The Chinese NVIDIA Problem
Over the past few years, the relationship between the West and China has been somewhat rough. On one side you have the US constantly throwing accusations of IP theft, unfair trade practices, and lack of respect for human rights, while China constantly accuses the West of hegemony, imperialism, and wanting to effectively control the world. In fact, everything from trade wars to unexpected tariffs and theft of IP have been thrown around like a frisbee by both sides in what can only be described as a never ending war (with no sight in victory). But while this has seen each side engage in tit-for-tat actions, it has been the West that has taken the lead in this fight. One such action has been restricting China’s access to the latest technologies, especially those that power AI, through export controls and restrictions on sales of graphics cards, high-performance CPUs, and other components to Chinese businesses. Of course, China has still found ways to get these devices, but doing so has introduced practical challenges for the West's export-control efforts. For example, when China announced the development of DeepSeek, a reported AI system that some reports said used components obtained from Western sources, the AI community was left utterly shocked as China had developed something that could match Western technologies. Thus, when looking at the disputes over the past few years, it has become abundanetly clear that the West restricting China’s access to hardware is a double edged sword.
NVIDIA to track GPUs
Recently, reports have emerged of Nvidia testing a new software solution that would allow it to track the location of its AI chips. While Nvidia has not made any official statements regarding this move, rumours suggest that the move comes after numerous reports of Nvidia chips finding their way into Chinese servers despite strict trade restrictions by the US government. According to recent reports, the software being developed by Nvidia would be able to verify the location of servers that are looking to purchase its latest AI hardware, the Blackwell chips and track chip performance, using server communication delays to estimate a device's location. Reports say the feature would initially be available for Blackwell chips and would be optional for customers. Allegations have surfaced that some AI models in China (such as DeepSeek) were trained using Nvidia Blackwell GPUs that may have been smuggled there, but Nvidia has not confirmed those allegations. Nvidia has said it has not seen evidence of “phantom data centers” or any smuggling activities involving its chips and has said it investigates any credible tips. Separately, the U.S. government recently approved sales of older H200 AI chips to approved customers in China, but that approval does not extend to Nvidia’s newer Blackwell chips. While this allows Chinese engineers to continue developing their own solutions on older hardware, the new software tools being developed by Nvidia are described in reports as intended for location verification and optional use by customers, rather than as an automatic lock that would disable older chips.
Is this a step too far for NVIDIA?
Now, before we go any further, the report that NVIDIA will start tracking its hardware cannot be proven 100% true, but it is probably more likely than not. Of course, the reason NVIDIA would want to do this is to try to prevent Chinese smugglers from diverting NVIDIA hardware into China. But considering how governments around the world are trying to introduce new legislation to limit what companies can and cannot do, NVIDIA might take this further and implement tracking on hardware not destined for China, potentially collecting data on what hardware is running in the West. Now, this does not necessarily mean that NVIDIA would use this data to spy on users or monitor their activities, but it does raise some serious concerns regarding security. For example, the moment NVIDIA starts tracking hardware, they could potentially observe or infer the GPU model, power consumption patterns, and the software or workloads in use. This could be used to help infer what GPU users are purchasing, how much processing power they need, and what software they are using. This data could then be used to better target customers with advertising, but this would also give NVIDIA a very good insight into their customers, something many find objectionable from a privacy perspective. There are plenty of other implications that come with NVIDIA tracking hardware, and it is clear that the US government is putting pressure on companies to restrict Chinese access to technology. But is this a step too far? Will this cause problems for customers down the line? How will this affect the relationship between the West and China? The answer to these questions is not clear, but one thing that is clear is that NVIDIA having the ability to track hardware could be a significant security and privacy concern for those who legitimately purchase NVIDIA products. However, as we move towards open standards, this threat may fade away. For example, compatibility layers and alternative software stacks (for example, AMD's HIP/ROCm, Intel's oneAPI, and portability efforts) or open ISA efforts (such as RISC-V) could reduce dependence on NVIDIA's proprietary CUDA ecosystem. Implementing full CUDA compatibility is non-trivial, but third parties could develop hardware or translation layers that run similar AI workloads. So, in the future, it is possible that we see a world where Chinese hardware becomes more prevalent, and the West may have to rely more on foreign tech in some areas. The future is unclear, but NVIDIA having the power to track graphics cards is one that makes me shudder.
You may also like
