32 Banned Nvidia Chips Found Inside China Telecom Network
Israeli security firm Lava identified 32 Blackwell Ultra GPUs in the infrastructure of a state-owned Chinese carrier, despite US export bans.
By Noah Ross ·

Thirty-two Nvidia Blackwell Ultra B300 chips have been identified within the network of China Telecom, a state-owned telecommunications provider. The discovery was made by Lava, an Israeli security firm, which located the high-performance graphics processing units despite a ban on their export by the United States. The presence of these specific chips raises significant questions about how they bypassed strict trade controls to end up in a major Chinese data center.
Under current US export regulations, Nvidia’s most advanced data center chips can only be sold to whitelisted buyers. These buyers must agree not to resell the hardware to sanctioned parties, including entities in China. However, the latest finding highlights that the flow of export-controlled technology into China remains a persistent issue. The chips found are part of the Blackwell Ultra series, which are currently restricted from direct export to the country.
Nvidia has responded to the report by downplaying the significance of the finding while refusing to deny that the chips are present. A company spokesperson stated that the diversion of controlled US products has been found by governments, researchers, and media to be insignificant. The spokesperson noted that recent court cases have shown a diversion rate of less than one half of one percent. This figure was described as a drop in the bucket when compared to the vast amount of domestic compute power China already possesses.
The Nvidia representative added that with tens of millions of GPUs already in service in approved countries, the report further demonstrates that smuggling is a non-factor. The company insists that the practice is effectively blocked by governments on both sides of the Pacific. This stance contrasts with previous reports from 2025, which indicated that Chinese companies had smuggled one billion dollars worth of Nvidia AI chips into the country over a three-month period.
The method used to detect the chips provides an interesting technical detail. Lava identified the hardware using Nvidia’s own DCGM Exporter tool. This software is typically installed by data center operators to monitor server health and performance. Although designed for internal use, the tool can often be accessed from outside the data center network. In some instances, it does not even require password authentication. Nvidia has confirmed it is working on a tool to locate its chips but has not yet released it.
The B300 chips are highly sought after due to their performance capabilities. Announced in March of last year, the data center GPU offers significant improvements over its predecessor. It provides 50 percent more memory and FP4 compute power than the previous-generation B200. Nvidia originally marketed the chip as built for the age of reasoning, citing advanced large language models like DeepSeek R1. The chip was expected to ship by the end of 2025 and quickly became a hot commodity.
The difficulty of tracing the specific path the B300 units took into China Telecom’s server racks remains unresolved. Lava has confirmed the chips are present but cannot pinpoint their exact location within the network. The broader context of chip smuggling has led to high-profile legal actions. In August 2026, nine people in Taiwan were indicted for the illegal export of B300 GPUs to China. This followed an investigation by Supermicro, which fired several employees related to smuggling activities. The scheme reportedly involved a Taiwanese server trading firm with whitelisted status.
The high demand for Nvidia silicon has also affected the market for older hardware. Prices for A100 servers, which are five years old, have tripled and can now reach up to 82,000 dollars. The intense crackdown on smuggling has even led some operators to retool gaming GPUs to run AI workloads where possible. This market dynamic underscores the critical role these chips play in modern AI infrastructure and the lengths taken to acquire them.
Source: Tom's Hardware · original report published





