ExoBrain
chips and hardwarecompute infrastructuregeopolitics and sovereignty

Three layers of AI sovereignty

A new study reveals that AI sovereignty depends on physical data centre locations, ownership structures, and chip supply chains, creating a global divide between US-aligned and Chinese-aligned nations.

ExoBrain

1 min read
Three layers of AI sovereignty

This image shows the construction of OpenAI’s $60 billion Stargate project in Texas, a facility larger than Central Park that epitomises the new geography of AI power. Only 33 nations host public cloud AI compute, with just 24 possessing training-capable infrastructure, according to a new study from Oxford University. The New York Times covers the research in an interactive piece entitled; “The Global AI Divide”.

The sovereignty question operates on three levels: where data centres physically sit, who owns them, and who makes the chips inside. Countries face an uncomfortable choice; align with either US or Chinese infrastructure, or hedge by using both. Twelve nations hedge, while 18 have picked sides. All but China depend on US-designed NVIDIA chips.

Without compute, nations can’t train AI models, losing their brightest minds to GPU-rich countries. The divide could start to create dependencies as profound as oil in the 20th century.