AI Runtime is Generally Available

Why it matters

Databricks has announced the General Availability of AI Runtime, a fully managed GPU infrastructure offering that lets customers train demanding models inside the same platform where their enterprise data resides. The company frames the launch as a response to a common pain point: AI teams lose time securing GPU capacity, handling failures, and waiting for experiments to start, while running GPU infrastructure apart from the Lakehouse creates governance gaps that put sensitive training data at risk. Databricks says hundreds of companies, including Robinhood, PicPay, Merck, Flo Health, TinyFish, and FactSet, already use AI Runtime in production, with applications spanning healthcare, manufacturing, and consumer-facing agents. Its own AI Research team uses the service to train the models behind KARL and AI Parse. The announcement highlights integration with Lakeflow for orchestration, a new observability experience in MLflow for debugging distributed training, and Genie Code, which embeds agentic debugging skills throughout the product. Customer testimonials from Robinhood, Flo Health, FactSet, Superhuman, and QuantHealth emphasize faster iteration and reduced infrastructure overhead.