32 Prohibited Blackwell Ultra Chips Discovered in Chinese State Telecom

Thirty-two Nvidia Blackwell Ultra B300 graphics processing units have been detected within the internal network of China Telecom, a state-owned telecommunications giant. This discovery, made by the Israeli security firm Lava, highlights a persistent gap in global export controls designed to restrict the flow of high-performance computing hardware to the People's Republic of China.

The presence of these specific chips is significant because they are part of the Blackwell Ultra series, which is currently barred from export to China under U.S. regulations. The B300 is not merely a standard component; it is one of the most powerful data center GPUs on the market, offering substantial performance improvements over its predecessor. Its inclusion in a Chinese state network suggests that the mechanisms intended to keep these advanced processors out of the country are being circumvented.

Lava did not discover these chips through physical inspection of server racks, but rather through a software vulnerability. The security firm used Nvidia’s own DCGM Exporter tool, a utility designed for data center operators to monitor server health and performance. While intended for internal use, this tool can sometimes be accessed from outside the data center's secure network. Compounding the issue, the software does not always require password authentication, allowing external researchers to identify the specific hardware models running behind the scenes.

Nvidia has responded to the findings with a mix of acknowledgment and dismissal. The company did not deny the presence of the chips but characterized the incident as statistically negligible in the broader context of China’s computing infrastructure. An Nvidia spokesperson noted that recent court cases have found diversion rates of less than 0.5% for controlled U.S. products. The spokesperson described this as “a drop in the bucket compared to the ocean of domestic compute China already has,” arguing that with tens of millions of GPUs already in service in approved countries, the impact of smuggling remains a non-factor.

Despite the corporate downplaying, the pattern of diversion is not new. In 2025, reports indicated that Chinese entities had smuggled approximately $1 billion worth of Nvidia AI chips into the country in just three months. At that time, the B200 was the primary target, but distributors were already positioning themselves to move the newer B300 series as soon as it became available. The current discovery confirms that the pipeline for these high-value components remains active and effective.

The mechanics of how these specific B300 units reached China Telecom’s infrastructure remain unclear. Lava has not confirmed the exact location of the chips within the company’s vast network, nor has it detailed the specific route they took. However, the broader context of recent smuggling scandals provides likely parallels. In August 2026, nine individuals in Taiwan were indicted for the illegal export of B300 GPUs to China. This legal action followed an investigation by Supermicro, which fired several employees after uncovering a complex smuggling operation involving a whitelisted server trading firm known as Flying Tiget Tech.

The B300 chip is a highly sought-after asset in the AI industry. Announced in March of last year, the processor was marketed as being “built for the age of reasoning,” with capabilities suited for sophisticated large language models like DeepSeek R1. It offers 50% more memory and FP4 compute capacity than the previous-generation B200. Originally slated to ship by the end of 2025, it quickly became a hot commodity, driving intense demand among data center operators worldwide.

The scarcity and high value of these chips have created a secondary market with dramatic price fluctuations. Demand for Nvidia’s silicon has reached such a fever pitch that even older models are commanding premium prices. The A100 server, which is five years old, has seen its price triple, with some units costing up to $82,000. This economic pressure has led some operators to retool gaming GPUs for AI workloads as a stopgap measure, further illustrating the critical role these chips play in modern computing infrastructure.

While Nvidia maintains that its new tracking tools will eventually help locate such diversions, the company has not yet released this software. Until then, the DCGM Exporter remains a primary means for external parties to audit data center hardware. The findings from Lava serve as a reminder that while export controls form a legal barrier, the physical and digital movement of advanced technology often finds ways to bypass these restrictions, keeping the debate over U.S.-China tech policy alive.