FinOps for AI: Why GPU cost control breaks at scale

Javier Abrego explores why conventional FinOps approaches struggle when applied to AI infrastructure and GPU-heavy workloads.

As organizations scale inference and training, cost visibility alone often fails to prevent runaway spending because of concurrency patterns, static reservations, and fragmented ownership models.

Attendees learn:
• Why traditional FinOps models break with AI workloads
• Hidden cost drivers in multi-tenant GPU environments
• The gap between allocation and real utilization
• Platform-level strategies to regain economic control

This session is aimed at platform and infrastructure teams supporting AI at scale.