Cfoto | Future Publishing | Getty Images
Nvidia is developing software that could provide location verification for its AI graphics processing units (GPUs), a move that comes as Washington ramps up efforts to prevent restricted chips from being used in countries like China.
The opt-in service uses a client software agent that Nvidia chip customers can install to monitor the health of their AI GPUs, the company said in a blog post on Wednesday.
Nvidia also said that customers “will be able to visualize their GPU fleet utilization in a dashboard, globally or by compute zones — groups of nodes enrolled in the same physical or cloud locations.”
However, Nvidia told CNBC in a statement that the latest software does not give the company or outside actors the ability to disable its chips.
“There is no kill switch,” it added. “For GPU health, there are no features that allow NVIDIA to remotely control or take action on registered systems. It is read–only telemetry sent to NVIDIA.”
Telemetry is the automated process of collecting and transmitting data from remote or inaccessible sources to a central location for monitoring, analysis and optimization.
The ability to locate a device depends on the type of sensor data collected and transmitted, such as IP-based network information, timestamps, or other system-level signals that can be mapped to physical or cloud locations.
A screenshot of the software posted on Nvidia’s blog showed details such as the machine’s IP address and location.
A screenshot of the software posted on Nvidia’s blog showed details such as the machine’s IP address and location.
Nvidia blog screenshot | Opt-In NVIDIA Software Enables Data Center Fleet Management
Lukasz Olejnik, a senior research fellow at the Department of War Studies, King’s College London, said that while Nvidia indicated that its GPUs do not have hardware tracking technology, the blog did not specify if the data “uses customer input, network data, cloud provider metadata, or other methods.”
“In principle, also, the sent data contains metadata like network address, which may enable location in practice,” Olejnik, who is also an independent consultant, told CNBC.
The software could also detect any unexpected usage patterns that differ from what was declared, he added.
The latest features from Nvidia follow calls by lawmakers in Washington for the company to outfit its chips with tracking software that could help enforce export controls.
Those rules bar Nvidia from selling its more advanced AI chips to companies in China and other prohibited locations without a special license. While Trump has recently said he plans to roll back some of these export restrictions, those on Nvidia’s cutting-edge chips will remain in place.
In May, Senator Tom Cotton and a bipartisan group of eight lawmakers introduced the Chip Security Act, which, if passed, would mandate security mechanisms and location verification in advanced AI chips.
“Firms affected by U.S. export controls or China-related restrictions could use the system to verify and prove their GPU fleets remain in approved locations and state, and demonstrate compliant usage to regulators,” Olejn noted.
“That could actually help in compliance and indirectly on investment outlook positively.”
Pressure on Nvidia has intensified after Justice Department investigations into alleged smuggling rings that moved over $160 million in Nvidia chips to China.
However, Chinese officials have pushed back, warning Nvidia against equipping its chips with tracking features, as well as “potential backdoors and vulnerabilities.”
Following a national security investigation into some of Nvidia’s chips to check for these backdoors, Chinese officials have prevented local tech companies from purchasing products from the American chip designer.
Despite a green light from U.S. President Donald Trump for Nvidia to ship its previously restricted H200 chips to China, Beijing is reportedly undecided about whether to permit the imports.
