32 Blackwell Ultra Chips Found in China Telecom Network
An Israeli security firm has located thirty-two banned Nvidia processors within the infrastructure of a major Chinese state telecom carrier.
By Ji-eun Choi ·
The flow of high-performance computing hardware into China continues despite strict regulatory barriers. This week, new findings revealed that a significant number of restricted Nvidia chips are currently operating within the network of China Telecom, a state-owned telecommunications provider.
Lava, an Israeli cybersecurity company, announced that it identified thirty-two Nvidia Blackwell Ultra B300 GPUs inside the systems of China Telecom. These specific processors are subject to U.S. export controls, which prohibit their sale to certain parties in China. The discovery raises questions about how the chips entered a data center that should have been excluded from the approved buyer lists.
The presence of these components is not an isolated incident in the broader context of chip smuggling. In 2025, reports indicated that Chinese entities imported approximately one billion dollars worth of Nvidia AI chips over the course of three months. While the B200 model was the primary target during that period, distributors were already preparing for the arrival of the more advanced B300 series. The B300 was announced in March of last year as a next-generation data center GPU. It offers fifty percent more memory and FP4 compute capability compared to its predecessor, making it a highly sought-after asset for advanced AI workloads.
Nvidia has responded to the report with a measured tone, choosing to minimize the impact of the discovery rather than deny the findings. A company spokesperson stated that the diversion of controlled U.S. products has historically been insignificant. They noted that recent court cases have shown such instances to be less than half of one percent of total volume. The spokesperson described this leakage as a drop in the bucket compared to the vast amount of domestic compute power China already possesses. With tens of millions of GPUs in service across approved countries, Nvidia maintains that smuggling remains a non-factor, effectively blocked by governments on both sides.
The method by which Lava discovered the chips highlights a potential vulnerability in data center monitoring. The firm used Nvidia’s own DCGM Exporter tool to identify the hardware. This software is designed for internal use by data center operators to monitor server health and performance. However, it can often be accessed from outside the local network and does not always require password authentication. It remains unclear exactly where within China Telecom’s network the chips are located, nor how they physically arrived at the site. Nvidia has confirmed it is developing a tool to help locate its chips, but that solution has not yet been released.
The enforcement landscape has tightened in recent months. In August of this year, nine individuals were indicted in Taiwan for the illegal export of B300 GPUs to China. This legal action occurred shortly after Supermicro confirmed it had fired several employees following an internal investigation into smuggling practices. The complex smuggling routes involved a Taiwanese server trading firm known as Flying Tiget Tech, which held whitelisted status for GPU purchases. This case underscores the difficulty of tracking hardware once it leaves the hands of approved distributors.
The demand for Nvidia silicon has created a secondary market where even older hardware commands premium prices. A100 servers, which are five years old, have seen their prices triple. They can now cost up to eighty-two thousand dollars. In some cases, operators have resorted to retooling gaming GPUs to handle AI workloads when newer chips are unavailable. The continued presence of the B300 in China suggests that the appetite for top-tier computing power remains strong despite the logistical and legal hurdles.
As the situation unfolds, the focus remains on how effectively export controls can be enforced in a global market. The identification of these thirty-two chips serves as a reminder that the gap between policy and practice can be difficult to close. For now, the chips remain active in the network of a major Chinese state entity, continuing their role in the country’s expanding AI infrastructure.
Source: Tom's Hardware
