LTAP, short for Lake Transactional/Analytical Processing, unifies transactional and analytical data by storing Postgres-native transactional data directly in Delta and Iceberg formats, eliminating ETL pipelines and enabling a single copy to serve both OLTP and OLAP.

LTAP builds on Lakebase, Databricks’ serverless PostgreSQL storage on lake storage, using safekeepers, page servers, and caches to deliver reliable, fast transactional processing directly from inexpensive object storage.

By storing transactions in Apache Iceberg, LTAP supports OLTP and OLAP from one copy, avoiding CDC and simplifying the data stack under the motto: One data. Zero compromises. Zero copies.

Databricks argues this architectural shift is essential for agents to move faster by removing bottlenecks from traditional pipelines and multiple specialized systems.

Analysts see the agentic AI framing as a differentiator, emphasizing live governance, retrieval, and write-back in a unified workflow, while noting questions about latency, reliability, and true single-copy sharing.

Databricks introduced two products, Lakehouse//RT and LTAP, to collapse the operational and analytical data split and provide real-time, low-latency access for AI agents.

CEO Ali Ghodsi denies fundraising rumors, stating Databricks is not pursuing a new round and is focused on agents and a unified data architecture rather than IPO timing or valuation.

Databricks positions LTAP as solving the long-standing 40-year divide between transactional and analytical databases, aiming to be a foundational enterprise AI architecture rather than chasing mere integration.

The architecture targets AI agents as primary users, with roughly 80% of databases on its platform now created by agents, signaling a shift toward agent-driven data management.

LTAP arose from agents’ need for instant data access and rapid experimentation, driving the Lakebase–Lakehouse convergence and the elimination of traditional ETL-like pipelines.

Databricks frames its strategy as an architectural replacement, advocating the removal of data management fragmentation rather than simply stitching systems together.

Market context shows consolidation away from separate operational stores toward a unified, live-data governance model with reduced data copies to empower AI agents.