Smuggled Nvidia B300 Chips Discovered in Major Chinese Telecom Network

For ordinary tech consumers and business operators reliant on advanced artificial intelligence, the supply chain for high-performance computing remains volatile. A recent discovery highlights the persistent difficulty in controlling the flow of powerful graphics processing units from the United States to China, a dynamic that continues to shape availability and pricing for enterprise-grade hardware across the Pacific Northwest and beyond.

Israeli cybersecurity firm Lava announced that it has located 32 Nvidia Blackwell Ultra B300 chips inside the network of China Telecom, a state-owned telecommunications provider. These specific chips are subject to strict U.S. export controls, which generally bar their sale to China. According to a report by The Information, the chips are only legally sold to whitelisted buyers who agree not to export them to sanctioned parties. The presence of these 32 units in a Chinese data center raises questions about how export restrictions are being circumvented in practice.

The incident adds to a growing list of cases where high-performance AI hardware has reached China despite regulatory hurdles. In 2025, reports indicated that Chinese entities smuggled approximately $1 billion worth of Nvidia AI chips into the country over a three-month period. While the B200 chips were the primary target at that time, distributors were already preparing to move the newer B300-series models as soon as they became available. This suggests that the demand for Nvidia’s latest silicon remains robust enough to drive significant illicit trade volumes.

Nvidia has attempted to minimize the impact of the finding, stating that it does not deny the chips’ presence but views the scale as negligible. A spokesperson told Tom's Hardware that the diversion of controlled products has historically amounted to less than half of one percent in recent court cases. The company described this figure as a "drop in the bucket" compared to the vast amount of domestic compute infrastructure China already possesses. Nvidia noted that with tens of millions of its GPUs in service in approved countries, the current smuggling issue remains a non-factor, supported by enforcement actions from governments on both sides of the globe.

The method used to detect the chips offers insight into potential security gaps in data center operations. Lava identified the B300 units using Nvidia’s DCGM Exporter tool, a software utility designed for data center operators to monitor server health and performance. Although intended for internal use, the tool can sometimes be accessed from outside the data center's network and does not always require password authentication. Nvidia has acknowledged the need for better location tracking and is working on a new tool for this purpose, though it has not yet been released. It remains unknown exactly where within China Telecom’s network the chips are housed or the specific route they took to arrive there.

The B300 chip, announced in March of last year, is a top-tier data center GPU designed for advanced reasoning tasks. It offers substantial performance improvements over the previous-generation B200, including 50% more memory and enhanced FP4 compute capabilities. Nvidia originally marketed the chip as "built for the age of reasoning," citing compatibility with sophisticated large language models. The hardware was expected to ship by the end of 2025 and quickly became a high-value commodity in the AI market.

The smuggling of these high-end chips has also triggered legal consequences for intermediaries. In August 2026, nine individuals in Taiwan were indicted for the illegal export of B300 GPUs to China. This action followed an internal investigation at Supermicro, which resulted in the termination of several employees. The scheme involved a Taiwanese server trading firm known as Flying Tiger Tech, which held whitelisted status for purchasing GPUs. This case underscores the complex networks often used to bypass export controls.

The scarcity of new-generation chips has driven up prices for older Nvidia hardware. A100 servers, which are five years old, have seen their prices triple, with some units costing up to $82,000. This price inflation reflects the intense demand for AI processing power. In some cases, operators are even retooling gaming GPUs to handle AI workloads as a stopgap measure. For buyers in the region, this market dynamic means that securing reliable access to high-performance compute resources requires careful navigation of a complex and sometimes opaque supply chain.

As Nvidia continues to refine its tracking tools and governments tighten enforcement, the gap between intended export policy and real-world hardware distribution persists. For now, the discovery of 32 B300 chips in China Telecom’s network serves as a reminder that while regulatory frameworks are in place, the physical movement of critical AI infrastructure remains a challenge for global markets.