Designing for AI: Compute, Power, and Connectivity in the Next Data Center
Wed, Nov 18
|
03:40 PM - 04:25 PM
Session details:
- Where AI infrastructure is shifting from server-by-server upgrades to rack-scale design — with compute, power, cooling, and interconnect increasingly engineered as one integrated system rather than separate layers
- Why learning/training and inference workloads are beginning to pull architecture in different directions — from centralized hyperscale training environments to more distributed, networked inference deployments closer to end users
-
What it takes to connect and operate these emerging distributed data center environments — from middle mile capacity and fiber density to latency requirements and interconnection with regional exchange points
- How rising rack densities and sustained accelerator loads are pushing liquid cooling from a specialized option to a baseline design requirement in next-generation AI facilities
