Exclusive: Databricks rolls out AI spend controls
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Sarah Grillo/Axios
Databricks is launching new tools to help companies cap AI costs, after finding customers had accidentally spent tens of millions of dollars on their broader AI bills in a single month.
Why it matters: AI agents are making corporate software bills harder to predict, and Databricks wants to be the layer companies use to keep them under control.
The big picture: The rise of AI agents that autonomously interact with models without much human supervision is increasing AI usage and forcing companies to blow budgets.
- Traditional cloud cost tools often flag overspending only after the damage is done.
- Databricks hopes its new tool helps firms pivot from token maxing to "value maxing," Patrick Wendell, Databricks co-founder, tells Axios.
What they're saying: Wendell says he's seen companies go from little or no AI spending to accidentally spending tens of millions of dollars a month.
- "We've definitely seen mistakes that are in the millions," he added.
- Wendell said AI token costs are entering the top three highest expenses among customers behind salaries and other IT expenses. That's a problem the team has been working to address.
Driving the news: The new Databricks product — Unity AI Gateway — will include AI spend limits, protections to prevent "runaway spend" and recommendations designed to help companies manage AI costs across multiple providers.
- The gateway can recommend cheaper models for tasks that don't require the most token-heavy or expensive options.
- It will monitor individual user sessions, which will then inform feedback on efficiency of AI usage, which could involve removing employee access or moving them to a cheaper model if they aren't using the tools efficiently.
Zoom out: Monitoring individual AI use could be a tough sell to employees.
- Wendell argued the pushback is usually stronger when companies use employee activity for AI training, rather than cost control.
- "The amount of data we need for this purpose is relatively narrow, and it only is related directly to their use of these coding tools," Wendell said.
The bottom line: Better cost controls could help companies rein in AI bills — but they could also pressure model providers counting on fast-growing enterprise usage.
