Since we announced the Public Preview of Lakebase in the summer, thousands of Databricks customers have been building Data Intelligent Applications on top of Lakebase, using it to power application data serving, feature stores, and agent memory, while keeping that data closely aligned with analytics and machine learning workflows.
As we approach the end of the year, we’re thrilled to release an exciting new set of improvements:
- Autoscaling that dynamically adjusts compute based on load
- Scale to zero, allowing compute to shut down when idle and resume automatically in hundreds of milliseconds
- Instant provisioning to create new database instances in seconds
- Instant database branching, enabling git-like workflows with isolated, copy-on-write environments for development, testing, and staging
- Automated backups and point-in-time recovery for fast restore and safer operations
- Postgres 17, alongside continued Postgres 16 support
- Increased storage capacity up to 8TB for larger production workloads
- A new Lakebase UI that simplifies common workflows
These features represent a significant milestone in defining the lakebase category, a serverless database architecture that separates OLTP storage from compute. They are made possible by combining the serverless Postgres and storage technology from our Neon acquisition with Databricks' enterprise-grade, multi-cloud infrastructure.
Autoscaling for dynamic application workloads
Modern application workloads rarely follow predictable traffic patterns. User activity fluctuates throughout the day, background jobs generate bursts of writes, and agent-based systems can create sudden spikes in concurrency. Traditional operational databases require teams to manually plan for peak usage and adjust capacity, often resulting in overprovisioning and unnecessary complexity.
Since Lakebase builds on an architecture that separates the storage layer from the compute layer and allows independent scaling of the two, we are now releasing the compute autoscaling capability that can adjust compute dynamically based on active workload demand. When traffic increases, compute scales up to maintain performance. When activity slows, compute scales down. Idle databases suspend after a short period of inactivity and resume quickly when new queries arrive. Compute adjusts dynamically to match workload demand across both production and development environments.
The result is less time spent managing capacity and more time focused on application behavior.
Fast startup and instant provisioning
Creating a new database or resuming an idle one should not slow down development. With this update, new Lakebase databases are provisioned in seconds, and suspended instances resume quickly when traffic returns. This makes it easier to spin up environments on demand, iterate during development, and support workflows where databases are created and discarded frequently.
For teams building and testing applications, faster startup reduces friction and keeps iteration cycles tight, especially when combined with branching and autoscaling.
Branching for faster, safer iteration
Building and evolving production applications means constant change. Teams validate schema updates, debug complex issues, and run CI pipelines that depend on consistent views of data. Traditional database cloning struggles to keep up because full copies are slow, storage-heavy, and operationally risky.
The Lakebase storage service implements copy-on-write branching, and we now expose this functionality as database branching to our customers. Branches are instant, copy-on-write environments that remain isolated while sharing underlying storage. This makes it easy to spin up development, testing, and staging environments in seconds and iterate on application logic without touching production systems.
In practice, branching removes friction from the development lifecycle and helps teams move faster with confidence. (But testing in production is still not recommended!)
Automated backups and point-in-time recovery
Not every data issue is an outage. Sometimes the problem is subtler: a bug that quietly writes incorrect data over time, a schema change that behaves differently than expected, or a backfill script that touches more rows than intended. These issues often go unnoticed until teams need to rely on historical data for analysis, reporting, or downstream application behavior.
In traditional environments, recovering from scenarios like this can be painful. Teams are forced to reconstruct history by hand, replay logs, or stand up temporary systems just to recover a known good version of their data. That process is time-consuming, error-prone, and often requires deep database expertise.
Lakebase now makes these situations much easier to handle. With automated backups and point-in-time recovery, teams can restore a database to an exact moment in time within seconds. This enables application teams to quickly recover from data issues caused by application bugs or operational errors, without requiring manual replay or complex recovery workflows.
Supporting larger production workloads
Beyond recovery, production systems also need room to grow as data volumes increase. With this update, Lakebase increases its supported storage capacity to up to 8TB, a fourfold increase over previous limits, making it suitable for larger and more demanding application workloads.
Expanded Postgres version support
Lakebase now also supports Postgres 17, alongside continued support for Postgres 16. This gives teams access to the latest Postgres improvements while maintaining compatibility with existing applications.
Together, these updates make Lakebase a stronger foundation for running production-grade operational workloads on Databricks.
Simpler workflows with a new Lakebase UI
Lakebase now includes a refreshed new user interface designed to simplify everyday workflows. Creating databases, managing branches, and understanding capacity behavior is more straightforward, with better defaults and faster provisioning. This new UI is accessible in the App Launcher icon for the new Lakebase autoscaling offering. The previous Lakebase provisioned offering will appear in the UI in the coming weeks.
Adoption
As indicated earlier, thousands of Databricks customers have been building applications on top of Lakebase. Because Lakebase is fully integrated into the Databricks Data Intelligence Platform, operational data resides in the same foundation that supports analytics, AI, applications, and agentic workflows. Unity Catalog provides consistent governance, access control, auditing, and lineage. Databricks Apps and agent frameworks can utilize Lakebase to integrate real-time state with historical context, eliminating the need for ETL or replication.
For practitioners, this creates a unified environment where operational and analytical data remain aligned, without the need to juggle multiple systems to keep applications connected to intelligence.
Quoting two early adopters:
“Lakebase lets an agentic team quickly self-serve the data they need for their models, whether it’s historical claims or real-time transactions, and that’s really powerful.” — Dragon Sky, Chief Architect, Ensemble Health
“Lakebase gives us a durable, low-latency store for application state, so our data apps load quickly, refresh seamlessly, and even support shared page links between users.” — Bobby Muldoon, VP of Engineering, YipitData
What’s next for Lakebase
These new features are available today in AWS us-east-1, us-west-2, eu-west-1 and will be gradually rolled out to more regions in the coming weeks. Check out the product documentation to learn more and try out the latest capabilities.
This update represents a meaningful step forward for Lakebase. But we are not standing still. Expect a lot of exciting updates after the holidays next year!
Happy Holidays from the Lakebase team!
Instagram Engineering
Shopify Engineering