Nvidia Blackwell Ultra B300 GPUs Detected in China Telecom's Infrastructure

Thirty-two Nvidia Blackwell Ultra B300 processors have been identified within the network of China Telecom, a state-owned telecommunications provider. The discovery by Israeli security firm Lava highlights a persistent challenge in global technology trade: the movement of high-performance chips that are subject to strict U.S. export controls. These specific units are designed for data center use and are currently barred from direct export to China, making their presence in a domestic Chinese network a notable anomaly that continues to fuel debate over the effectiveness of current regulatory frameworks.

The presence of these chips raises significant questions regarding how they bypassed established trade restrictions. Under current U.S. rules, such hardware can only be sold to whitelisted buyers who agree not to transfer the technology to sanctioned parties. The fact that these specific Blackwell Ultra units, which were recently restricted, have appeared in a major Chinese data center suggests that loopholes or secondary markets remain active. This finding adds to a growing body of evidence indicating that while official channels are closed, informal routes for acquiring advanced compute power continue to operate.

Nvidia has responded to the report by downplaying the scale of the issue, though the company did not dispute Lava’s technical findings. A spokesperson for the chipmaker stated that the diversion of controlled products represents a tiny fraction of the total market. The company noted that in recent court cases, the diverted amount was less than half of one percent of the total volume. Nvidia argued that compared to the vast amount of domestic compute power China already possesses, this specific instance is a "drop in the bucket." The firm maintained that with tens of millions of GPUs already in service in approved countries, smuggling remains a non-factor that is effectively managed by governments on both sides of the Pacific.

The method of detection offers a glimpse into the technical vulnerabilities of data center operations. Lava identified the chips using Nvidia’s DCGM Exporter tool, a software utility designed for monitoring server health and performance. While primarily intended for internal use, this tool can sometimes be accessed from outside the data center’s network. Compounding the issue, the tool does not always require password authentication, creating a potential window for external observation. Nvidia has acknowledged the utility of such tracking methods and is reportedly working on a dedicated tool to locate its chips, although this solution has not yet been released to the public.

The demand for the Blackwell Ultra B300 is driven by its substantial performance upgrades over its predecessor. Announced in March of the previous year, the chip offers 50% more memory and FP4 compute capability than the B200. Nvidia positioned the hardware as "built for the age of reasoning," specifically citing its suitability for advanced large language models. The chip was originally expected to ship by the end of 2025, but it quickly became a high-value target in the global market. Its scarcity and power have made it a focal point for entities seeking to enhance their AI capabilities without waiting for formal export approvals.

Recent legal actions underscore the complexity of the smuggling networks involved. In August of this year, nine individuals in Taiwan were indicted for the illegal export of B300 GPUs to China. This prosecution followed an investigation by Supermicro, which resulted in the termination of several employees. The case highlighted the use of intermediaries, including a server trading firm with whitelisted status, to facilitate the transfer of hardware. These incidents illustrate that the flow of chips often involves a multi-step process that leverages legitimate business structures to move restricted goods.

The economic impact of these restrictions is visible in the pricing of older hardware. As demand for new, high-performance chips remains high, the value of previous-generation silicon has surged. Nvidia’s A100 servers, which are five years old, have seen their prices triple, with some units commanding up to $82,000. This price inflation reflects the intense competition for available compute resources. In some cases, operators have even resorted to retooling gaming GPUs to handle AI workloads, demonstrating the extent to which data center operators are willing to adapt their hardware strategies to meet demand.

While Nvidia maintains that the volume of smuggled chips is statistically insignificant, the detection of 32 B300 units in a state-owned network serves as a concrete example of the ongoing tension between export controls and market demand. The incident does not necessarily indicate a systemic failure of the regulatory system, but it does confirm that the flow of advanced technology is not fully contained. As China continues to build out its domestic AI infrastructure, the balance between restriction and availability remains a dynamic and closely watched aspect of the global tech landscape.

The situation leaves several open questions about the long-term efficacy of export controls. If even a small number of chips can be traced to specific networks, it suggests that the monitoring tools available to both buyers and regulators are becoming more sophisticated. The upcoming release of Nvidia’s dedicated tracking tool may provide additional clarity on the location and usage of its hardware. For now, the discovery in China Telecom’s network stands as a reminder that the global trade of AI compute power remains a complex and evolving puzzle.