US Media: Global Distribution of AI Chips Lacks Balance

On September 2, a report on the Time magazine website highlighted an article by Billy Perrigo titled "New Study Reveals a Significant Gap in Global Ownership of Advanced AI Chips." The following is a refined summary:

As artificial intelligence expands its capabilities, the geopolitical significance of high-end chips and their locations is increasingly pronounced. The United States and China are in a competitive race to accumulate these resources, with Washington implementing sanctions to prevent Beijing from acquiring the most advanced chip categories. Surprisingly, despite the stakes, there is a lack of public data regarding the actual distribution of AI chips worldwide. A new peer-reviewed paper aims to fill this gap.

The research reveals that GPU (graphics processing unit) resources are highly concentrated in just 30 countries, with the US and China leading the pack. Many regions globally find themselves in what the authors term “computing deserts,” where no GPUs are available. This finding is crucial not only for the upcoming geopolitical competition but also for AI governance—especially concerning which governments have the authority to oversee the development and deployment of AI technologies.

Willie Leichtenberg, the first author and a professor at the Oxford Internet Institute, states, “If the actual infrastructure for running or training AI is located on your territory, you can enforce regulations.” He argues that countries lacking jurisdiction over AI infrastructure face limited legislative options, leaving them vulnerable in a world shaped by others.

Co-author Wu Boxi emphasizes, “This will influence which countries shape AI development and establish norms about what constitutes good, safe, and beneficial AI.” The paper notes that to date, the US and China possess the largest public GPU clusters in the world. While China has more regions with GPU capabilities, advanced GPU technology is highly concentrated in the US. The US has eight regions where H100 GPUs, targeted by government sanctions against China, can be rented. Conversely, while China has access to similar technology, cloud providers state they do not have any H100 GPU locations within the country.

This research sheds light on the intricate dynamics of AI chip ownership and its broader implications for international relations and regulatory frameworks in AI governance.