Summit Season Cheat Sheet: What Snowflake and Databricks Will Announce in June

The April 2026 preview was a deliberate pre-Summit teaser. Snowflake's pattern of announcing preview 2-3 months before Summit, then announcing GA at Summit, is well-established. Expect a live demo showing governance policies enforced across Snowflake, Spark, Trino, and at least one other engine reading from Polaris-managed tables. This is the headline announcement.

Expect a "Polaris Partner Ecosystem" announcement with connectors and integrations for DuckDB, Dremio, Starburst/Trino, AWS Athena, Google BigQuery, and others. Snowflake will frame Polaris as the Switzerland of catalogs.

Iceberg v3 brings row-level lineage, deletion vectors, VARIANT type, geospatial types, and nanosecond timestamps. Snowflake will position themselves as having full v3 support on day one, ahead of competitors.

Snowflake will likely announce that Policy Exchange Standards are being contributed to the Apache Iceberg community or a new cross-vendor initiative. This is the "governance should be an open standard, not a vendor lock-in tool" move, a direct shot at Unity Catalog.

Cortex Agents (autonomous AI agents querying data, building visualizations, taking actions within Snowflake) and Cortex Fine-Tuning (fine-tune LLMs on enterprise data within Snowflake's security perimeter) were in preview through 2025. Expect both to GA.

Snowflake needs to close the AI governance gap with Unity Catalog. Expect an announcement around model governance, AI output auditing, LLM guardrails, and compliance tracking, all tied to the Polaris governance layer.

SPCS has been in extended preview. GA with NVIDIA GPU support (H100/H200) for model training and inference within Snowflake would make the compute story more competitive with Databricks.

Expect native support for reading S3 Tables as an external catalog source in Polaris/Snowflake. This turns AWS's boldest storage-layer move into an on-ramp rather than a threat, and lets Snowflake emphasize that S3 Tables lacks the governance layer that Polaris provides.

A feature that reads Unity Catalog metadata into Polaris, essentially making Databricks customers' catalogs portable. This would be the most aggressive competitive move of the Summit.

Notebooks supporting Python, SQL, Markdown with Cortex-powered code generation, collaborative editing, and Git integration.

Dynamic Tables improvements (sub-second latency), possibly a native Kafka-compatible streaming endpoint. Streaming has been Snowflake's gap relative to Databricks; expect investment here.

Especially for AI/ML workloads. Possibly a "Cortex credits" model that simplifies pricing for inference and fine-tuning.

This is the single most predictable announcement. Iceberg read has been GA; write has been in Public Preview. Azure GA evidence already appeared in Snowflake documentation as of April 2026. Expect full cross-cloud GA at the summit.

Currently in Public Preview: native Iceberg tables in Unity Catalog (not Delta-with-UniForm). GA at the summit would be the proof point that Databricks is genuinely format-neutral, not just format-tolerant.

Iceberg v3 entered Public Preview on Databricks on April 9, 2026, just two months before the summit. Full GA in two months is aggressive, but expect either GA announcement or a firm GA date. The three headline v3 features: Row Lineage, Deletion Vectors, and VARIANT.

The MCP (Model Context Protocol) Servers tab (Beta), MCP in Marketplace (Public Preview), and the Supervisor Agent (Beta) were all announced in early 2026. The summit is the natural venue to GA these. MCP is the hottest protocol in the AI agent ecosystem, and Databricks is positioning Unity Catalog as the enterprise governance layer for MCP.

Lakebase (the PostgreSQL-based operational database for AI workloads) went GA on AWS in early 2026, is in public preview on Azure. The summit is the natural venue for Azure GA and GCP preview. This is important for the AI agent story: agents need operational data access, and Lakebase provides it within the Unity Catalog governance perimeter.

Expect a major new capability for agent observability: tracking every tool call, model invocation, and data access across agent chains with full lineage in Unity Catalog. The analyst prediction of "MLOps and LLMOps merging into governed loops" is likely the framing.

Databricks already federates with AWS Glue, Hive Metastore, and Snowflake Horizon. The elephant in the room is Polaris. With Polaris at Apache TLP, Databricks needs a story. Expect either direct Polaris federation or a generic "federate with any Iceberg REST catalog" capability.

Databricks One (simplified business user interface) and Genie (natural language analytics) will likely GA with an "agentic Genie" that autonomously routes questions, chains queries, and reasons across data sources.

Read from Microsoft OneLake through Unity Catalog (currently in public preview) moves to GA. Write support to OneLake enters public preview. Deepens the Microsoft partnership and the Azure enterprise story.

Natural language to pipeline definitions, automatic optimization suggestions, intelligent monitoring. This extends the "AI everywhere" narrative from analytics into data engineering.

Databricks engineers are driving three Iceberg v4 proposals: adaptive metadata tree (single-file commits), relative path support, and a modernized statistics model. A keynote segment from Matei Zaharia or Reynold Xin on "the future of open table formats" with an Iceberg v4 timeline would let Databricks frame the convergence narrative on their terms.

UC OSS v0.5 is expected with Metric View support. A v1.0 at the summit would be symbolically important, signaling the open-source project is production-ready independent of Databricks commercial offering. This counters the "UC OSS is a second-class citizen" criticism.

Snowflake's April 2026 governance portability announcement forces a Databricks response. Watch for "portable governance" or "governance federation" language: the ability to export/import Unity Catalog RBAC policies in a standard format.

With Dario Amodei (Anthropic) speaking, expect at least a deep Anthropic partnership announcement. A DBRX-2 (multimodal MoE successor) is possible but Databricks has been more focused on the platform than proprietary model training.

At least 2-3 industry-specific packages (financial services, healthcare, retail/CPG) with pre-built data models, compliance templates, and AI agent configurations.

Summit season is two months away. Both companies are preparing announcements designed to win the catalog war, the AI governance race, and your architecture review simultaneously.

The value is not in what they announce. It is in knowing what to watch for when they do. Bring this cheat sheet to both Summits and score the predictions in real time.