Google Cloud rolled out updates to its Agentic Data Cloud on Nov. 4 that let companies query business information across Amazon Web Services and Microsoft Azure without relocating the original datasets, a move that helped one enterprise slash data-movement expenses by roughly 40%. The announcement includes expansions to the company's borderless Lakehouse architecture and Knowledge Catalog, designed to connect AI agents with distributed enterprise information while keeping source files in their current locations. Though the underlying data remains in place, retrieved blocks can be stored temporarily in Google Cloud.
The platform now supports querying information housed in AWS and Azure environments, alongside catalog federation and connections to supported software-as-a-service platforms. AlloyDB integrations enable developers to search live operational records next to historical Iceberg data, while AlloyDB Express entered preview as a PostgreSQL database that halts automatically during idle periods and can scale up to AlloyDB when application demands increase. Cross-cloud interconnect and caching for AWS and Azure reached general availability, and Knowledge Catalog is now active by default in Gemini in preview. Delivery Hero's talabat used the borderless Lakehouse to federate single-copy Apache Iceberg information from Amazon S3 tables to BigQuery, reducing data-movement costs by approximately 40% and making streaming transactional events searchable across both clouds within minutes of production.
According to Google Cloud, the extended Knowledge Catalog delivers "self-learning, reusable context sets, and a reimagined user experience," and the vendor's internal testing indicates the catalog boosts AI agent accuracy nearly fourfold. The company states that agents require business context to generate precise answers and execute suitable actions, and the new Agentic Data Cloud capabilities aim to link scattered enterprise information, anchor AI agents in business logic, and handle agent workloads at scale. Harish Ramesh, staff data engineer at Delivery Hero's talabat, said in Google's announcement that the borderless Lakehouse implementation let the company query federated data across clouds shortly after production events occurred.
Channel partners face practical hurdles when deploying these systems for customers. The announcement recommends partners map each client's information sources to supported integrations, catalogs, and networking choices before proposing a rollout, and distinguish which required features are generally available versus still in preview to confirm support commitments align with production needs. Cost reviews should account for query processing, networking, caching, catalog usage, and staff time to run the environment, and infrastructure assessments should separate hardware budget constraints from the cloud consumption fees that drive an agent deployment's recurring expenses. Leaving source data in its original location doesn't eliminate the obligation to model those charges.
Data access requires a business-level review beyond technical connectivity. Partners should collaborate with customer data owners to settle on definitions for metrics like revenue or customer churn, confirm the records agents will reference, and verify that permissions block access to restricted material—a technically sound connection can still yield an unusable result if the foundational definitions or information are incorrect. Before widening deployment, partners should test one workflow and measure answer quality, response time, and operating cost against the customer's current process, with upfront acceptance criteria giving both sides a clearer foundation for judging whether the deployment provides sufficient value to proceed. Clear ownership and service boundaries help customers grasp what ongoing support includes and help partners price the work realistically. These requirements could underpin assessment, integration, governance, and continuous optimization services, provided partners define who maintains business definitions, reviews access changes, investigates faulty answers, and monitors spending to make those services commercially viable. For organizations juggling siloed data across multiple platforms, the tension between federation convenience and governance complexity will shape whether cross-cloud architectures deliver sustainable operational gains or simply relocate bottlenecks from migration pipelines to permission audits and cost reconciliation.

