Review of Databricks Data + AI Summit 2026

Open original

Databricks introduced Lakehouse//RT—powered by the new Reyden engine—to deliver millisecond analytics directly on Delta Lake, removing the need for separate serving databases. The summit also revealed LTAP, which merges OLTP and OLAP on a single data copy, and Lakebase, a PostgreSQL-compatible engine, signaling a push to unify transactional, analytical, and real-time workloads.

  • Lakehouse//RT (via the Reyden engine) claims 10–100ms query latency and ~12,000 QPS directly on Delta Lake, aiming to eliminate separate serving databases like ClickHouse or Pinot.
  • LTAP merges OLTP and OLAP on a single data copy, and Lakebase adds a Postgres-compatible engine, pushing toward unifying transactional and analytical workloads.
  • Most of the new AI-agent tools (Genie Ontology, ZeroOps, Omnigent, Unity AI Gateway) are dismissed as undifferentiated experiments, with Databricks' real edge seen in core infrastructure, not agent frameworks.
  • The bigger bet is architectural: killing data copies, sync pipelines, and duplicated security rules rather than just raw speed gains.

Databricks’ Lakehouse//RT aims to serve real-time applications directly from Delta Lake, cutting out separate serving databases like ClickHouse or Pinot. Powered by a new execution engine called Reyden, it promises 10–100 ms query latencies, up to 16× speedups in preview, and sustained throughput of around 12,000 queries per second. Instead of spinning up full warehouse clusters, it auto-scales nodes to match concurrent demand, all while reusing Delta Lake storage and Unity Catalog governance. Alongside Lakehouse//RT, Databricks announced LTAP—an OLAP+OLTP layer on a single data copy—which tackles transactional and analytical workloads together. The analyst dismissed most of the AI-tool releases (Genie Ontology, ZeroOps, Omnigent, Unity AI Gateway) as experiments without clear differentiation. He believes Databricks still leads with disruptive core technology rather than flashy agent frameworks. The writer flags two open questions: is Reyden simply the engine, or does it bundle with Lakehouse//RT as a standalone real-time data warehouse? Time will tell, but the bold bet is on eliminating data copies, sync pipelines, and duplicated security rules. That architectural shift, more than raw speed, could reshape how companies build dashboards, SaaS apps and AI-driven services.

Ask about this article