Databricks used its annual summit to argue that the AI race is now an infrastructure race — and to show 30,000 attendees it’s positioned to win it.
Databricks closed its 2026 Data + AI Summit after four days, arguing that enterprise AI is now an infrastructure contest, not a model contest, and that its stack is built to win it. More than 30,000 data and AI professionals attended in person at Moscone Center in San Francisco from June 15-18, with tens of thousands more joining virtually, together representing 150-plus countries.
The four-day program included more than 800 breakout sessions, 25-plus hands-on training and certification courses, and a multi-day hackathon hosted with OpenAI to build agentic data apps for social impact. Evenings closed with Data After Hours at Oracle Park, a reminder that Databricks still treats the summit as much as a recruiting and brand event as a product launch.
The headline news was Genie One, an agentic coworker that helps business teams automate and orchestrate work across structured or unstructured, analytical or operational data, inside or outside the platform. Where the original Genie answered questions about data, Genie One is meant to take action: drafting documents, generating reports, scheduling work, and orchestrating tasks on its own.
Underpinning it is Genie Ontology, a self-improving context layer that continuously learns the business from internal and external data, AI tools, and connected workplace apps, intended to ground answers in governed reality rather than guesswork. Databricks has cited internal benchmarks showing query accuracy climbing from 32 percent to more than 90 percent against a leading coding agent — a number worth treating as directional rather than independently verified, since it’s the company’s own testing rather than a third-party benchmark.
Also Read: “Disheveled,” “Not Coherent” — The Bias Is in the Notes, Not the AI
The Genie family expanded accordingly. Genie Agents lets non-technical users build and share tailored conversations as reusable agents, Genie App Builder is a governed low-code environment for building custom apps on live data, Genie ZeroOps removes traditional database administration overhead, and Genie Code is an autonomous partner for building, debugging, and operating data and AI workflows. Pricing is shifting too: starting July 6, the Genie suite moves to pay-as-you-go, with each user getting roughly $10.50 worth of free usage per month before metered billing kicks in.
Governance got its own platform-level push. Unity AI Gateway now governs models, MCP services, agents, and skills through Unity Catalog while enforcing policies at runtime, with cost controls, smart routing, and a centralized inventory of approved tools. A companion capability, Lakewatch, captures end-to-end traces and helps investigate incidents — Databricks’ answer to what happens when an enterprise runs dozens of agents simultaneously.
On compliance, Databricks announced Automatic Identity Management for Entra ID, now generally available on AWS and Google Cloud with Okta support in public preview, alongside Context-Based Ingress for zero-trust access policies and a path to FedRAMP High on Azure later this summer.
On the infrastructure side, Lakebase — the serverless Postgres database built for agents — added decoupled compute and storage with instant, copy-on-write database branching, letting engineers safely reproduce production bugs without compliance risk. A new product, Lakehouse//RT, brought real-time analytics directly into the lakehouse; the healthcare software company PointClickCare reported running 35% faster on average and achieving tenfold faster queries after adopting it.
Perhaps the most consequential announcement for the wider AI ecosystem was OpenSharing, a new Linux Foundation open protocol for sharing AI agent skills across platforms and vendors — a direct answer to the lock-in problem enterprises face as they wire together agents from multiple AI providers.
Databricks also pushed into two new categories: it agreed to acquire security company Panther, further establishing a “security lakehouse” category, and launched CustomerLake, an agentic customer data platform built directly into the lakehouse with autonomous agents that assemble customer profiles and run campaigns — Databricks’ first real entry into martech proper, competing for the same budget as standalone CDPs.
Also Read: Are Banks Losing the Race Against Instant Fraud?
The guest list underscored how far Databricks has moved from its data-engineering roots: Greg Brockman of OpenAI and PepsiCo’s global chief data and AI officer, Magesh Bagavathi, joined the keynote stage, alongside Microsoft’s Satya Nadella in a pre-recorded fireside chat, with case studies from AstraZeneca, Mastercard, Mercedes-Benz, Nasdaq, Rivian, and Zillow, and more than 240 sponsors, including Anthropic, AWS, Google, Accenture, and Deloitte on the expo floor.
Taken together, the announcements reinforce the framing Databricks has pushed all year: the lakehouse, the semantic layer, the agent runtime and the governance layer converging into one platform built to run agents safely at enterprise scale. Whether that consolidation holds against rivals making the same case — Microsoft, Snowflake, the AI labs themselves — is the question the next twelve months will answer.
