Databricks has unified analytical and transactional processing on its Lakehouse, adding large-scale real-time query and agentic marketing capabilities. The updates were announced at its San Francisco Data + AI Summit, attracting over 30,000 on-site attendees.
The new LTAP (Lake Transactional/Analytical Processing) architecture unifies OLAP and OLTP via a shared storage layer. It integrates serverless Postgres-based Lakebase and Lakehouse under one unified governance, single source of truth and shared storage for operational, analytical and streaming data.
Unlike compute-layer OLAP/OLTP unification tools, LTAP converges data at the storage layer. It makes operational data instantly analytics-ready without extra pipelines, supporting independent, fully isolated scaling of transactional and analytical workloads with zero performance loss. Built on open standards, it works with all Postgres-compatible apps and Delta/Iceberg table readers.
“Complex data infrastructure has long been a business overhead. AI agents have rapidly expanded enterprise workforce capacity, yet legacy systems have become a major bottleneck — LTAP solves this challenge,” said Databricks Co-founder and CEO Ali Ghodsi.
LTAP is built on Lakebase, which brings Postgres transactions to Lakehouse object storage. Its compute-storage separation drastically cuts costs for running massive concurrent apps and AI agents. Launched last year, Lakebase serves thousands of customers including Block, Zillow and Superhuman, handling 12 million daily database launches.
Previously, Lakebase and Lakehouse maintained separate data copies and formats despite shared storage. LTAP eliminates redundant duplication, storing all data in Unity Catalog with standard Lakehouse open formats. Unified identity, permission and audit rules deliver consistent governance, letting AI agents operate on one trusted data source.
Transactional workloads retain full Postgres ACID compliance, while analytical workloads run at unlimited scale and concurrency. Zero data movement guarantees consistent operational and analytical results with no ETL required. Lakebase adds three key upgrades: cross-cloud/cross-region disaster recovery, Git-style branching/snapshots for safe production testing, and AI-powered autonomous database operations for monitoring, optimization and recovery.
Databricks launched Lakehouse//RT, a real-time Lakehouse driven by the new Reyden compute engine. It delivers millisecond-level latency for massive concurrent users and AI agents, running natively on governed Delta Lake and Iceberg tables under Unity Catalog. It eliminates separate permissions, proprietary formats and sync pipelines, removing the cost of standalone real-time serving layers.
“We have unified core data workloads on open infrastructure for a decade. Lakehouse//RT completes the stack, delivering the low-latency layer critical for modern analytics and AI agents, proving Lakehouses also lead in real-time workloads,” Ghodsi added.
Lakehouse//RT achieves 10ms latency for small datasets and sub-100ms latency for large-scale data. Benchmarks show sub-100ms latency at 12,000 QPS, with customer workloads gaining up to 16x performance over traditional real-time serving systems.
Databricks also debuted CustomerLake, a native Lakehouse-based agentic CDP. As defined by CDPI, CDPs unify multi-source customer data into unified profiles. CustomerLake consolidates data from websites, apps, CRM and POS systems, replacing one-off marketing campaigns with continuous agent-driven “infinity campaigns” to deliver billion-scale personalized real-time customer experiences.
Equipped with dedicated campaign and profile agents, an open partner ecosystem and native reverse ETL, it unifies and activates customer data across martech stacks. Following Lakewatch, it is Databricks’ second vertical enterprise product, adopting a value-based consumption model instead of traditional software licensing.
Availability: Lakehouse//RT is in beta; LTAP will launch soon as a Lakebase feature; CustomerLake is in private preview with early adopters including HP, Circle K, AB InBev and Santander’s Getnet. Notably, vector database firm Zilliz applies the “Lakebase” term differently for its Milvus Vector Lakebase.
Beijing Qianxing Jietong Technology Co., Ltd.
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!
Ваше сообщение должно содержать от 20 до 3000 символов!
Russian
