Nvidia graphics cards receive tracking system to curb smuggling (Update)

On December 10, 2025, Reuters reported that Nvidia had developed a software-based location verification feature that would indicate in which country AI accelerators are operating. Nvidia had not shown the function publicly in the months before, but in presentations; a broad product announcement was still missing at that time. At the same time, Nvidia published its own blog post on December 10, 2025, in which the company describes opt-in software for the fleet management of data center GPUs, i.e. software to be installed by the customer. Nvidia explicitly emphasizes that there is no hardware tracking technology in the GPUs, no kill switches and no backdoors, and that the telemetry is read only, i.e. without the possibility of writing back to the systems.

This means that the core of the process is no longer just a media report, but is in part supported by Nvidia’s own account. However, the specific location function is framed rather indirectly in the Nvidia blog post, as part of a monitoring service with inventory, health and integrity overview, not as a stand-alone location module.

Technical approach according to Reuters and Nvidia

Reuters describes location verification as a software option that customers can install and that uses the confidential computing capabilities of Nvidia GPUs. According to this, the location is not determined using GPS, but via typical Internet procedures, specifically via time delays in communication with Nvidia servers, which should make it possible to classify the location at country or region level, comparable to other Internet-based location services.

In the blog post, Nvidia describes a solution that records telemetry such as performance, temperature and power consumption as well as errors and configuration statuses and transmits them to a portal service hosted by Nvidia. The data is managed and customizable by the customer, the telemetry is read only, and the software is positioned as an optional service aimed at availability, efficiency and operational stability. The component relevant to site proximity, namely latency measurement to the Nvidia infrastructure, is described in the Reuters report as part of the solution. Nvidia also states in the Reuters report that there is no function with which Nvidia can remotely control registered systems or trigger measures, nor is there any mechanism that could deactivate a GPU.

Open source announcement, what is secured and what remains open

Reuters reports that Nvidia is planning to make the software open source so that external security researchers can examine the code. Nvidia itself writes in the blog post that an open source client agent is planned, which will serve as an example and as an auditable component. Which license is actually used, when the source code will be published and which parts of the overall system, i.e. only the agent or other components, will actually be open in the end, remains open in the published description. However, this cannot yet be clearly proven.

Classification in export controls, smuggling allegations and political dynamics

Reuters places the development in the context of US export controls and points to political pressure from the White House and Congress to establish location verification as an instrument against circumvention and smuggling. In December 2025, it was also reported that US President Donald Trump wants to allow exports of the predecessor model H200 to China again in principle, combined with a publicly stated levy of 25 percent. The statement mentioned yesterday that there could be revenue sharing or fee models is therefore at least partially covered by Reuters reports. However, there is no evidence of any further technical link between H200 export approval and the new location service.

Chinese reports and alleged internal leaks

Chinese-language articles, including those from major financial and news aggregators, take up the Reuters report almost word for word, including the points, optional software, use of confidential computing, latency-based classification, no official initial announcement in the original phase. As it stands, these texts do not provide any additional, independently verifiable technical details, but are predominantly secondary use. Specific internal leaks, for example on concrete implementation details, backend architecture, data fields, key material, attestation flows or an exact roadmap, cannot be found in verifiable form in the reliable sources that are accessible here. Unfortunately, anything beyond this cannot be verified at present.

Security, control and acceptance issues, what can be deduced from this

In the current, publicly comprehensible presentation, Nvidia is clearly trying to bring together two contradictory requirements, namely compliance pressure towards location verification on the one hand and concerns about hidden control functions on the other. In this context, Reuters also reports that the debate about location verification in the USA has contributed to China’s cyber security authority questioning Nvidia about possible backdoors and security risks; Nvidia has denied such backdoors.

Technically, the decisive factor remains how strong the location statement really is in the end. Latency and communication patterns are noisy in real networks, can be influenced by routing, peering, VPN, proxies and distributed infrastructures, and are only unambiguous to a limited extent in multi-region deployments. The fact that Nvidia and Reuters describe the function more as a country indicator and a rough classification fits in with these limitations. Based on the information published to date, it is not possible to make a reliable statement about the effectiveness against smuggling.

Conclusion

It is currently verifiable that Nvidia is developing an opt in software for data center fleet management using GPU telemetry, that Nvidia emphasizes the absence of kill switch, backdoor and remote control, and that Reuters describes this software as the basis for location verification at country level, which is flanked by latency to the Nvidia infrastructure and confidential computing capabilities.

It is also verifiable that Nvidia holds out the prospect of the open source status of a client agent, and that the political context, export controls, investigations into circumvention routes and the discussion about location verification are strongly influencing the development. The level of detail of the location logic, the concrete scope of an open source release, as well as alleged internal leaks on implementation and roadmap, which go beyond Reuters and Nvidia’s own information, remain unverifiable at present.

According to consistent media reports, the US graphics chip developer Nvidia has introduced a new software-based system in recent months that can be used to determine the approximate location of certain graphics cards. The aim of the development is to make the unauthorized resale and smuggling of powerful AI accelerators into countries with existing trade restrictions more difficult. A public product announcement has not yet been made.

According to company sources, this is an optional telemetry software that is based on the existing security functions of Nvidia architectures. Specifically, so-called confidential computing functions are to be used to indirectly determine location information. The technology is not designed as a stand-alone location module, but as a supplement to existing management and monitoring tools for data center operators.

The location is not determined using classic GPS or satellite data. Instead, the software combines latency times and communication patterns between the operated systems and Nvidia servers with other technical metadata. On this basis, a regional assignment should be possible, which, according to Nvidia, deliberately only allows a rough geographical classification. According to the company, precise localization of individual systems is not planned.

Nvidia emphasizes that the use of the software is voluntary and can only be activated by operators of large server installations. The recorded telemetry data is read-only and does not allow external access to the hardware. A remote-controlled shutdown mechanism or so-called kill switch is not part of the system. These statements cannot currently be independently verified.

According to the Reuters news agency, the technology has already been shown several times in non-public presentations. A formal introduction is apparently planned in connection with the upcoming Blackwell generation of AI accelerators. These graphics units are primarily intended for use in high-performance data centers and AI training environments.

Another aspect of the development is the planned release of the software under an open source license. Nvidia justifies this step with the aim of enabling independent security researchers to review the code. Whether and to what extent the source code will actually be published is currently unclear.

The economic and political background to this measure is closely linked to the US government’s current export controls. High-performance GPUs for AI applications are currently not allowed to be exported to China. According to the US authorities, there have been repeated investigations and arrests in connection with suspected GPU smuggling in recent months. At the same time, there are indications that individual export restrictions could be relaxed under certain conditions. These include reports of possible revenue sharing for the US state when exporting selected models such as the H200 accelerator. This information cannot be conclusively verified at present.

The introduction of software-supported location determination marks a new approach in the enforcement of export regulations at hardware level. It remains to be seen whether the system is effective in practice or whether it raises new questions about control, transparency and international regulation.

Conclusion

The software-based location verification for AI graphics processors developed by Nvidia represents a technical attempt to enforce existing export controls more effectively without directly interfering with the functionality of the hardware. According to the information known to date, the approach is based on indirect measurement methods such as network latencies and telecoordination data and only enables a rough regional classification of the systems in operation. According to the company, precise localization of individual servers or remote intervention in the operation of GPUs is not planned and has not yet been verified.

The voluntary activation by data center operators and the planned open source provision indicate that Nvidia is attempting to combine regulatory requirements with transparency and customer acceptance. At the same time, the project underlines the increasing political importance of high-performance computers and AI accelerators in international trade. Whether the system can actually effectively curb smuggling or merely represents an additional formal layer of control cannot be conclusively assessed at this time.