Databricks Data+AI Summit 2026 runs June 15-18 at the Moscone Center in San Francisco, with a free virtual experience available to anyone who registers. If you are building AI applications that sit on top of data infrastructure — which in 2026 means most serious AI applications — this is the conference where the Databricks stack evolves in public. This year has a specific reason to pay attention: Lakebase is now generally available and its implications for AI app developers are still not widely understood.

Here’s a builder-focused preview of what to watch.

Quick Facts

If you cannot travel, the free virtual experience gives you live keynotes plus a curated set of 8 technical breakout sessions (not the full 800+ session catalog — that’s an in-person-only benefit, with the full session library available on-demand after the event). For data/AI practitioners who prefer watching announcements live to reading summaries the next day, the virtual option is worth registering for now.

The Keynote Lineup

The keynote stage is the Databricks co-founder lineup: Ali Ghodsi (CEO), Matei Zaharia (CTO), Arsalan Tavakoli-Shiraji, and Reynold Xin, each expected to drive live product demos and announcements. This is where Databricks historically drops its major feature releases — not in a press release before the event, but live on stage.

External keynote guests include a pre-recorded fireside with Satya Nadella (Microsoft Chairman and CEO) and a session with Greg Brockman (OpenAI President, Chairman, and Co-Founder). The Nadella/Brockman pairing is notable given Microsoft’s recent MAI model rollout and its complicated relationship with OpenAI; don’t expect dramatic disclosures, but it frames the competitive context for everything else on stage.

More useful for builders: featured technical sessions from Anthropic, Cognition, CrewAI, LangChain, LlamaIndex, Lovable, OpenAI, and Replit, covering agentic architectures, foundation models, and real-world AI applications. This is the ecosystem showing up in one place to discuss how their tools sit on top of Databricks infrastructure.

Why Lakebase Is the Announcement to Watch

The most underreported development coming into this summit is Lakebase GA.

Lakebase is Databricks’ serverless PostgreSQL product. It launched in preview in June 2025 and reached general availability on AWS on February 3, 2026 (with an Azure GA following on March 3, 2026). At its simplest, it’s a managed Postgres database. At its architecturally interesting level, it unifies OLTP transactional workloads, analytical queries, and AI/vector workloads on a single governed data foundation — inside the same Unity Catalog governance layer you’re already using for your data lakehouse.

The practical implications for AI application builders:

Instant database branching. You can spin up isolated copies of your production database for dev, test, or CI environments in seconds via zero-copy clones and point-in-time recovery. For teams building AI apps with complex data dependencies, this is a meaningful workflow change.

No separate vector database. Lakebase supports pgvector alongside standard SQL, which means you can avoid maintaining a separate Pinecone or Weaviate instance if your data already lives in Databricks. This matters for RAG pipelines built on structured + unstructured data.

Governance without compromise. Unity Catalog governance applies to Lakebase the same way it applies to Delta Lake tables. If you’re building in a regulated industry (finance, healthcare) or for an enterprise that requires audit trails, this is a meaningful differentiator versus managing a standalone Postgres instance.

Since its June 2025 launch, Lakebase adoption has grown at more than twice the rate of Databricks’ data warehousing product, with thousands of companies running production workloads on it — a striking figure given how established the warehousing business is. Expect the June 15 keynote to feature Lakebase prominently, with at least one major announced capability or partnership that extends the product.

The Technical Track: What Else Is Worth Attending

The summit’s 800+ sessions span data engineering, MLOps, governance, analytics, agentic applications, and platform development. For builders, the highest-signal tracks:

MLflow. MLflow 3 shipped in June 2025 with native GenAI tracing and observability built in from the start; more recent point releases have extended it further — multi-turn conversation evaluation arrived in MLflow 3.10 (February 2026) and the MLflow AI Gateway (also February 2026) added centralized, governed LLM routing and usage tracking. Summit sessions will cover practical patterns for production ML and LLM observability. If you’re not using MLflow for experiment tracking and deployment, the summit is a good forcing function to understand whether you should be.

Delta Lake and Apache Iceberg interoperability. The table format wars have moved from “pick a side” to “use both.” Databricks has been investing in Iceberg compatibility for Delta tables via UniForm, which writes Delta and Iceberg metadata simultaneously so the same tables can be read natively from Snowflake, BigQuery, or other Iceberg engines; expect further announcements on parity and migration tooling.

Databricks Apps. This is Databricks’ hosted application runtime for deploying Streamlit, Gradio, Dash, and custom Python/JS apps directly inside the Lakehouse, governed by Unity Catalog. It reached general availability in June 2025 and adoption has reportedly accelerated quickly since; sessions here will address whether it’s ready for production AI app deployments.

Unity Catalog for AI governance. With enterprise AI governance becoming a compliance requirement rather than a best practice, Unity Catalog’s AI governance features — model registration, lineage, and access control for models and model-serving endpoints — are increasingly the reason enterprises standardize on Databricks.

The Agent Infrastructure Angle

Databricks is not an agent framework company. But it is the data layer that most enterprise AI agents run on. The presence of Anthropic, CrewAI, LangChain, and LlamaIndex in the featured sessions signals that the summit is explicitly positioning Databricks as the data foundation for agentic applications, not just batch analytics.

What to watch for in these sessions: patterns for giving agents access to governed data without bypassing Unity Catalog controls; patterns for agent memory and state management on Lakebase; and integration points between Databricks’ inference tables (automatic logging of model serving inputs/outputs to a Unity Catalog Delta table) and agent observability frameworks.

How to Make the Most of It

If you’re attending in-person: The networking events are notable. There’s a Welcome Reception on the evening of June 16, and a Data After Hours party on June 17 at Oracle Park headlined by The Chainsmokers, for whatever that’s worth to you. More usefully, the in-person community skews heavily toward data engineers, ML engineers, and platform architects — a different slice of the technical community than you’d find at, say, a pure AI developer conference.

If you’re watching virtually: Register in advance rather than hunting for recordings after the fact. The free virtual experience runs June 16-17 with live keynotes and a curated set of breakout sessions; the full session catalog is only available on-demand after the event closes.

The sessions to block first: Ghodsi’s keynote for major product announcements; any Lakebase deep-dive session for architectural detail; the Anthropic or LangChain session for the agent-on-data-infrastructure angle; and the Delta Lake/Iceberg interoperability session if table format decisions are live questions in your stack.

What We’re Writing After the Summit

If the Lakebase session produces architectural details that change how AI app builders should think about data persistence, we’ll cover that. If any of the framework integrations (CrewAI, LangChain, LlamaIndex on Databricks) surface patterns worth documenting, we’ll document them.

The summit runs June 15-18. We’ll have post-summit analysis up the week of June 22.

This article is based on pre-summit research. Session details and keynote announcements are subject to change. Registration and agenda information accurate as of June 8, 2026.