ÌÇÐÄÆƽâ°æ

The United States was home to ** public cloud AI compute regions as of late 2024, with ** of these classified as training-relevant regions. 

AI sovereignty concerns mount

With AI considered central to digital transformation and broader economic growth, attention is increasingly focused on who controls the infrastructure needed to train the most advanced AI models. This concern falls under the broader notion of AI sovereignty, a term describing the strategic influence over AI development, covering infrastructure, ownership, data, and policy. Tech leaders argue that control of the most powerful AI infrastructure will be key to shaping the next era of AI-driven innovation, spurring governments to develop policies that protect and advance their strategic interests.

Training vs. inference

AI models are trained at specialized data centers equipped with the latest high-performance GPUs. These facilities are often grouped into clusters of facilities, allowing providers to scale capacity efficiently while minimizing latency and ensuring redundancy. Once trained, the models are deployed for inference, the process of generating predictions or outputs in response to user queries. Inference infrastructure typically has different priorities than training, emphasizing widespread availability, low-latency access, and scalability across regions, rather than the raw computational intensity required during training.