Brickster Newsletter
|
|
01
Genie pricing. 150 free DBUs, the service-principal trap, and why AI Gateway budgets landed the same week.
|
|
02
This week, in brief. Asset Bundles get a new name, Genie lands in Slack, an SDK refresh, and the LTAP push.
|
|
03
From Brickster.ai: Now in five more languages, plus this week's digest.
|
|
|
|
PRICING · EFFECTIVE TODAY
The free lunch is over. The meter is on.
Genie now bills by usage. 150 free DBUs per user per month, then pay-as-you-go. No seat fees, but no more unlimited.
|
|
Effective July 6, 2026, the entire Genie family (Genie Spaces, Genie Code, and Genie One) moved to a pay-as-you-go pricing model. Instead of being bundled into the platform, each Genie session is now billed in DBUs that reflect the LLM models and agents powering it.
The softener: every identified user gets 150 DBUs of free LLM usage per month, roughly $10.50 in US East, or about 80 to 100 Genie questions and 20 to 30 Genie Code sessions. Databricks is not adding seat-based fees. Go over the free tier, and the overage is billed by consumption.
It's a consumption model, and consumption models reward the disciplined and punish the careless. The teams that get surprised will be the ones running Genie through automation, because the free tier has a very specific hole in it.
|
|
|
💸 The real cost math
|
What's billed, and what isn't
The free 150 DBUs cover Genie LLM usage only. That's the part people fixate on, but it's not the whole bill. Two costs sit outside the free tier and are easy to forget:
First, Genie compute is separate. The SQL Serverless warehouse that actually runs the query Genie generates is billed on its own, as it always has been. The new pricing is about the intelligence layer, not the query execution underneath it.
Second, and this is the one that bites, the free tier applies to identified users only, not service principals. Every Genie call made by a service principal is billed from the first DBU. If you've wired Genie into a scheduled job, an app backend, or an agent pipeline, that traffic has no free allowance at all.
|
|
🔄 Free vs. billed from DBU #1
|
|
Gets the free 150 DBUs
Identified human users
- Analysts asking Genie questions
- Interactive Genie Code sessions
- ~80–100 questions / user / month
- Resets monthly, per user
|
|
Billed from DBU #1
Automation & overage
- Service principals (jobs, apps, agents)
- Any user usage past 150 DBUs
- Genie compute (SQL Serverless)
- Heavy Genie Code / One workloads
|
|
|
150
Free DBUs / user
|
~$10.50
Free value / mo
|
$0
Seat fees
|
80–100
Free Q&A / mo
|
|
|
🛡️ Not a coincidence: the budget controls arrive
|
|
The same month Genie switched on the meter, Databricks put Unity AI Gateway budgets into Public Preview across AWS, Azure, and Google Cloud. That timing is the whole story: they turned on consumption billing and, in the same breath, shipped the guardrails to keep it from running away.
AI Gateway budgets let admins define spending thresholds on requests flowing through the Gateway and Genie. You get shared or per-user limits, with configurable actions when a threshold is hit: email alerts, or hard-blocking further usage. Pair that with per-Genie-space budgets and you can cap the exact automation traffic that has no free tier.
Do this in your first billing cycle
- Set an AI Gateway budget with an alert (not block) to learn your baseline
- Add a hard cap on any service-principal Genie traffic
- Watch DBU consumption in system tables before trusting a fixed limit
|
|
|
🔭 What this means for your team
|
|
1. Most human users won't notice. 80 to 100 questions a month covers the typical analyst. If your usage is interactive and human, the free tier likely absorbs it, so don't over-react.
|
|
2. Audit every service principal calling Genie. This is where surprise bills come from. Automation has zero free allowance, so inventory it, then cap it.
|
|
3. Don't forget the compute line. Genie LLM cost and Genie SQL Serverless cost are two separate meters. Budget for both.
|
|
4. Consumption pricing is a signal. Databricks is treating Genie as a metered product, not a loss-leader. That's a sign it's confident enough in adoption to charge for it directly.
|
|
|
|
|
|
|
|
Outside the pricing news, a few things worth catching up on:
|
|
RENAME
Asset Bundles are now Declarative Automation Bundles
Databricks renamed Asset Bundles to Declarative Automation Bundles back in March (because "assets" was overloaded), and a wave of tutorials put it back on the radar this week. The rename is non-breaking: the CLI command and your existing config stay the same.
|
|
|
PUBLIC PREVIEW
Genie comes to Slack
The Databricks Genie app for Slack has been in Public Preview since June. Mention @Genie in a channel and get answers without leaving Slack. Remember, those calls now count against the new pricing.
|
|
|
RELEASES · SDK
SDK refresh: workspace grants and job-trigger conditions
databricks-sdk-go v0.153.0 adds workspace-grant listing methods and a SQL condition field for job triggers. The Java SDK v0.124.0 shipped the same day. Both carry breaking changes (removed catalog secret fields), so read the notes before you bump.
|
|
|
ARCHITECTURE
Lakebase and LTAP blur operational and analytical
Databricks is pushing LTAP, built on Lakebase (serverless Postgres on object storage), to keep one copy of data for transactional and analytical work and skip the ETL hop. A bid for the operational-database seat, not just the warehouse.
|
|
|
|
|
Two things from our side this week. First, FlutterBricks now reads in five more languages: Japanese, Korean, German, French, and Brazilian Portuguese, each on its own URL. Switch with the globe icon in the top bar. Second, this week's digest breaks down the Genie pricing change with worked cost examples, plus the full July release list. Full breakdown at brickster.ai/digest.
|
|
🤖 Trying to model your Genie bill?
The Brickster Assistant searches our full archive and answers with citations.
|
|
|
Enjoying Brickster Weekly?
Forward this to a colleague who works with Databricks.
Share this issue ↗
|
|
