Just 32 countries have an AI data center, but half are concentrated in just 5 of them

Only 32 countries in the world host AI computing data centers, resulting in an uneven distribution of resources for AI researchers, according to a new study from Oxford University. This leaves entire continents like South America and Africa with scarce AI resources, each with data centers in the single digits.

The researchers ran a census of the world’s publicly available AI data centers by looking at the regions where providers offered access. America and China effectively control access to the world’s AI infrastructure, and a majority of the data centers for training and inference (running AI models) are owned by US companies like Microsoft, Amazon, and Google.

Adding to the uneven distribution are limitations on the specialized hardware needed to develop and run AI models. Nvidia’s specialized GPUs are in high demand around the world, and export controls are limiting who has access. The New York Times profiled several AI researchers around the world who are desperate for the hardware to build their own domestic AI computing centers.

The study also looked at the distribution of computing resources through the lens of AI sovereignty. The study notes that the US and China are the only countries that host AI accelerators sourced from domestic chip suppliers.

The US has a major advantage when it comes to domestic chips. Thanks to market leader Nvidia’s dominance in the field, the study estimates that 95.5% of the AI accelerators in the world are powered by chips from US companies.