Databricks Adds Hard Caps to AI Spend Controls in Unity Gateway
Summary
Databricks releases AI Spend Controls in Unity Gateway, letting organizations set budgets per user, workspace, use case, or account, and hard caps that halt requests once a limit is exceeded. The company cites a $2,000 per-user monthly budget as an example. Databricks says the move targets runaway retry loops and accidental agent experiments that inflate AI bills.
Key Points
- Databricks' Kevin Stumpf says workspace budgets can hold production to $50,000 a month, sandbox to $5,000 a month.
- Every request logs to Unity Catalog system tables with DBU costs, not just token counts.
- Users raise a hard cap by editing the budget's spend limit or waiting for the next billing period.