cisco routers experienced multiple backdoor incidents over several years, including an undocumented backdoor discovered in 2014 that allowed unauthorized access to small business routers1.

The company had also proposed a “lawful intercept” backdoor in 2004 for law enforcement access, which was later exploited by attackers, demonstrating how intended access points can become security vulnerabilities1.

In 2017, security researchers identified backdoor accounts in Cisco’s network management software that provided root access, showing how these vulnerabilities can persist even in enterprise-grade products1.

These precedents explain why Chinese authorities are conducting due diligence on AI chips that will be integrated into critical infrastructure, as backdoors in networking equipment have repeatedly created serious security risks.

2️⃣ Export controls create a complex compliance burden that extends beyond trade policy

Nvidia’s situation illustrates how U.S. export restrictions force companies to navigate both regulatory compliance and international security scrutiny simultaneously.

The company reported a $5.5 billion financial impact from export restrictions, demonstrating the substantial business costs of these policies2.

The H20 chip itself exists specifically because of U.S. export controls. It is a modified version of Nvidia’s H100 GPU designed to meet export requirements while still serving the Chinese market3.

This creates a challenging dynamic where companies must engineer products that satisfy U.S. regulators’ national security concerns while also convincing international customers that the same modifications don’t create security vulnerabilities.

Recent Nvidia developments