Nvidia's Most Powerful AI Chips Detected in Chinese State Network Despite Export Bans
Security firm Lava has identified 32 Blackwell Ultra B300 GPUs within China Telecom's infrastructure, highlighting persistent challenges in enforcing U.S. export controls on high-performance silicon.
By Adrian Romano ·

Israeli cybersecurity firm Lava has confirmed the presence of 32 Nvidia Blackwell Ultra B300 graphics processing units within the internal network of China Telecom, a state-owned telecommunications provider. The discovery is significant because these specific chips are subject to strict U.S. export restrictions, which generally prohibit their sale to buyers in China. The finding raises critical questions about the efficacy of current enforcement mechanisms and the pathways through which advanced hardware continues to reach sanctioned entities despite regulatory barriers.
The B300 chips are not only restricted by export policy but are also limited in their initial distribution. Nvidia sells these high-performance AI data center components primarily to whitelisted buyers who contractually agree not to resell or transfer the hardware to sanctioned parties, including those in China. The identification of 32 such units inside a major Chinese state network suggests that either the whitelist agreements were bypassed, or the chips were diverted after being sold to approved intermediaries. This incident adds to a growing list of examples where controlled technology has reached Chinese infrastructure, challenging the assumption that export bans create a complete barrier.
Nvidia has responded to the report with a mix of acknowledgment and dismissal. While the company did not deny Lava’s findings, a spokesperson characterized the issue as minor in the broader context of global computing resources. The spokesperson stated that recent court cases have shown the diversion of controlled U.S. products to be less than half of one percent. They described this amount as a "drop in the bucket" compared to the vast volume of domestic compute capability China already possesses. Nvidia argued that with tens of millions of GPUs already in service in approved countries, such smuggling incidents are a non-factor in the overall balance of AI hardware.
The context of this discovery is underscored by previous reports of large-scale smuggling. In 2025, it was documented that Chinese firms imported approximately $1 billion worth of Nvidia AI chips over a three-month period. At that time, the B200 was the primary target, but distributors were already preparing for the arrival of the B300 series. The current finding indicates that the market for these chips remains active and that distributors are quickly adapting to new product releases. The B300, announced in March of last year, offers substantial performance improvements, including 50% more memory and FP4 compute capabilities compared to the previous-generation B200. Nvidia originally positioned the chip as being "built for the age of reasoning," citing its utility for sophisticated large language models like DeepSeek R1.
The mechanism used to locate the chips provides insight into potential vulnerabilities in data center security. Lava identified the GPUs 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 often be accessed from outside the immediate network, and it does not always require password authentication. This accessibility allows external security firms to detect specific hardware signatures. Nvidia has previously confirmed it is developing a tool to help locate its chips, but that solution has not yet been released, leaving a temporary gap in visibility for data center operators.
The desire for B300 chips in China is driven by their superior specifications and the critical role they play in modern AI infrastructure. The chips were expected to ship by the end of 2025 and quickly became high-demand items. In August 2026, the smuggling issue reached a legal milestone when nine individuals in Taiwan were indicted for the illegal export of B300 GPUs to China. This indictment followed closely on the heels of Supermicro firing several employees after an internal investigation into smuggling activities. The case involved a Taiwanese server trading firm, Flying Tiget Tech, which held whitelisted status for GPU purchases, illustrating how legitimate channels can be exploited for illicit transfers.
The scarcity of Nvidia’s latest silicon has also distorted market prices for older generations. A100 servers, which are five years old, have seen their prices triple, with some units selling for up to $82,000. This price surge reflects the intense demand for any available Nvidia hardware capable of running AI workloads. The economic pressure is so significant that some organizations are resorting to retooling gaming GPUs to handle AI tasks where possible, a move that underscores the premium placed on dedicated data center chips like the B300.
For the Midwest and broader U.S. tech sector, the stakes involve both economic competitiveness and strategic advantage in the AI race. If China can consistently acquire the latest Nvidia hardware, it maintains the ability to train and run complex AI models that could rival those developed in the United States. While Nvidia maintains that the volume of smuggled chips is statistically insignificant, the presence of the most advanced B300 units in a state-owned network serves as a tangible reminder that export controls require constant vigilance. The incident highlights the complex interplay between global supply chains, regulatory enforcement, and the rapid pace of innovation in artificial intelligence.
The discovery by Lava serves as a case study in the challenges of controlling technology flow in an interconnected global market. As Nvidia continues to push the boundaries of performance with new architectures, the pressure to secure these assets will likely increase. For data center operators, the incident may prompt tighter security protocols around monitoring tools like the DCGM Exporter. Meanwhile, policymakers and industry leaders must continue to evaluate whether export bans are sufficient to maintain a lead in AI capabilities or if additional measures are needed to prevent the diversion of critical hardware to strategic rivals.






