Smuggled Nvidia Blackwell Ultra Chips Detected in China Telecom Network

Lava, an Israeli cybersecurity company, has reported the discovery of 32 Nvidia Blackwell Ultra B300 graphics processing units within the infrastructure of China Telecom, a state-owned telecommunications giant. The presence of these specific chips is significant because U.S. export controls have already barred their sale to China, raising questions about how they bypassed regulatory restrictions to reach a domestic data center.

According to a report by The Information, these B300 chips are subject to strict export rules that limit sales to whitelisted buyers who agree not to resell to sanctioned parties, including those in China. Despite these controls, the detection of the hardware highlights the persistent difficulty in preventing high-performance AI silicon from reaching Chinese operators. In 2025, it was estimated that Chinese firms smuggled $1 billion worth of Nvidia AI chips into the country over a three-month period, with B200 models being particularly sought after before the B300 series became available.

Nvidia has responded to the findings by downplaying the impact of the discovery while acknowledging the presence of the chips. A spokesperson stated that recent court cases have shown that the diversion of controlled U.S. products amounts to less than half of one percent, describing it as "a drop in the bucket" compared to the massive amount of domestic computing power China already possesses. The company emphasized that with tens of millions of GPUs in service in approved countries, smuggling remains a non-factor that is actively blocked by governments on both sides of the globe.

The exact method by which the B300 chips entered China Telecom’s server racks remains unknown, and Lava has not confirmed their specific location within the network. However, the security firm identified the hardware using Nvidia’s own DCGM Exporter tool, a software utility designed for data center operators to monitor server health and performance. While intended for internal use, this tool can often be accessed from outside the data center’s network and may not always require password authentication, a vulnerability that facilitated the detection.

The Blackwell Ultra B300, announced in March 2025, represents one of the most powerful data center GPUs currently available. It offers substantial performance improvements over the previous-generation B200, including 50% more memory and FP4 compute capability. Nvidia originally marketed the chip as being "built for the age of reasoning," citing its compatibility with advanced large language models. Although initially expected to ship by the end of 2025, the chip quickly became a target for illicit trade. In August 2026, nine individuals in Taiwan were indicted for the illegal export of B300 GPUs to China, a move that followed an internal investigation at Supermicro that resulted in several employee firings.

The high demand for Nvidia’s silicon has driven up prices for older models as well. A100 servers, which are five years old, have seen their prices triple, with units now costing up to $82,000. Some operators have even resorted to retooling gaming GPUs to handle AI workloads in an effort to circumvent supply constraints. While Nvidia continues to develop a dedicated tool to track its chips, it has not yet been released, leaving security firms like Lava to rely on existing monitoring utilities to uncover illicit hardware.

Despite the ongoing challenges with smuggling, Nvidia maintains that the volume of diverted chips is minimal relative to the total number of units deployed globally. The company’s stance suggests that while export controls are not perfect, the overall impact on the balance of AI computing power remains limited. As the race for AI dominance intensifies, the ability to track and secure high-value hardware will likely remain a critical focus for both manufacturers and national governments.