Nvidia Blackwell Ultra B300: State-Owned Telecom Found in Possession of Restricted Chips

Nvidia has reacted to the identification of 32 Blackwell Ultra B300 graphics processing units within the infrastructure of China Telecom, a state-owned telecommunications provider. The Israeli security firm Lava disclosed that it detected these specific chips, which are currently subject to U.S. export controls, residing inside the Chinese carrier's network. This discovery highlights the continuing challenge of enforcing restrictions on high-performance AI hardware destined for the region.

The B300 series was barred from export to China, with sales limited to whitelisted buyers who agreed not to transfer the technology to sanctioned parties. The presence of these units in a data center operated by a major state entity raises questions about the efficacy of current oversight mechanisms. Previous reports indicated that Chinese entities smuggled approximately $1 billion worth of Nvidia AI chips into the country in a single quarter in 2025, suggesting that the flow of restricted silicon remains a significant factor in the AI market.

In response to the report, Nvidia downplayed the significance of the find, noting that while it did not deny the findings, the volume involved is statistically small. An Nvidia spokesperson stated that recent court cases have shown diversion of controlled products to be less than half of one percent, describing it as "a drop in the bucket" relative to the existing domestic compute capacity in China. The company argued that with tens of millions of GPUs already in service in approved countries, such instances of smuggling do not materially impact the overall supply chain.

The method used to locate the chips provides insight into potential security vulnerabilities in data center monitoring. Lava identified the B300 units using Nvidia’s DCGM Exporter tool, a software utility designed for internal server health monitoring. This tool can sometimes be accessed from outside the data center network and does not always require password authentication. While Nvidia has acknowledged it is developing a dedicated tool to track its chips, that solution has yet to be released, leaving operators to rely on existing monitoring software.

The demand for the Blackwell Ultra B300 is driven by its substantial performance improvements over the previous generation. Announced in March of the previous year, the chip offers 50% more memory and FP4 compute capability compared to the B200. Nvidia marketed the B300 as "built for the age of reasoning," citing its suitability for advanced large language models. The chip became a target for smuggling shortly after its release, with recent legal actions in Taiwan indicting nine individuals in August 2026 for illegal exports, and Supermicro dismissing employees following an internal investigation into the practice.

The scarcity of Nvidia’s latest silicon has driven up prices for older models as well. Servers equipped with the A100, a chip released five years ago, have seen prices triple, reaching as much as $82,000. This price surge reflects the critical role these chips play in AI data centers, leading some operators to retool gaming GPUs for AI workloads where possible. The situation underscores the tight balance between supply constraints and the rapid expansion of AI infrastructure globally.

Open questions remain regarding the specific pathway the B300 chips took to enter China Telecom’s racks. While Lava confirmed their presence, it could not pinpoint their exact location within the network. The incident serves as a reminder that while export controls are a primary tool for managing technology flow, they are not an absolute barrier, and the market for high-end AI compute continues to evolve in response to these pressures.