$NVIDIA(NVDA.US) Jensen Huang sold the H20 chip with 80% of its computing power cut to China, but Chinese engineers cracked it and found hidden "tracking and positioning" and "remote shutdown" backdoors. This stems from the 2024 U.S. export controls, which banned Nvidia's high-end H100 chip, forcing the company to create a "castrated version" H20 for China, with performance only 15%-30% of the original, specifically for AI inference.

Chinese engineers discovered during disassembly that the H20 firmware layer contains a covert communication module that automatically connects to a Singapore data center upon startup, leaking device location, operational status, and user privacy. More seriously, the chip integrates a remote control interface that can instantly paralyze computing clusters. This design is not accidental; U.S. senators proposed a bill in 2025 requiring AI chips exported to China to be pre-installed with tracking systems.

Hardware-level backdoors pose significant risks. Financial institutions using H20 for risk control models would have their data transmitted to the U.S. in real-time; smart city projects could also experience mass shutdowns. In July 2025, China's Cyberspace Administration urgently summoned Nvidia because its backdoor bypassed traditional protections, directly connecting to U.S. government servers, handing over the "lifeline" of China's critical infrastructure to others.

Nvidia denied the existence of backdoors in the chip, but its explanation failed to convince the public. U.S. AI experts revealed that tracking technology is already mature in GPUs and can be activated by implanting code in drivers. The H20's design logic is similar to Intel's ME engine, which has been proven to remotely access system memory—a classic hardware backdoor.

The H20 is priced high at $13,000 per unit, comparable to the original version, but its performance is far inferior. Supporting software further limits computing power, with actual performance falling below 60% of the claimed value, effectively constituting technological blackmail. The U.S. move is a "retreat to advance," superficially relaxing restrictions on H20 while actually locking China's AI industry into its ecosystem. The H20 is compatible with the CUDA system, making it costly for companies to switch to domestic chips.

However, through reverse engineering, Chinese engineers not only uncovered the backdoor but also mastered the H20's architecture, providing critical references for domestic chip development. Huawei's Ascend 910B is already compatible with most CUDA interfaces, with minimal performance loss.

The H20 incident highlights the battle for computing sovereignty. The U.S. aimed to slow China's AI development but underestimated Chinese engineers' capabilities. Now, domestic chip companies are fully committed to independent R&D, from architecture design to packaging and testing, vowing to break the "stranglehold." After all, only by mastering core technology can one control the future.

The copyright of this article belongs to the original author/organization.

The views expressed herein are solely those of the author and do not reflect the stance of the platform. The content is intended for investment reference purposes only and shall not be considered as investment advice. Please contact us if you have any questions or suggestions regarding the content services provided by the platform.