32 Banned Chips Found in Chinese Network

Thirty-two. That is the specific number of Nvidia Blackwell Ultra B300 chips that Israeli security firm Lava claims to have located within the network of China Telecom. This state-owned telecommunications giant operates across China, yet the hardware it now houses has been explicitly barred from export by U.S. regulations. The discovery forces a re-examination of how strictly export controls are being enforced, particularly given that these specific processors are only supposed to be sold to whitelisted buyers who agree not to pass them to sanctioned parties. The presence of these units in a major Chinese data center suggests that the leak is not merely theoretical but operational.

The path to this finding is rooted in a long-standing struggle to keep high-performance silicon out of the hands of specific buyers. In 2025, reports emerged that Chinese firms had smuggled approximately $1 billion worth of Nvidia AI chips into the country in just three months. At that time, the B200 model was the primary target of illicit trade, though distributors were already positioning themselves for the B300 series. Now, with the B300s in circulation, the gap between regulation and reality appears to be widening rather than closing.

Nvidia has responded to the latest findings with a mix of acknowledgment and minimization. The company did not deny the presence of the chips but characterized the issue as negligible in the broader context. An Nvidia spokesperson stated that the diversion of controlled products has been found to be "insignificant," citing recent court cases that showed a rate of less than half of one percent. The spokesperson described this leakage 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 incident proves smuggling remains a non-factor managed by governments on both sides.

The methodology behind the discovery offers a glimpse into the vulnerabilities of modern data center management. Lava identified the chips using Nvidia’s own DCGM Exporter tool, a software utility installed by data center operators to monitor server health and performance. While designed for internal use, this tool is often accessible from outside the data center's network. Furthermore, it does not always require password authentication, a security quirk that allowed Lava to pinpoint the location of the B300s. Nvidia has confirmed it is developing a dedicated tool to help locate its chips, but that solution has not yet been released.

The motivation for acquiring these specific chips is clear when examining their specifications. The Blackwell Ultra B300, announced in March of last year, offers substantial performance gains over its predecessor. It provides 50% more memory and double the FP4 compute power compared to the B200. Nvidia originally marketed the chip as being built for the "age of reasoning," specifically highlighting its compatibility with sophisticated large language models like DeepSeek R1. The B300 was expected to ship by the end of 2025, and it quickly became a highly sought-after asset in the global market.

The illicit movement of these chips has become a criminal enterprise with significant consequences. In August 2026, nine individuals were indicted in Taiwan for the illegal export of B300 GPUs to China. This legal action followed closely on the heels of Supermicro firing several employees after an internal investigation into smuggling practices. The complex supply chain involved a Taiwanese server trading firm called Flying Tiget Tech, which held whitelisted status for purchasing GPUs, allowing the hardware to move through legitimate channels before reaching its final destination in China.

The scarcity of these advanced chips has triggered a ripple effect throughout the hardware market. Demand for Nvidia’s older silicon has surged, causing prices to skyrocket. The A100 server, a product that is now five years old, has seen its price triple and can now cost up to $82,000. This price inflation underscores the desperation for compute power in the AI sector. In some cases, operators have resorted to retooling gaming GPUs to handle AI workloads, a stopgap measure that highlights the imbalance between supply and demand.

For readers in California, a hub for both tech innovation and AI development, this story serves as a reminder that the hardware powering the industry is subject to complex geopolitical constraints. The ability of a security firm to identify specific banned chips in a foreign state-owned network suggests that the digital footprint of these machines is harder to erase than previously thought. As Nvidia continues to refine its tracking tools and governments tighten their nets, the question remains whether the current export control framework is robust enough to keep pace with the ingenuity of those looking to acquire the world's most powerful computing resources.

What happens next depends on whether Nvidia’s forthcoming tracking tool can be deployed quickly enough to prevent further leaks. Until then, the 32 chips in China Telecom’s racks stand as a tangible symbol of the ongoing challenge. The industry must now decide if a "drop in the bucket" is truly an acceptable loss, or if it represents a crack in the foundation of the current export control strategy that needs immediate repair.