NVIDIA Denies Backdoors in Chips Amid Security Concerns

In the early hours of August 6, NVIDIA released a lengthy statement on its Chinese official website and other channels, asserting that its chips do not contain backdoors, kill switches, or monitoring software. The company emphasized that these features are not, and will never be, part of its approach to building trustworthy systems.

NVIDIA's statement addressed concerns raised by some experts and policymakers who have suggested the need for hardware-based "kill switches" or built-in controls to remotely disable GPUs without the user's knowledge or consent. NVIDIA firmly denied the existence of such mechanisms in its GPUs, stating that they do not and should not be implemented.

Recently, NVIDIA's AI chips were reported to have serious security vulnerabilities. Previously, U.S. lawmakers had called for advanced chips exported from the U.S. to include "tracking and location" features. AI experts in the U.S. revealed that NVIDIA's AI chips already possess mature "tracking and location" and "remote shutdown" technologies. To safeguard the network and data security of Chinese users, the Cyberspace Administration of China held a meeting with NVIDIA on July 31, 2025, requiring the company to address and provide evidence regarding the security risks of backdoors and vulnerabilities in the H20 AI chips sold in China.

In response to these concerns, NVIDIA stated, "Network security is of utmost importance to us. NVIDIA's chips do not have 'backdoors' and do not provide any means for remote access or control."

In July, the founder and CEO of NVIDIA visited China to participate in the Chain Expo and other events, marking their third visit to the region this year. During this visit, NVIDIA was reported to be submitting an application to resume sales of the NVIDIA H20 GPU, with assurances from the U.S. government that the necessary permits would be granted, and the company hoped to commence deliveries as soon as possible.

Additionally, the founder and CEO announced the launch of a new, fully compatible NVIDIA RTX PRO GPU, designed as an ideal choice for creating digital twins of AI in smart factories and logistics.

In April of this year, the U.S. government decided to prohibit NVIDIA from selling its H20 chips to the Chinese market. The H20 chip, designed to comply with U.S. export restrictions, is an AI accelerator specifically tailored for the Chinese market. Based on NVIDIA's Hopper architecture and featuring CoWoS advanced packaging technology, the H20 is more suited for vertical model training and inference but cannot meet the demands of trillion-level large model training, with overall performance slightly better than the 910B.

During the visit, the founder and CEO expressed that the Chinese market is vast, dynamic, and highly innovative, serving as a hub for many AI researchers. Therefore, it is crucial for U.S. companies to establish a strong presence in the Chinese market.