Since July 8, 2026, Azure Databricks has billed Genie, its natural-language data querying assistant, on a pay-as-you-go basis for usage beyond a free monthly allowance. Each identified human user gets 150 DBUs of free LLM usage per month, worth roughly 10.50 USD at US East rates. Usage beyond that is billed at 0.07 USD per DBU. It is a modest per-user rate, and that is exactly what makes it easy to underestimate at scale.
Where the cost surprise comes from
The allowance applies only to identified users, meaning actual people signed into a workspace. Service principals, the automated identities that run scheduled jobs, pipelines, and integrations, get no free allowance whatsoever. Any organisation that wired Genie into an automated workflow using a service principal, rather than a named user account, has been paying pay-as-you-go rates on every query since the transition took effect, with no buffer at all.
The more common exposure is simpler: teams that rolled Genie out broadly during its earlier availability, encouraging wide adoption across a data team or an entire department as a way to drive usage, now have that same broad user base each individually capable of exceeding their personal 150 DBU allowance without any central visibility into who is doing it or how often. A feature that felt free during evaluation does not stay free once genuine daily usage habits form across dozens or hundreds of users.
The tools exist, but only if someone configures them
Microsoft’s own guidance points administrators toward Genie budgets, which let account admins track and cap spend at the account, workspace, user group, or individual user level, with alerts before limits are reached. Those controls are opt-in. An organisation that has not explicitly set them up has no ceiling on Genie spend at all beyond whatever appears on the next Azure invoice.
What to check this week
If your organisation has enabled Genie for any group of users, confirm whether budgets and alerts are actually configured, not just available. Identify any automation using a service principal to query Genie, since that usage has been billed in full since day one with no allowance to offset it. And if Genie access was granted broadly during an evaluation period, it is worth reviewing who is actually using it productively versus who simply has access left switched on from a pilot that never got a follow-up review.
If your organisation needs help auditing Databricks or wider Azure AI spend, setting up budget controls before costs accumulate, or reviewing which workloads should have consumption-based AI features enabled in the first place, contact Excello Digital. We help European data and platform teams keep cloud AI costs predictable instead of finding out about them on an invoice.