Google Cloud has announced a package of updates to what it calls Agentic Data Cloud, its set of databases, analytics and data catalog built for AI agents. Among the most relevant points for infrastructure teams are cross-cloud interconnection with AWS and Azure with no variable egress fees for its Apache Iceberg lakehouse, the new AlloyDB Express edition, Spanner Omni for on-premises and multicloud environments, and BigQuery Graph, now generally available. Sirish Chandrasekaran and Raj Pai, Google Cloud vice presidents, announced the news on November 4, 2026.
Google’s Agentic Data Cloud in 20 seconds
- The Iceberg lakehouse connects to AWS and Azure with no variable egress costs.
- AlloyDB Express spins up in under two seconds and scales to zero.
- Spanner Omni brings Spanner to on-premises, edge and multicloud environments.
- Several features remain in preview, and the figures are Google’s own.
The logic behind the announcement is that agents need fast access to data spread across clouds, transactional databases and analytical warehouses, and that copying that data from one place to another is expensive and slow. Google’s answer is more federation and fewer copies. According to the company, the data it manages in Apache Iceberg has more than tripled over the past year.
A Lakehouse That Doesn’t Force You to Move Data
The cached cross-cloud interconnection for AWS and Azure is now generally available: it offers a private connection with no variable egress fees and keeps a local copy in Google Cloud of the remote data blocks that are queried. The open Iceberg catalog federates with Databricks Unity, Snowflake Horizon and AWS Glue, and Google is announcing copy-free integrations with SAP, Workday and Salesforce. Gemini can work directly on top of that lakehouse so users can query data across multiple clouds in natural language.
Google cites the case of talabat, a Delivery Hero subsidiary, which according to its engineer Harish Ramesh cut its data-movement costs by around 40% by federating Iceberg tables from Amazon S3 with BigQuery, and was able to query transactional events from both clouds within minutes of them happening.
AlloyDB, Spanner and BigQuery
AlloyDB, Google’s PostgreSQL-compatible database, will be able to query Iceberg tables in cloud storage with no ETL processes or copies, combining live operational data with the lakehouse’s historical data in a single query. Google says it is up to seven times faster on aggregation queries than Amazon’s comparable PostgreSQL service, without detailing how it measured this. It is also introducing two preview variants:
- AlloyDB Express, an edition for developers and coding agents that provisions in under two seconds, pauses and resumes by scaling to zero, offers a private IP and built-in observability, and has a small footprint: 1 to 2 vCPUs and 20 GB of storage.
- PostgreSQL for agents on AlloyDB, aimed at production, which lets thousands of agents query live transactional data with full workload isolation, sub-millisecond I/O and more than 3 million queries per second, according to Google. Compute scales with active reasoning sessions and drops to zero at rest.
Spanner Omni, now generally available, brings Spanner’s consistency and scale to hybrid applications on-premises, at the edge and across multiple clouds. In analytics, Google says BigQuery has sped up its queries by more than 40% over the past year and cut processing costs by more than 50%. BigQuery Graph, generally available, enables petabyte-scale graph analysis on relational tables with no ETL and no separate graph database, and Google Research is contributing TimesFM 3.0 in preview, for SQL forecasting, and TabFM, for predictive machine learning inside BigQuery. The managed Apache Spark service now includes Spark 4.0 and improvements to its Lightning Engine.
The Catalog as Context for Agents
The other piece is Knowledge Catalog, which Google describes as its universal context engine. According to the company’s internal tests, it makes agents nearly four times more accurate, with up to 33% less latency and up to 75% fewer tokens. It now automatically indexes unstructured data, integrates with Gemini to resolve business terms and pick tables before generating SQL while respecting each user’s permissions, and extends metadata discovery to Oracle AI Database@Google Cloud, Snowflake, Databricks, Atlan and Collibra. For developers, Data Agent Kit, already available, brings Google’s data tools into VS Code, Claude Code and Antigravity via MCP.
As with any announcement of this kind, it’s worth separating what’s available from what’s promised: Spanner Omni, the cross-cloud interconnection, BigQuery Graph and Data Agent Kit are already generally available, while AlloyDB Express, PostgreSQL for agents and several Knowledge Catalog features remain in preview. The performance and cost improvements are Google’s own figures.
For more on how Google Cloud is cutting down on moving data between tools, see our coverage of Alteryx and Google Cloud bringing “in-place” analytics to BigQuery.
Frequently Asked Questions
What is Google’s Agentic Data Cloud?
The name Google Cloud uses for its set of databases, analytics and data catalog built so AI agents can access company data.
What is AlloyDB Express?
A new lightweight edition of AlloyDB, in preview, that provisions in under two seconds, scales to zero, and uses 1 to 2 vCPUs and 20 GB of storage.
Do you have to pay to move data out of AWS or Azure?
With Google’s new cross-cloud lakehouse interconnection there are no variable egress fees, according to the company, and the data blocks queried are cached in Google Cloud.
Sources:
This article is also available in Spanish: “Google prepara su nube de datos para agentes: AlloyDB Express, Spanner Omni e Iceberg,” on revistacloud.com.
