从Databricks数据与AI峰会2026中获得的要点

North AmericaSoftware Date17 June 2026 US Software Takeaways from Databricks Data + AI Summit Brad ZelnickResearch Analyst+1-212-250-8563 We attended Databricks' 2026 Data + AI Summit in San Francisco including anInvestor Session hosted by the company with presentations from CEO Ali Ghodsi,CFO Dave Conte, and other members of the management team. We have longviewed Databricks as a leadingmulti-cloud data and AI platform, and come awayfrom this year’s DAIS with that view reinforced as the company continues pushingthe innovation frontier particularly into agentic workflows. To us, the biggesttakeaway was Databricks’ ambition to become the agentic system of record for Bhavin Shah, CFAResearch Analyst+1-212-250-6775 Daniel KnauffResearch Associate Yash KejriwalResearch Associate What made the event particularly compelling was the degree to which Databricksis building for this moment across multiple layers of the stack, evolving in supportof agentic applications and taking a holistic platform approach. In our view, thecompany is attacking the core reasons enterprise AI has yet to fully translatemodel intelligence into broad productivity gains, namely the lack of high-qualityenterprisecontext,governed access,cost control,and low-latency datainfrastructure required for agents to operate in production. As Mr. Ghodsi put it, While on the surface the company's vision is very compelling and suited to themoment, we think underlying momentum in the financials is a testament to howproduct innovation from Databricks is translating to customer success, withrevenue of $4.1bn+ exiting FY26 (>55% y/y) and expectations for a $6.9bn runrate exiting F1H27 (representing 80%+ y/y overall and 65%+ y/y core growth).The acceleration at scale also supports our broader view that data and analyticsplatforms are key beneficiaries of the AI theme as customers continue migratingdata estates to the cloud and into platforms like Databricks and Snowflake that Genie Ontology was the standout announcement, and in our view, the mostdifferentiated part of the product story.CEO Ali Ghodsi framed the core limitationof today's agents as the lack of a high quality map of enterprise context ratherthan shortcomings of underlying LLMs. Without this, agents are forced to searchacross a wide variety of sources in real time oftento only cover a small subset ofthe broader enterprise knowledge graph while burning time and tokens. Here,Genie Ontology was positioned as alearned, permissions aware enterpriseknowledge graph that connects to systems beyond just Databricks and uses what Lakehouse RT / Rayden was another important announcement particularly for anagentic world.Databricks argued that data architecture today commonly involveshaving separate real time instances with copies of the same underlying data toachieve sub-second response times. Not only does this call for operationalcomplexity, but management argued that existing engines across vendors oftenstruggle to sustain consistent low latency performance as the number of queriesper second (QPS) scales, with internal benchmarks showing latency spikes or An expanded Genie portfolio enables enterprise agentic workflows.Genie Onewas positioned as the front end experience for business users to reason acrossenterprise data, while Genie Agents, Genie Code, and Genie Zero Ops extend thatconceptinto agent creation,developer workflows,and autonomous dataoperations respectively. Genie Code more specifically was noted to be particularlycapable across data engineering as well as ML workflows, consistent withDatabricks'unique heritage across these areas and powered by whatmanagement referred to as the secret sauce of Genie Ontology. Unity AI Gateway Several infrastructure and ecosystem updates bolster platform value.WithinLakeflow, ZeroBus Ingest is a fully managed ingestion layer that managementdescribed as wire-compatible with Kafka and able to land streaming data directlyintothe lakehouse.LTAP was one of the more incremental infrastructure Processing architecture that enables analytical workloads on transactional datawithout traditional ETL or CDC pipelines. On the security side, the plannedPanther acquisition extends the Lakewatch push by adding an AI SOC platformwith 100+ out of the boxintegrations (see our full thoughts here). Finally,Databricks continued to lean into openness through Iceberg v3 support and Financialsappear robust all around.Disclosure at the investor sessionunderscored Databricks’ continued growth at significant scale, withcompanyrevenueincreasing from $2.6bn+ in FY25 to $4.1bn+ in the recently ended FY26,representing 55%+ y/y growth with an NRR of 140%+. Management expects thetotal revenue run-rate to exceed $6.9bn in F1H27, representing 80%+ y/y growthoverall, with the core businessexpected to grow 65%+ y/y (i.e. excluding pass-through token revenue which is new to the model). Growth here remains largelybroad-based across regions, products, and clouds and we note that the cor