Nvidia B300 Chips Discovered in China Telecom Network
Security firm Lava found 32 banned Blackwell Ultra processors within the state-owned carrier's infrastructure, prompting Nvidia to call the leak insignificant relative to China's domestic compute capacity.
By Amy Swan ·

The ongoing struggle to contain the flow of advanced American semiconductor technology into China has revealed another gap in the export control net. This week, Israeli security firm Lava announced that it has identified 32 Nvidia Blackwell Ultra B300 graphics processing units operating within the network of China Telecom, a state-owned telecommunications provider. The presence of these chips is significant because the B300 series has been barred from export to the country by U.S. regulations, raising immediate questions about how they bypassed restrictions to reach a major Chinese data center.
Under current U.S. export controls, these high-performance AI chips can only be sold to buyers on an approved whitelist who agree not to resell them to sanctioned parties, including those in China. Despite these rules, the leakage of hardware has become a persistent issue. In 2025, reports indicated that Chinese entities smuggled approximately $1 billion worth of Nvidia AI chips into the country in just three months. While the B200 models were the primary target at that time, distributors were already preparing to move B300-series units as soon as they became available, indicating a sustained demand for top-tier silicon.
Nvidia has responded to the findings with a mix of confirmation and minimization. The company did not deny the discovery but characterized the event as a minor anomaly. In a statement, an Nvidia spokesperson noted that recent court cases have shown that the diversion of controlled U.S. products is less than 0.5 percent of the total, describing it as "a drop in the bucket compared to the ocean of domestic compute China already has." The spokesperson further argued that with tens of millions of GPUs already in service in approved countries, the smuggling incident remains a non-factor, largely blocked by governmental efforts on both sides of the Pacific.
The method used to locate the chips offers a glimpse into the vulnerabilities of modern data center monitoring. Lava identified the B300 units using Nvidia’s own DCGM Exporter tool, a software suite designed for data center operators to monitor server health and performance. Although intended for internal use, this tool can sometimes be accessed from outside the data center’s network. Complicating the security picture, the tool does not always require password authentication. While Nvidia has previously confirmed it is developing a specific tool to help locate its chips, that solution has not yet been released, leaving operators reliant on existing monitoring systems.
The motivation for acquiring these specific chips is rooted in their superior performance. Announced in March 2025, the Blackwell Ultra B300 was marketed as being "built for the age of reasoning," with significant capabilities for sophisticated large language models like DeepSeek R1. The chip offers a 50% increase in memory and FP4 compute power compared to the previous-generation B200. Its high performance made it a highly sought-after asset, with initial expectations for shipment by the end of 2025 quickly turning it into a commodity of high value in the global AI market.
The discovery by Lava is the latest chapter in a series of enforcement actions related to illegal exports. In August 2026, nine individuals in Taiwan were indicted for the illegal export of B300 GPUs to China. This legal action followed closely behind an internal investigation at Supermicro, which resulted in the termination of several employees involved in smuggling. The complex logistics of these transfers often involve intermediaries with whitelisted status, such as the Taiwanese server trading firm Flying Tiget Tech, which was implicated in the recent indictments.
The scarcity of new high-end chips has also driven up the cost of older hardware, creating a secondary market with inflated prices. Nvidia’s A100 servers, which are five years old, have seen their prices triple, with some units commanding up to $82,000. As the crackdown on smuggling intensifies, some operators have even resorted to retooling gaming GPUs to handle AI workloads where possible, highlighting the critical role these chips play in the current AI infrastructure landscape.
Source: Tom's Hardware


