H2 and AI: Lessons from the World's Most Energy Efficient Data Center
Abstract
Artificial Intelligence• Artificial intelligence (AI) is the ability for computers to mimic human intelligence-or at least do things that can seem intelligent.• AI's applications fall into three categories: classifications, forecasting, and generation, which is the new kid on the block.• Web services traditionally have ~10 kW racks and loosely coupled CPUs.• Newer "AI" GPU chips consume dramatically more power than traditional Intel CPUs.• Rising power densities require component-level liquid cooling.• Traditional "supercomputer" HPC has 60-70 kW racks.• Nvidia roadmap for 600 kW racks in ~2027.• A large "hyperscalar" data center used to be 100 MW; AI data centers are in the range of 1-5 GW.
Fields of Study & Subjects
Energy Efficiency in Computing
Cloud Computing and Resource Management
Green IT and Sustainability
Publication details
- Journal / source
- OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information)
- Publication date
- 27 January 2026
- Language
- EN