Databricks makes AI Runtime generally available for GPU training

Databricks on October 21, 2026 announced general availability of AI Runtime (AIR), a fully managed serverless GPU platform built into its Lakehouse for training and fine-tuning deep learning models on enterprise data. Hundreds of companies including Robinhood, Merck and FactSet already use it, with over 100,000 GPU-hours of training runs.

• AIR is a serverless GPU platform inside the Lakehouse with managed clusters and networking
• Over 100,000 GPU-hours of training and hundreds of customer companies
• Superhuman cut training from 7 days to 48 hours, a 3.5x speedup
• Supports PyTorch, Ray, Docker and integrates with MLflow and Lakeflow