Databricks Makes AI Runtime Generally Available

Databricks announced on its website that AI Runtime is generally available, bringing managed GPU infrastructure directly into its platform. The service handles GPU provisioning, cluster configuration, networking, infrastructure health checks and workload optimization.

AI Runtime supports distributed training across hundreds of GPUs, with workloads launching in seconds on clusters that use RDMA. Teams can use their own code and containers, run Ray workloads, access GPUs through a command line interface, notebooks or development environments, and inspect streaming logs through MLflow.

The service applies Unity Catalog permissions to training data, checkpoints and logs. It also provides virtual machine level workload isolation, PrivateLink support, network access controls, Lakeflow orchestration and usage records for tracking GPU costs.

Databricks says hundreds of companies already use AI Runtime, including Robinhood, PicPay, Merck, Flo Health, TinyFish and FactSet, and that its own AI research team uses it to train its KARL and AI Parse models.