Thirty-Two Blackwell Ultra Chips Exposed in China Telecom Network

Thirty-two high-performance Nvidia Blackwell Ultra B300 graphics processing units have been detected within the infrastructure of China Telecom, a state-owned telecommunications giant, marking a significant breach in U.S. export control protocols. The discovery highlights the persistent difficulty in preventing advanced American silicon from entering the Chinese market, despite explicit restrictions designed to limit the proliferation of AI computing power to sanctioned entities. This incident adds to a growing list of evidence that suggests existing export controls may be insufficient to fully contain the flow of critical hardware.

The specific chips in question were barred from export by U.S. regulations, which permit sales only to whitelisted buyers who agree not to transfer the hardware to restricted parties. Lava, an Israeli cybersecurity firm, identified the thirty-two units using Nvidia’s own Data Center GPU Manager Exporter tool, a software utility designed for monitoring server health and performance. Although this tool is intended for internal use by data center operators, it can sometimes be accessed from outside the local network and often lacks strict password authentication, a vulnerability that allowed for the identification of the hardware. It remains unclear exactly where within China Telecom’s vast network these servers are located or the specific mechanism by which they arrived there.

The involvement of key industry players underscores the complexity of the smuggling routes. In August of this year, nine individuals in Taiwan were indicted for the illegal export of B300 GPUs to China, a case that followed an internal investigation at Supermicro that resulted in several employee terminations. The operation reportedly utilized Flying Tiger Tech, a Taiwanese server trading firm with whitelisted status for GPU purchases, to facilitate the transfer. This incident follows a broader trend identified in 2025, when reports indicated that Chinese entities smuggled approximately one billion dollars worth of Nvidia AI chips into the country over a three-month period, with distributors actively preparing to move B300-series units as soon as they became available.

Nvidia has responded to the findings by characterizing the incident as statistically minor relative to the broader market. A spokesperson stated that recent court cases have shown an insignificant diversion of controlled U.S. products, estimating the loss at less than half of one percent. The company argued that with tens of millions of GPUs already in service in approved countries, the smuggling represents a negligible factor compared to the vast domestic compute capabilities China already possesses. Despite this downplaying, Nvidia did not deny the presence of the chips, noting that it is currently developing a dedicated tool to help locate its hardware, though this utility has not yet been released.

The stakes for the B300 are high, given its role in the current AI race. Announced in March of last year, the chip offers a fifty percent increase in memory and FP4 compute capabilities over its predecessor, the B200, positioning it as a key component for next-generation large language models. The intense demand for such hardware has driven up prices across the board, with five-year-old A100 servers now tripling in value to reach up to eighty-two thousand dollars. As governments on both sides of the globe continue to tighten their grip on chip movements, the discovery within China Telecom serves as a reminder that the battle over AI supremacy is being fought not just in boardrooms, but in the server racks of state-owned enterprises.

The exposure of these chips reinforces the notion that while export controls create friction, they do not create an impenetrable barrier. For analysts and policymakers in the Southeast and beyond, the case illustrates the ongoing tension between commercial interests and geopolitical strategy. As Nvidia continues to push its "built for the age of reasoning" narrative, the presence of its most powerful silicon in a restricted network suggests that the demand for AI compute remains a driving force capable of bending regulatory frameworks.