Databricks serverless micro apps are an announced but unreleased runtime for Databricks Apps: each app runs in its own lightweight virtual machine, starts when needed and scales down to zero when idle. Databricks announced them at Data + AI Summit 2026 and set them out in a blog post dated June 16, alongside App Spaces and Genie App Builder. The post says all three are "coming to Databricks Apps, with private previews coming soon." Three months later, that sentence is still the entire published status on the pages we checked: as of September 16, 2026, micro apps appear on none of the 44 Databricks Apps docs pages, in none of the AWS platform release notes from June through September, and nowhere on the Databricks Apps product page or the Azure what's-coming page. Nothing published lets you create one today, and none of the pages we checked says when that changes.
Key facts
- Status: announced June 16, 2026; "private previews coming soon", with no date, no sign-up link and no preview name. We record any status change on our Serverless Micro Apps roadmap tracker.
- Runtime: "a new micro VM-based runtime" in which each app can "start up fast when needed, scale all the way down when idle and stay isolated from other apps."
- Claim: a lightweight app "becomes usage-based, not reserved capacity-based." Today apps are "billed per hour of compute time while running, based on provisioned capacity."
- Not published: a price, a cold start time, resource limits, a framework list, regions, or a migration path for existing apps.
What Databricks said about serverless micro apps
Everything we can quote on micro apps comes from one blog post, "Enabling Governed Vibe Coding for Enterprise Apps on Databricks", dated June 16, 2026 and signed by Evan Pandya, Justin DeBrabant and Cong Xu; we keep it at our news permalink for the post, which links to the original. It names "three new key capabilities for Databricks Apps":
- App Spaces, "a new governance boundary to configure resources, access management, and security policies for groups of apps."
- Genie App Builder, "a purpose-built AI app authoring tool for Databricks with native awareness of your data assets, Unity Catalog semantics and workspace context." For the rest of the Genie family, see Genie One vs Genie Agents vs Genie Code Explained.
- Serverless micro apps: "A new class of Serverless Micro Apps that start up fast when needed and scale down to zero when idle, so organizations can support a broad portfolio of apps without the cost of always-on infrastructure."
Databricks' own sentence on how the three fit: "App Spaces establishes the governance layer before a single app gets built, Genie App Builder creates inside that layer, with full awareness of your data and workspace context, and microVM-based serverless micro apps make the whole model economically sustainable at scale." Micro apps are the runtime underneath the other two. The technical description is the two sentences in the key facts above, and the post gives nothing more.
What it would change
Today a Databricks app "runs as a containerized service on the Databricks serverless platform" and is "billed per hour of compute time while running, based on provisioned capacity." You choose an instance size when you create or edit the app; the docs list two and default to Medium.
| Size | CPU | Memory | Rate | Where it is listed |
|---|---|---|---|---|
| Medium (default) | Up to 2 vCPUs | 6 GB | 0.5 DBU per hour | Docs and pricing page |
| Large | Up to 4 vCPUs | 12 GB | 1 DBU per hour | Docs and pricing page |
| XLarge | Not documented | Not documented | 3 DBU per hour | Pricing page only |
The pricing page also lists an XLarge size the docs do not. Its worked example prices the DBU at $0.75 on the AWS Premium tier: "5 Medium apps for 30 days and 24 hours each day" is 1,800 DBUs and "a final charge of $1,350 for the entire month." Divide by five, our arithmetic not theirs, and in that example an always-on Medium app is $270 a month whether or not anyone opens it. There is no idle state: an app is "Running, Stopped, Deploying, or Crashed", and Running is billed.
That is the model the post argues against. "Using always-on, dedicated compute means paying for capacity that is often not in use," it says. "The economics of traditional app infrastructure are less favorable for smaller, more specialized apps," meaning departmental tools and line-of-business workflows that are "used in bursts and idle most of the time." The promised change is one sentence: "The practical outcome is that deploying a lightweight app on Databricks becomes usage-based, not reserved capacity-based." So the $270 is the number a micro app is meant to cut. By how much, nobody has said.
What Databricks has not published. No price or billing unit for a micro app. No cold start time. No word on whether today's Python and Node.js frameworks carry over. No regions, no resource limits, no maximum idle window, and no migration story for existing apps. And nothing on how scale to zero would interact with the documented rule that "Apps don't preserve in-memory state after restarts." Anything that fills those gaps is a guess.
What already exists that is close
Stopping the app yourself. A Stopped app "is not accessible and doesn't incur any costs. Databricks preserves the app's configuration and environment, so you can restart it without reconfiguring." The CLI has a stop command, databricks apps stop [NAME]; the app details page documents six tabs and does not document stopping. This is scale to zero by hand: free while off, unreachable while off, and someone has to restart it. In-memory state does not survive the restart, though anything written to a Lakebase database does: "Any schemas or tables created by an app remain even after you re-deploy or stop the app."
Horizontal scaling, which goes the other way. An app can already run "across multiple instances behind a single app URL for higher availability and concurrency." It is in Beta behind the Previews page, capped at 5 instances per app, and "Compute cost scales linearly with the number of instances." Azure will switch it on for compliance-profile workspaces in early October 2026. It scales out, not down; the floor is still one running instance.
Lakebase, the platform's scale-to-zero precedent. The same Azure page says "Lakebase Autoscaling adds autoscaling and scale-to-zero" and that "With elastic compute replacing fixed-size instances, most customers will see a reduction in compute costs." Same shape of claim, made for a database, and already rolling out. It says nothing about when Apps gets it.
How to follow it
Our Serverless Micro Apps tracker is where we record a status change. It stands at Announced with one source, the June 16 blog post. The App Spaces and Genie App Builder trackers stand at the same status on the same source, and all three are listed as sub-features on the parent Databricks Apps roadmap tracker, where a fourth sibling, Apps on Databricks Marketplace, has already moved to Public preview. Recent items mentioning Databricks Apps land hourly on the Databricks Apps topic hub. The trackers read public sources only; we do not see the private roadmap.
Three signals would make it real: a docs page carrying a preview notice like the horizontal scaling one, where "Workspace admins can control access to this feature from the Previews page"; a release-notes entry under Databricks Apps; or a micro app line on the pricing page. For the record, the only Apps entry in the September 2026 release notes, "Databricks Apps is now on by default for workspaces with the compliance security profile enabled", dated September 18 on a page that says releases are staged, is not that signal.
Sources
- Databricks blog, June 16, 2026: Enabling Governed Vibe Coding for Enterprise Apps on Databricks
- Databricks docs: Databricks Apps, key concepts and Configure compute resources for a Databricks app
- Databricks pricing: Databricks Apps pricing and the pricing calculator
- Databricks docs: Horizontal scaling for Databricks apps (Beta), CLI apps commands, view app details and Lakebase with Databricks Apps
- Databricks platform release notes, AWS: June, July, August and September 2026
- Azure Databricks: What's coming
- Databricks Apps product page and the docs sitemap, 5,930 URLs, for the absence checks
Every quotation, rate and status above was checked against the Databricks blog post, the 44 Databricks Apps docs pages, the CLI reference, the June to September 2026 AWS platform release notes, the Databricks Apps pricing page, the Databricks Apps product page, the docs sitemap and the Azure what's-coming page on September 16, 2026. Databricks can open a preview or publish a docs page without notice, so check the tracker before you plan around this. brickster.ai is an independent community project, not affiliated with, endorsed by, or sponsored by Databricks, Inc.
