Nvidia builds location tool to fight chip smuggling: sources

Nvidia has developed a new location verification feature that could help track where its AI chips are being used, according to sources familiar with the project.

The company has privately demonstrated the technology, which is not yet publicly available and would be rolled out as optional software for customers.

The system uses GPU telemetry and confidential computing capabilities to monitor chips across large data centers, measuring communication delays with Nvidia servers to estimate a chip’s location.

Nvidia plans to introduce the feature first on its latest Blackwell chips, which include enhanced security for attestation, and is exploring options for earlier models.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

Nvidia’s location verification must contend with detection speed and network variability

Nvidia estimates location from delays between chips and servers. Timing swings with distance, cable quality, and router hops 1.
GNSS spoofing can shift timing within 15 seconds, while one system spots spoofing in under 3 seconds for time-server protection (servers that distribute precise time to other systems) 2.
Blackwell chips get the feature first, plus attestation security (a cryptographic proof that hardware and software are in an expected state). The system must tell normal delays from tampering enough to trigger protection. GPU Trusted Execution Environments (TEEs) support attestation, like NVIDIA H100/H200 3.

GPU attestation creates opening for third-party compliance platforms

On CPUs, AMD Secure Encrypted Virtualization (SEV) provides cryptographic isolation with remote attestation logs 4. Intel Trust Domain Extensions (TDX) enables confidential virtual machines (VMs) with attestation 5.
AMD’s MI350 GPUs include Secure Boot that verifies firmware at startup. They add Device Identifier Composition Engine (DICE) attestation plus encrypted GPU-to-GPU communication for multi-tenant environments (infrastructure shared by multiple customers) 6. Hardware root-of-trust components sign attestation evidence (built-in cryptographic identity anchored in silicon) 7.
Enterprises with mixed GPU fleets need visibility across Nvidia, AMD, and Intel to check export controls and data sovereignty rules (data must remain within specific jurisdictions). Cloud and colocation providers plus enterprise IT teams can build audit services that combine telemetry with attestation data so companies can prove compliance without relying on chip vendors’ proprietary systems.