Databricks makes AI Runtime, managed GPU training infrastructure, generally available
Databricks announces general availability of AI Runtime, fully managed GPU infrastructure that lets customers train models where their enterprise data already lives.
By GennnAi··1 min read
The serverless, on-demand service promises launches within seconds on RDMA-enabled GPUs and works with Ray and custom containers. It operates within the governance perimeter of Unity Catalog. Hundreds of companies, including Robinhood, Merck and FactSet, already use it. At Superhuman, training moved from roughly seven days to 48 hours, a 3.5x speedup, after shifting from a single-node setup of eight A100 GPUs to 16 H100 GPUs.
Source: Databricks, published on October 21, 2026.