China Telecom Network Found Hosting 32 Banned Nvidia B300 Chips

Despite strict U.S. export controls designed to limit access to advanced AI hardware, illegal transfers of high-performance graphics processing units continue to reach Chinese data centers. On October 23, 2026, new findings revealed that 32 Nvidia Blackwell Ultra B300 chips are operating within the network of China Telecom, a state-owned telecommunications provider.

The discovery was made by Lava, an Israeli cybersecurity firm, which identified the specific hardware inside the Chinese company’s infrastructure. These B300 chips are currently barred from export to China under U.S. regulations, which generally restrict sales to whitelisted buyers who agree not to transfer the technology to sanctioned parties. The presence of these specific units raises questions about the effectiveness of the current enforcement mechanisms surrounding high-value semiconductor shipments.

This incident follows a broader pattern of smuggling that has plagued the industry. In 2025, reports indicated that Chinese entities had smuggled approximately $1 billion worth of Nvidia AI chips into the country over the course of three months. At that time, the B200 model was the primary target, but distributors were already positioning themselves to move B300-series units as soon as they became available. The continued flow of these chips suggests that the barrier of export controls remains difficult to enforce completely.

Nvidia has responded to the findings with a mix of acknowledgment and minimization. While the company did not dispute Lava’s detection, a spokesperson argued that the volume of diverted chips is negligible compared to the total number of units in service. The spokesperson stated that recent court cases have found insignificant diversion of controlled U.S. products, citing figures less than one-half of one percent. Nvidia described this as a drop in the bucket relative to the massive volume of domestic compute resources China already possesses, maintaining that smuggling remains a non-factor given the tens of millions of GPUs operating in approved countries.

The method used to locate the chips highlights a specific vulnerability in data center operations. Lava identified the B300 units using Nvidia’s DCGM Exporter tool, a software utility designed for monitoring server health and performance. Although intended for internal use by data center operators, the tool can often be accessed from outside the local network and does not always require password authentication. It is not yet known exactly where within China Telecom’s network the servers are located, nor how they were initially installed, but their presence has been confirmed through this monitoring software. Nvidia has previously confirmed it is developing a dedicated tool to help locate its chips, but that solution has not yet been released.

The desire for the B300 chip is driven by its significant performance advantages over previous generations. Announced in March of the previous year, the B300 was marketed as a processor built for the age of reasoning, offering 50% more memory and FP4 compute power than the B200. It was positioned to support sophisticated large language models, including DeepSeek R1. Originally expected to ship by the end of 2025, the chip quickly became a high-demand asset for AI infrastructure developers.

The high value of these chips has led to complex smuggling schemes, most recently highlighted in August 2026 when nine individuals in Taiwan were indicted for the illegal export of B300 GPUs. The indictment came shortly after Supermicro, a major server manufacturer, fired several employees following an internal investigation into smuggling activities. The scheme reportedly involved a Taiwanese server trading firm known as Flying Tiget Tech, which held whitelisted status for purchasing GPUs, allowing it to act as a conduit for the hardware.

The scarcity of new chips has also driven up the price of older inventory. Nvidia’s A100 servers, which are now five years old, have seen their market prices triple, with units commanding up to $82,000. This economic pressure has led some operators to retool gaming-grade GPUs to handle AI workloads where possible, further complicating the tracking of hardware deployment. The combination of high demand, complex supply chains, and monitoring vulnerabilities continues to make the enforcement of export controls a challenging endeavor for regulators and manufacturers alike.