US explores a Find My-style tracker for Nvidia AI chips

Kyle Miller and Jacob Feldgoise, senior researchers at Georgetown University's Center for Security and Emerging Technology (CSET), argue the focus should be on the most powerful data-center GPUs. Manufacturers could build PLV into new chips, while chips already shipped could get it through a firmware update. Encrypted certificates could also confirm a device's location at a given moment.

Can AI chip tracking actually work?

Not perfectly. IEEE Spectrum notes that signal checks cannot follow chips that are switched off, sitting in a warehouse or in transit. And if a chip stops answering, nobody can tell whether it is a technical fault or an attempt to hide. People with physical access could also block the signals or pull out cryptographic keys. So experts suggest pairing digital checks with physical inspections of data centers.

Hardware fingerprinting is another route. Researchers have already shown they can create unique fingerprints for Nvidia H200 GPUs from manufacturing quirks and random numbers.

Why now?

US export limits introduced in 2022 did not stop advanced accelerators from reaching China. In March 2026 the House Foreign Affairs Committee approved a bill requiring location checks on advanced AI chips, and in September lawmakers urged a floor vote. Experts also warn that enforcement would be expensive and that buyers could still get around it through foreign cloud services.