Nvidia developed and demonstrated an optional software agent to
The agent helps data-centre operators and regulators estimate the country in which high-end AI GPUs are running
The effort is a technical fix aimed at curbing illicit export/smuggling of advanced AI chips
Nvidia has developed and quietly demonstrated an optional software agent that can help data-centre operators and its vendors and regulators, estimate the country in which high-end AI GPUs are running, Reuters reported.
The tool, which Nvidia says originated as a fleet-management telemetry feature, uses GPU attestation data and network-latency signals generated when the processors communicate with Nvidia servers to infer a device’s location. The capability is being introduced first for the company’s newer Blackwell-generation chips.
1 December 2025
Get the latest issue of Outlook Business
The effort comes as US lawmakers and enforcement agencies intensify pressure to stop advanced AI processors from being illicitly exported to countries subject to US restrictions. Prosecutors have brought criminal cases against groups accused of trying to smuggle large volumes of Nvidia chips, and lawmakers have sought technical fixes to make diversion harder.
Nvidia frames the software primarily as a customer-installed fleet-management feature to monitor GPU health, inventory and integrity, not as a “kill switch”, and says any location verification would be opt-in for customers.
Smuggling Allegations & Denials
The announcement follows recent reports, notably in The Information and Bloomberg, that China-based startup DeepSeek trained models using Blackwell chips that were allegedly smuggled into the country after being bought in third-party markets and moved through complex routes.
Nvidia has denied seeing verified evidence that Blackwell cards were used in smuggling rings and told reporters it pursues any credible tip, but the wave of accounts has helped drive urgency around technical controls.
How It Works?
According to people familiar with the prototype, the software leverages confidential-computing features and telemetry the GPU already exposes for performance monitoring; round-trip timing to Nvidia’s attestation servers can be correlated with network geography to produce an approximate location.
Nvidia says the data is read-only, that the agent is optional and customer-installed, and that the system does not remotely disable hardware. Security experts say such telemetry can make diversion harder without embedding secret backdoors, but critics and foreign regulators may still demand transparency and proof.
