- Lead. Israeli security company Lava reported on October 22 that it detected 32 Nvidia Blackwell Ultra B300 GPUs operating on network endpoints linked to China Telecom, a state-owned carrier that U.S. export controls bar from receiving the chips.
- Fact. Lava identified the chips through publicly reachable Nvidia monitoring instances lacking authentication; the physical location of the hardware could not be established from network data alone, though IP ranges resolved to China Telecom infrastructure.
- Stake. The B300 is Nvidia’s most advanced AI accelerator and is subject to stringent export controls; the disclosure arrives as regulators are tightening oversight of cloud-based and third-party chip access routes.
Lava, an Israeli cybersecurity firm, disclosed on October 22 that its researchers had identified 32 Nvidia Blackwell Ultra B300 graphics processing units connected to network infrastructure associated with China Telecom. The chips, representing the leading edge of U.S. AI accelerator technology, are subject to export controls that prohibit their sale to entities on U.S. government restricted lists, including Chinese state-owned telecommunications carriers. The discovery was first reported by The Information and confirmed through Lava’s public technical disclosures.
How the Chips Were Found
Researchers found the chips by scanning for Nvidia DCGM Exporter instances — a monitoring tool that reports GPU utilization — that were accessible over the open internet without authentication. Each instance exposed identifying data that allowed Lava to determine the GPU model. The IP address ranges hosting the instances resolved to infrastructure linked to China Telecom, though network analysts noted that geolocation of IP addresses cannot independently establish the physical location of the underlying hardware. In the same disclosure, Lava published CVE-2026-47483, a high-severity vulnerability in the same monitoring tooling.
Nvidia characterized chip smuggling as “a drop in the bucket” in a statement to Tom’s Hardware, without disputing the core finding. Analysts suspect the chips reached China through routes similar to those documented in previous enforcement cases: a California CEO was charged earlier this year in a $300 million GPU smuggling ring that routed Nvidia chips through shell companies.
The Enforcement Gap
The Lava disclosure arrives as U.S. regulators are actively trying to close the pathways the firm identified. The Commerce Department has drafted a rule that would block China from renting U.S. AI chips through overseas data centers, precisely because physical location controls are difficult to enforce once chips leave U.S. jurisdiction. The Treasury’s proposed COINS Act rules, published this week, take a parallel approach by restricting outbound investment in AI and semiconductor sectors.
The 32 chips represent a small fraction of the estimated tens of thousands of high-end GPUs that U.S. officials believe have reached restricted parties through circumvention networks. Their significance lies less in volume than in visibility: operating on openly reachable monitoring endpoints accessible to any researcher with a port scanner, the chips suggest that end-use compliance after export is not being enforced rigorously enough to conceal the fact of their presence. Washington regulators are expected to scrutinize the case; whether it produces new compliance mandates for chipmakers — such as requirements to remotely monitor or deactivate chips shipped to third parties — is the policy question now before the Bureau of Industry and Security.