Databricks LTAP eliminates ETL for AI workloads
Key insights
- LTAP runs OLTP and OLAP on a single data copy in open Delta and Iceberg formats, eliminating ETL pipelines and replicas by design.
- Lakebase, the serverless Postgres foundation of LTAP, already handles 12 million database launches per day across thousands of customers.
- LTAP is listed as coming soon within Lakebase, meaning it is not yet generally available at the time of launch.
Why this matters
AI agents that read historical data, reason over it, and write application state in a single pass have always required bridging separate OLTP and OLAP systems, adding latency, ETL maintenance costs, and failure points. LTAP's architecture bet is that collapsing those two systems onto a single open-format storage layer under unified governance removes that overhead at the infrastructure level rather than the application level. If it delivers at scale, data teams building agent-first applications may no longer need to maintain distinct ETL pipelines, replica databases, and analytical lakehouses as separate infrastructure layers.
Summary
Databricks launched LTAP (Lake Transactional/Analytical Processing) on June 16, 2026, claiming to be the world's first platform to run OLTP and OLAP on a single data copy without ETL pipelines or replicas.
The architecture runs on Lakebase, Databricks' serverless Postgres layer handling 12 million database launches per day across thousands of customers including Block, Ensemble, Superhuman, and Zillow.
Essentially: (Databricks) collapses the wall between transactional databases and analytical lakehouses into one governed storage layer on open object storage.
- All operational, analytical, and streaming data stores in open Delta and Iceberg formats under a single governance model
- New Lakebase capabilities include cross-cloud disaster recovery, git-style branching, and autonomous database operations
- LTAP is listed as coming soon as part of Lakebase, not yet generally available
For AI agents that must read history, reason over it, and write state in a single pass, the two-system architecture was the direct bottleneck. LTAP is built to remove it.
Potential risks and opportunities
Risks
- Customers who have built ETL pipelines and dedicated OLTP systems on top of Databricks face migration costs before LTAP delivers its promised simplification.
- If LTAP underperforms dedicated OLTP systems on latency-sensitive workloads, Databricks risks losing operational database customers to purpose-built alternatives.
- The coming soon status means organizations planning AI agent architectures around LTAP today are building on an unshipped product, creating delivery timing risk.
Opportunities
- ETL and replication vendors (Fivetran, Airbyte, dbt Labs) face accelerating customer churn pressure as LTAP targets the pipeline layer their tools serve.
- Enterprises already on Lakebase, including Block, Ensemble, and Zillow, gain an early window to consolidate OLTP and analytics stacks before LTAP reaches general availability.
- Open table format stakeholders benefit as LTAP's Delta and Iceberg storage bet positions open formats as the default converged layer for both transactional and analytical workloads.
What we don't know yet
- Pricing for LTAP and Lakebase at scale is not disclosed, leaving cost comparisons against dedicated OLTP vendors such as Aurora or CockroachDB unknown.
- The coming soon availability window for LTAP is unspecified, so the gap between the announced architecture and production-ready access for existing Lakebase customers is unclear.
- No public benchmarks compare LTAP's single-storage approach against purpose-built OLTP systems under high-write, low-latency workloads.
Originally reported by databricks.com
Read the original article →Original headline: Databricks Launches LTAP: First Architecture to Unify Transactional and Analytical AI Workloads on a Single Open-Format Data Layer