Why most AI GPU clusters waste 30–60% capacity

Javier Abrego Lorente examines a common but often hidden issue in AI infrastructure: large-scale GPU underutilization caused by static allocation and fragmented ownership models.

Drawing from real-world platform observations, he explains why traditional capacity planning fails once inference workloads grow and multiple teams compete for GPUs.

Attendees will learn:
• Where GPU fragmentation typically originates
• How multi-tenant inference changes capacity dynamics
• Why visibility alone is not enough
• Practical patterns to improve utilization and cost efficiency

This session is aimed at platform engineers and infrastructure teams running AI workloads at scale.