AI Can’t Wait: How to Overcome GPU Scarcity and Keep Workloads Running

GPU scarcity, cloud capacity constraints, and regional outages are challenges enterprises simply cannot afford: they block model training and serving, and leave mission-critical workloads queuing or failing with no good recourse.  By intelligently routing workloads to the optimal compute while preserving each customer's data residency, security controls, and compliance boundary, Databricks is solving both problems transparently. In this talk, we'll show how we're delivering GPU access to customers who are capacity-constrained in their home region, and offering enterprise-grade availability guarantees for serverless workloads without requiring customers to manage multi-cloud infrastructure themselves.