After a $300M smuggling case, the US explores 'Find My'-style tracking for AI chips

A $300 million diversion case

On 1 October 2026, US authorities arrested a man they accuse of moving more than $300 million worth of GPU-equipped servers into China while filing paperwork that described the shipments as bound for Malaysia and Singapore, IEEE Spectrum reports. The case illustrates the strain on Washington's export control regime: restrictions on advanced chips have been in place since 2022, yet restricted hardware keeps turning up in China regardless, and enforcement has largely depended on catching schemes after the equipment has already moved.

Congress moves toward location checks

In March, the House Foreign Affairs Committee approved a bill that would require location verification for advanced AI chips, and ahead of the US-China summit in late September legislators pushed for a floor vote on it and related measures, according to IEEE Spectrum. If a location requirement becomes law, chipmakers will need technical mechanisms to prove where their silicon physically sits — something like Apple's Find My, but for data center hardware.

How ping-based verification would work

A September report from the Center for Security and Emerging Technology (CSET) weighs the options and argues that a practical regime would combine ping-based location verification (PLV) with physical inspections.

PLV times signals as they travel from a chip to several trusted "landmark" servers around the world and back; the round-trip measurements narrow down the chip's position. Kyle Miller, a senior research analyst at CSET and coauthor of the report, says new chips could ship with PLV built into the hardware, while silicon already in the field could gain the capability through a firmware update. Tracking would not cover everything: coauthor Jacob Feldgoise says the priority is the highest-end data center GPUs rather than AI silicon in general.

Miller also sees a deterrent effect. A smuggler knows that powering on a chip will start it pinging landmark servers, and if a chip surfaces somewhere other than its intended destination, the timing data helps investigators work out when and how it moved.

Keys, certificates and fingerprints

Secure computing firm Lucid Computing has published a draft specification for encrypted "sovereignty certificates", cryptographic credentials meant to provide verifiable proof of where a CPU or GPU was located at a specific moment. CEO Kristian Rönn describes the mechanism: a chip keeps a private key inside a trusted execution environment, signs messages with it, and landmark servers time those messages to bound the chip's possible location.

That approach has weaknesses. An adversary with physical access could stop a chip from signalling, or extract its keys and spoof a signal from an approved location. Researchers are therefore also exploring hardware fingerprinting: one proof-of-concept study generated unique identifiers for Nvidia H200 GPUs from manufacturing variations, seeded with random numbers, offering a way to confirm a chip is what it claims to be even without a hardware root of trust.

What tracking cannot solve

CSET is candid about the limits. PLV only uncovers diversion after it happens; it cannot see chips that are offline, in storage, in transit or not yet activated, and it cannot distinguish a legitimate reason for silence from a malicious one. Physical audits of data centers remain essential, with Miller describing PLV as a tool that "generates leads" for investigators to chase. The report also warns that such a system would be costly, and Miller notes that if China begins producing advanced AI chips domestically, tracking US exports loses much of its strategic value.

A further gap sits outside the tracking debate entirely. Hanna Dohmen, another CSET coauthor, points out that Chinese firms can lawfully use advanced AI chips remotely through cloud providers, and data centers in Malaysia and elsewhere in Southeast Asia are increasingly serving Chinese customers.

Why it matters

The $300 million arrest signals that export controls are now being enforced through criminal charges rather than paperwork alone. If location-verification legislation passes, compliance obligations would land on chipmakers, server builders and data center operators, adding cost and engineering work to every high-end GPU deployment. Feldgoise expects due diligence requirements to keep tightening as compute is treated as a strategic national and economic resource, and Rönn frames sovereignty certificates as a "lie detector" for data centers. For anyone selling or operating advanced AI hardware across borders, exports taken at face value are on their way out.