Nvidia B300 GPUs Discovered Inside China Telecom Network Despite Export Bans
Israeli security firm Lava identified 32 banned chips in a major Chinese state network, prompting Nvidia to argue the incident does not significantly alter the balance of global AI compute power.
By Nadia Singh ·
For data center operators and tech analysts, the discovery of 32 Nvidia Blackwell Ultra B300 chips inside the infrastructure of China Telecom adds a new layer of scrutiny to how state-owned networks manage their hardware. Although the chips were officially barred from export to China under U.S. regulations, their presence in the network of a major telecommunications provider confirms that restricted hardware continues to find its way into the country despite strict controls. This finding highlights the ongoing difficulty in enforcing export bans on high-performance computing components.
The identification was made by Lava, an Israeli security firm, which detected the B300 units within China Telecom's systems. These chips are subject to strict export controls, meaning they can only be sold to whitelisted buyers who agree not to transfer them to sanctioned parties. The presence of these specific units raises questions about the supply chain, as the B300 is a newer, high-performance model that was expected to ship at the end of 2025. The discovery comes amid a broader trend of smuggling, with reports in 2025 indicating that Chinese companies moved approximately $1 billion worth of Nvidia AI chips into the country in just three months.
Nvidia has responded to the news by downplaying the impact of the finding. A company spokesperson stated that the diversion of controlled products is a drop in the bucket compared to the massive volume of domestic compute resources already available in China. The spokesperson noted that recent court cases have shown insignificant diversion rates of less than 0.5%. With tens of millions of GPUs already in service in approved countries, Nvidia maintains that smuggling remains a non-factor, a position supported by the fact that governments on both sides of the globe are actively working to block these transfers.
The exact method by which the chips entered the network remains unclear. Lava identified the units using Nvidia's DCGM Exporter tool, a software designed for internal monitoring of server health and performance. While this tool is typically used by data center operators, it can sometimes be accessed from outside the internal network and often does not require password authentication. Nvidia has previously acknowledged that it is developing a tool to help locate its chips, but that solution has not yet been released. The lack of precise location data for the 32 chips adds a layer of uncertainty to the incident.
The B300 chip is highly sought after due to its significant performance improvements over the previous generation. Announced in March of last year, the B300 offers 50% more memory and FP4 compute power than the B200. Nvidia marketed the chip as being built for the age of reasoning, specifically citing its suitability for advanced large language models like DeepSeek R1. Its high demand has made it a prime target for smuggling, with nine people in Taiwan indicted in August 2026 for the illegal export of B300 GPUs to China.
The smuggling of these chips often involves complex logistical operations. In the recent Taiwan indictment, a whitelisted server trading firm called Flying Tiget Tech was implicated in the process. Supermicro, another key player, also fired several employees following an internal investigation into smuggling activities. The high stakes involved are reflected in the market price of Nvidia's hardware. Older models like the A100 have seen their prices triple to as much as $82,000, and some operators are even retooling gaming GPUs to handle AI workloads to meet demand.
This incident underscores the persistent challenges in controlling the flow of advanced AI hardware. While Nvidia argues that the volume of smuggled chips is small relative to the total number of units in service, the discovery within a state-owned network serves as a reminder that export controls are not foolproof. As the demand for high-performance computing continues to drive prices and smuggling efforts, the gap between official policy and actual hardware distribution remains a critical area of focus for both tech companies and governments.
