Particle.news
Download on the App Store

Nvidia Introduces Opt-In GPU Fleet Software That Can Estimate Chip Locations

The opt-in design faces scrutiny as Washington seeks stronger tools to police diversion of advanced AI processors.

Overview

  • Nvidia detailed a customer-installed agent that monitors GPU health and uses telemetry such as IP data, timestamps and latency to approximate a device’s physical or cloud location in a dashboard.
  • Company statements emphasize read-only data collection with no remote control, confirming there is no hardware backdoor or “kill switch.”
  • Location checks are framed as a secondary benefit to fleet management and uptime monitoring, and participation remains voluntary.
  • The service is set to debut on Blackwell-generation accelerators and relies on existing security features rather than new on-chip locks.
  • The rollout unfolds as the Justice Department investigates smuggling schemes exceeding $160 million and lawmakers push the Chip Security Act, while Nvidia disputes reports that China’s DeepSeek used smuggled Blackwell chips and the U.S. recently allowed limited H200 exports to approved Chinese buyers.