Google Cloud expands agentic data tools for enterprises
Google Cloud expanded Agentic Data Cloud with cross-cloud Lakehouse tools, Knowledge Catalog features, Gemini functions and database updates; many remain in preview.

Why it matters
- November 18, 2026: Google Cloud announced updates across Agentic Data Cloud and related products.
- Delivery Hero talabat reported approximately 40% lower data-movement costs using borderless Lakehouse.
- Google Cloud claims BigQuery queries are more than 40% faster and processing costs more than 50% lower over the past year.
- Cross-cloud interconnect and caching for Amazon Web Services and Microsoft Azure are generally available.
- BigQuery Graph, Spanner Omni and Data Agent Kit are generally available; several agent and model features remain previews.
By James Calloway
What did Google Cloud announce?
Google Cloud said on November 18, 2026 that it had made BigQuery queries more than 40% faster over the past year while cutting query-processing costs by more than 50% through system optimisations. Google Cloud also introduced updates across Agentic Data Cloud and related data products for data management, databases, analytics and enterprise AI.
Most additions connect data held in public clouds, on-premises systems and software platforms. Google Cloud positions them as a way for AI agents and employees to query that information in natural language. That matters for enterprise deployments: agents need business data before they can produce useful answers.
The announcement supplies no independent benchmark, workload definition or cost methodology for the BigQuery figures. It also labels many of the new features as previews rather than generally available releases.
Takeaway: Google Cloud is selling a unified data layer for agents, but the evidence for its headline performance claims is incomplete.
How does the borderless Lakehouse connect clouds?
Google Cloud's borderless Lakehouse, a data warehouse built on the Apache Iceberg format, can now work directly with Gemini. Users can query records across different cloud environments in natural language.
The announcement makes the Lakehouse the connective layer for this release. Google Cloud does not provide an independent test of query accuracy, latency or scale.
Cross-cloud interconnect and caching for Amazon Web Services and Microsoft Azure are now generally available. Customers can query data across those clouds through a private connection, while the Lakehouse caches retrieved data blocks locally in Google Cloud.
Caching could avoid repeated cross-cloud retrieval for those blocks. Google Cloud did not provide transfer prices, cache limits or cross-cloud latency measurements.
Delivery Hero talabat provided the only quantified customer deployment in the announcement. Harish Ramesh, Staff Data Engineer at Delivery Hero talabat, said:
"By implementing the borderless Lakehouse that federates single-copy Apache Iceberg data from Amazon S3 tables to BigQuery, Delivery Hero's talabat reduced data-movement costs by approximately 40% while making streaming transactional events queryable across both clouds within minutes of production," said Harish Ramesh, Staff Data Engineer at Delivery Hero talabat.
That is a specific result, but not broad validation. The announcement gives no independent cost audit, baseline definition or event-volume detail.
Google Cloud also announced interoperability beyond the two major public clouds:
- Iceberg catalog federation with Databricks Unity, Snowflake Horizon and AWS Glue. Catalog federation links the metadata records that describe Iceberg tables.
- Zero-copy integrations with SAP, Workday and Salesforce, without requiring source data to be copied into Google Cloud.
The announcement did not give an availability label for either group, or list regions or service limits.
Google Cloud also added Lakehouse integrations for AlloyDB. Users can query Iceberg tables stored in cloud storage directly and combine live operational data with historical lakehouse data in a single PostgreSQL query.
The feature integrates with the Lakehouse runtime catalog, so Iceberg tables registered in AlloyDB also appear in Knowledge Catalog. Spanner Omni is now generally available for hybrid applications running across on-premises, edge and multi-cloud environments.
The source does not state whether the AlloyDB Lakehouse integration is a preview or generally available.
Takeaway: Google Cloud wants the Lakehouse to become the query layer across fragmented enterprise data, though only some components carry an explicit general-availability label.
What does Knowledge Catalog add for AI agents?
Google Cloud calls Knowledge Catalog 'the context engine for its AI agents,' the layer that supplies business information to AI systems. The Lakehouse integration is designed to index distributed and unstructured data in cloud storage. Google Cloud says this lets agents draw on a wider pool of business information.
The company publishes no retrieval-accuracy result and does not explain how the index handles conflicting records.
Knowledge Catalog gained three notable additions:
- Self-learning enrichment, in preview, learns from user intent and interactions. It mines Google Workspace, GitHub, Jira, Confluence and SharePoint to identify and fill context gaps.
- Reusable context sets. Google Cloud names the feature but does not explain its format, scope or lifecycle.
- A studio experience, also in preview, gives data teams a conversational interface to explore context and view data estates across cloud and software environments. Teams can review or roll back proposed enrichment changes.
Knowledge Catalog is now on by default in Gemini in preview. When customers connect BigQuery, AlloyDB, Cloud SQL or Spanner, Google Cloud says data questions will automatically draw on business context.
Because the default is itself in preview, enterprises should not treat current behavior or availability as settled. The announcement also omits access controls, audit logs and failure handling for self-learning enrichment.
Google Cloud also made semantic model support generally available for BigQuery, Looker, Databricks, Snowflake, dbt and SAP. Semantic models provide reusable business definitions for data relationships and metrics.
Metadata discovery and federation now extend to Oracle AI Database@Google Cloud, Snowflake, Databricks Unity Catalog and third-party catalog products including Atlan and Collibra. Those capabilities locate and link metadata across systems.
The announcement does not define the distinction between semantic models, metadata discovery and federation. It also gives no separate availability label for the metadata extensions.
Takeaway: Knowledge Catalog is the proposed context layer for agents, but its self-learning behavior and metadata controls remain largely undocumented.
What can Gemini and Data Agent Kit do?
Google Cloud introduced two new sets of Gemini functions, split between data and machine-learning engineers and business users.
- Engineer-focused functions can generate PySpark code, provide notebooks, train models and help fix pipeline issues.
- Business-focused functions can build operational reports from natural-language requests using BigQuery and Knowledge Catalog.
The announcement gives no model version, accuracy rate, availability label or validation mechanism for either set. Those omissions matter because the functions can sit close to production data workflows.
Data Agent Kit is now generally available. It integrates Google Data Cloud tools into development environments and command-line interfaces including VS Code, Claude Code and Antigravity.
The announcement gives no compatibility matrix, model choice or deployment details for those integrations. It also does not say whether customer code or prompts leave the managed environment.
Takeaway: The tools lower the barrier to building data workflows, but Google Cloud supplied no model-level evidence for their reliability.
Which database and analytics updates are ready?
AlloyDB Express is in preview as a version aimed at rapid testing and prototyping. It offers near-instant provisioning, pause-and-resume functions, private IP connectivity and a smaller infrastructure footprint for development work.
PostgreSQL for agents in AlloyDB is also in preview, aimed at production environments where large numbers of AI agents need to query live transactional data with workload isolation. Workload isolation separates agent workloads from other database activity, though the source does not describe its limits.
The broader analytics portfolio also received updates:
- BigQuery Graph is generally available, adding graph analytics to relational tables without moving data. Graph analytics examines relationships as well as row and column values.
- TimesFM 3.0 and TabFM from Google Research run natively in BigQuery in preview. They bring forecasting and predictive machine learning directly into SQL.
- Managed Service for Apache Spark has improved Spark 4.0 support and faster start-up times. Its premium tier combines upgraded hardware with the Lightning Engine, which Google Cloud enables by default.
Only BigQuery Graph carries an explicit general-availability label in this group. The announcement provides no accuracy figures for TimesFM 3.0 or TabFM, no service limits for AlloyDB Express and no performance baseline for the Spark changes.
Takeaway: Enterprises get more database and analytics choices, but AlloyDB Express, PostgreSQL for agents, TimesFM 3.0 and TabFM remain previews.
What should enterprises test next?
For enterprises, the release changes the operational question from whether agents can reach data to what happens when they do. Self-learning enrichment may alter an agent's context, while cross-cloud Lakehouse access may move queries beyond a customer's original cloud boundary.
Adopters should test retrieval accuracy, permissions, rollback behavior, cost and latency against their own data estates. They should also verify pricing and regional availability, which the announcement does not provide.
Google Cloud's central claim is that one context layer can connect data across clouds and enterprise applications. The current evidence is a product announcement and one customer account, not a broad benchmark.
Takeaway: The next proof point is independent production evidence on cross-cloud cost, latency, context accuracy and rollback; until then, preview features remain tests rather than settled infrastructure.
Source: itbrief.com.au
More from James Calloway
Show full bio
News editor covering people, launches and day-to-day practice in artificial intelligence and machine learning at AI In Context.
722 articles
Related articles
- Google Cloud Puts Agent Telemetry in Customer-Owned BigQuery Tables
- Google Cloud makes Pathways generally available on GKE
- Rubrik brings BigQuery backup to general availability, extends cover to Cloud Storage
- Google Cloud Doubled While Amazon Undercuts AI Compute Costs
- Google launches AlloyDB Express: sub-two-second Postgres for AI agents