Google Cloud is rolling out a series of updates for its Agentic Data Cloud. The borderless Lakehouse now runs queries on data in AWS and Azure without variable egress costs. Knowledge Catalog, AlloyDB, and BigQuery are also getting new features designed to make AI agents work more accurately and cost-effectively.
The updates primarily address one challenge: enterprise data is everywhere. That data is scattered across multiple clouds, on-premises systems, and SaaS platforms such as SAP, Salesforce, and Workday. Google wants agents to work with it directly, without moving the data first.
Google introduced the Agentic Data Cloud during Google Cloud Next 2026. The platform is designed to serve as a connecting layer between enterprise data and AI agents, featuring a Universal Context Engine that reduces hallucinations. Agents retrieve information via APIs, MCP, and A2A. However, integration with third-party applications is heavily dependent on partners and the availability of MCP servers.
Data stays where it is
The borderless Lakehouse is built on Apache Iceberg. Cross-cloud interconnect and caching for AWS and Azure are now generally available. Cross-cloud queries run over a private connection, with no variable egress costs. Retrieved data blocks are cached locally within Google Cloud. Additionally, there is catalog federation with Databricks Unity, Snowflake Horizon, and AWS Glue. Gemini also runs directly on the Lakehouse, allowing business users to query data in natural language.
AlloyDB can now query Iceberg tables directly, without ETL pipelines. This allows live operational data and historical Lakehouse data to be combined in a single PostgreSQL query. According to Google, this is up to seven times faster than Amazon’s comparable service for aggregation queries. Spanner Omni is now generally available.
Knowledge Catalog now features self-learning enrichment, currently in preview. This feature continuously extracts context from sources such as Google Workspace, GitHub, Jira, Confluence, and SharePoint. Domain knowledge is packaged into reusable context sets in the Open Knowledge Format. A new studio environment allows data teams to review suggested changes, including version control and rollback. Knowledge Catalog is also enabled by default in Gemini (preview). Semantic models from Looker, Databricks, Snowflake, dbt, SAP, and others are generally available.
Agent-scale infrastructure
AlloyDB Express is now in preview for developers. Provisioning takes less than two seconds; the database scales down to zero and runs on 1 to 2 vCPUs with 20 GB of storage. PostgreSQL for agents in AlloyDB supports more than 3 million queries per second. The Data Agent Kit, now generally available, brings Google’s data tools to VS Code, Claude Code, and Antigravity.
Google reports that BigQuery became over 40 percent faster and more than 50 percent cheaper per query over the past year. BigQuery Graph is generally available. The forecasting models TimesFM 3.0 and TabFM run directly in BigQuery in preview.