U.S. export controls have restricted sales of the most advanced AI chips to China since 2022, yet chips continue to reach the country through smuggling and diversion. A proposed U.S. requirement would make manufacturers verify where certain high-end chips are located. One approach, ping-based location verification, has chips send timed signals to trusted servers around the world; comparing travel times can estimate a chip’s location and reveal when it has moved somewhere unexpected. Chipmakers could build this capability into new hardware or add it through firmware updates. Researchers argue that tracking should focus on the most powerful data-center GPUs rather than every chip.
Location checks could deter some diversion and help investigators identify when and where chips moved, but they cannot locate devices that are switched off, stored, or in transit, and a missing response does not prove wrongdoing. Physical inspections would therefore be needed to follow up on suspicious signals. The system also faces attacks: someone with physical access could block signals or steal cryptographic keys and spoof a location. Proposed defenses include tamper-resistant hardware, cryptographic “sovereignty certificates” that attest to location, and distinctive hardware fingerprints; a proof of concept has fingerprinted Nvidia H200 GPUs. These protections would add cost and still leave policy gaps, including Chinese companies’ ability to rent access to chips in foreign cloud data centers. As computing power becomes a strategic resource, stricter export checks are likely even if tracking is not mandated.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.