Trends in AI Supercomputers

Konstantin Friedemann Pilz, James Sanders, Robi Rahman, Lennart Heim

Keywords: AI Supercomputers, Compute, AI Infrastructure, Trends in AI, AI Data Centers

TL;DR: Compiled a dataset of 500+ AI supercomputers over six years and analyzed trends in performance, cost, power, distribution; discussed policy implications

Abstract: Frontier AI development requires AI supercomputers with thousands of AI chips. Yet, analysis of developments in these systems is limited. We create a dataset of 500 AI supercomputers from 2019 to 2025 and quantify key trends. We find that computational performance of AI supercomputers has doubled every nine months, while hardware acquisition cost and power needs have doubled yearly. The leading system in March 2025, xAI's Colossus, had a hardware cost of \$7B, and required 300 MW of power, as much as 250,000 households. While the public sector owned 60\% of AI supercomputer performance in 2019, this share declined to only 20\% by 2025, which may limit access to frontier capabilities for academic researchers. The United States dominates AI supercomputers, owning 75\% of performance, suggesting a large degree of geographical concentration of compute. Our study provides visibility into AI infrastructure trends, allowing policymakers to make more informed AI governance decisions.

Submission Number: 24