Google Cloud and Oracle Connect AI Agents to Governed Enterprise Data

Oct 27, 2026 631 views

Google and Oracle are turning database access into the AI story

Oracle databases run much of the enterprise machinery companies can’t afford to get wrong: supply chains, financial ledgers, and billing systems. The more significant question raised at Oracle AI World in Las Vegas is how AI agents will reach that operational data—and what they’ll be allowed to do with it.

Google Cloud and Oracle are expanding their partnership across AI infrastructure, enterprise data, and agentic AI. The announcement includes making Google TPUs available through Oracle Cloud Infrastructure and deepening the connection among Oracle AI Database@Google Cloud, Google’s Agentic Data Cloud, and Gemini. The pitch is that companies can pair Oracle’s applications and transactional records with BigQuery and Gemini to build workflows that act on current business data, rather than stopping at static reports.

That’s an ambitious promise. The practical test is whether agents can find the right data, respect existing permissions, and operate without adding a tangle of custom plumbing. Google and Oracle are building their case around those foundations.

The groundwork: database access, governance, and context

At Google Cloud Next ’26, the companies first announced a set of capabilities intended to support agentic AI on Oracle AI Database@Google Cloud. The pieces address a real production concern: model capability alone doesn’t give an agent safe, useful access to enterprise systems. It also needs a governed way into the database and enough business context to interpret what it finds.

  • Remote MCP Server with Data API is now generally available. The Remote Model Context Protocol (MCP) Server for Oracle workloads works with the IAM-governed ExecuteSQL Data API to give AI agents, Google Cloud services, and enterprise applications a standardized way to run SQL against Oracle databases. Credentials are resolved through Secret Manager, while Google Cloud IAM policies control access. Google says that removes the need for custom natural-language-to-SQL middleware, bastion VMs, and plaintext passwords.
  • Knowledge Catalog integration is in preview. It brings metadata discovery and governance for Oracle databases into the broader Google Agentic Data Cloud. Cataloging schemas and business metadata can help organizations apply consistent access controls, trace data provenance, and give agents context for interpreting operational records. You can sign up for the feature.

Gemini moves closer to Oracle data and applications

A separate preview, the Gemini connector for Oracle AI Database@Google Cloud, lets business users and developers query their Oracle databases in natural language from the Gemini app. In the background, Gemini uses Knowledge Catalog to find relevant data and ground its responses in the governed Oracle information the user is permitted to access. If that authorization model works as described, it could make complex databases easier to query without loosening enterprise controls. You can sign up for the preview.

The integration runs in both directions: Oracle data and workflows can be brought into Gemini, while Gemini models are being added to Oracle’s application layer.

  • Oracle in Gemini is available today. It includes conversational analytics, cross-cloud database connectors, and domain-aware agents, including the Oracle AI Database Agent in Gemini. Google and Oracle cite more than 135 pre-built operational and developer skills, plus standardized MCP connectivity that avoids plaintext credentials. The companies say enterprises can use these capabilities to coordinate agents across JD Edwards, E-Business Suite, NetSuite, and Oracle Fusion without custom SQL or fragile middleware.
  • Gemini models in Oracle applications are being embedded in Oracle Fusion AI Agent Studio, Fusion Claw, Oracle Fusion Applications, and NetSuite. The aim is to combine Oracle’s database intelligence with Gemini’s reasoning so agents can carry out work in transactional systems.
  • Partner blueprints are meant to help customers put these workflows into practice. The companies are working with Argano, Automus, Calfus, HCL, IBM, Infosys, LTM, Pythian, TCS, and Tessell on pre-built operational skills, custom MCP connectors, and industry blueprints. See Argano, Automus, Calfus, HCL, IBM, Infosys, LTM, Pythian, TCS, and Tessell.

Three ways to connect Oracle data with BigQuery

There’s no single integration pattern that suits every workload. Latency needs, governance rules, and existing architecture all shape the choice. Google Cloud and Oracle are offering three approaches—shared lakehouse tables, continuous data streaming, and query federation—that customers can select or combine.

Shared Iceberg tables across the two lakehouses

Earlier this year, Google Cloud and Oracle connected Google’s borderless Lakehouse with Oracle AI Lakehouse using the open Apache Iceberg standard. That first step let Google’s lakehouse read Iceberg tables created by Oracle AI Lakehouse across public cloud and on-premises environments, using catalogs and storage supported by both platforms.

The new preview makes the connection bidirectional: Oracle Access to Google Cloud borderless Lakehouse lets Oracle Autonomous AI Database and Google’s borderless Lakehouse, powered by BigQuery and Managed Spark, query shared Iceberg tables in Google Cloud Storage and perform write and DML operations on them. The stated benefit is that teams can work from shared tables without ETL or data replication.

The integration uses a Lakehouse runtime REST catalog connected to Oracle through DBMS_CATALOG. In principle, analytical teams using BigQuery and Gemini and operational teams using Oracle SQL can work with the same data. You can sign up for the preview.

Sub-second change streaming with OCI GoldenGate

For teams that want a continuously refreshed copy of Oracle operational data in BigQuery, native OCI GoldenGate integration streams change data capture (CDC) events from Oracle redo logs. The source says this avoids reading table blocks or adding overhead to transactional workloads. Events travel over Google Cloud private networks into BigQuery managed tables, with sub-second analytical freshness for Gemini, feature stores, and live business intelligence dashboards. The same integration can support migrations from on-premises systems to Oracle AI Database@Google Cloud.

Query live Oracle data from BigQuery

BigQuery query federation with Oracle is in preview and targets a different need: occasional cross-system analysis or a live view of operational records, without moving the data first. Using the EXTERNAL_QUERY function, data teams can issue native Oracle SQL through BigQuery queries, dashboards, and pipelines, then combine returned records with BigQuery datasets that may span petabytes.

This zero-copy approach could cut storage duplication and ETL upkeep for ad hoc lookups, intraday risk reports, and live fraud detection. The feature is in preview; organizations interested in signing up are directed to contact their Google Cloud account team.

Three ways to connect Oracle data to BigQuery

The choice of integration pattern comes down to how fresh the data needs to be and how much work the query should place on Oracle’s transactional system. A shared Apache Iceberg lakehouse offers real-time, zero-copy access through the Lakehouse runtime REST catalog using DBMS_CATALOG and Glue Federation. It lets BigQuery, Spark and Oracle SQL work across shared analytical storage, supporting multi-cloud interoperability without tying the data to one vendor. For continuously updated data, OCI GoldenGate streams changes from Oracle redo logs to BigQuery through the Storage Write API. The stated latency is below one second, with no impact on Oracle OLTP. That makes it a fit for reflecting operational activity in BigQuery, feeding live AI feature stores, and keeping machine-learning and BI pipelines current. BigQuery Query Federation, currently in preview, takes a different route: EXTERNAL_QUERY through the BigQuery Connection API runs queries in place against Oracle. Results are live on demand, but Oracle’s workload impact varies with the query. This suits occasional lookups, cross-estate joins and regulatory reports; it’s a less predictable choice for sustained analytical traffic.

More regions, and a compliance plan

The service has added two locations: europe-west12-a in Turin, Italy, and northamerica-south1-b in Querétaro, Mexico. They join existing hubs in Milan, Iowa, São Paulo, Tokyo, Sydney and Mumbai. The stated benefit is lower-latency access and local data residency. The [regions and zones list](https://docs.cloud.google.com/oracle/database/docs/regions-and-zones#exadata) has the full details. At Oracle AI World, Oracle also announced plans to pursue FedRAMP High certification for Oracle AI Database@Google Cloud. The proposal would let US government and public-sector customers run Oracle Exadata Database Service on dedicated infrastructure within Google Cloud Assured Workloads. A FedRAMP High-authorized environment is intended to support sensitive, unclassified data while agencies colocate Oracle workloads with existing applications and Gemini. The pitch includes low-latency access to live Oracle data, unified billing and procurement through Google Cloud Marketplace or certified partners, all within agency compliance boundaries. Certification is still a plan, so the announcement describes a direction rather than a completed approval.

AI ambitions meet deployment choices

The broader argument is that Oracle’s transactional database and Google’s Agentic Data Cloud can turn enterprise data into an active input for AI agents. Iceberg sharing, GoldenGate streaming, BigQuery federation and Gemini access to governed Oracle data offer distinct ways to connect that data. The building blocks are there; whether they support autonomous agents at scale will depend on how well organizations govern access and match each workload to the right integration pattern. Oracle AI Database@Google Cloud is available through the [Google Cloud Marketplace](https://console.cloud.google.com/marketplace/product/oracle/oracle-database-at-google-cloud). Customers can deploy using existing Google Cloud commitments and receive unified billing and Oracle Support Rewards. More information is available in the [service overview](https://www.oracle.com/cloud/google/oracle-database-at-google-cloud/) and [Oracle Database@Google Cloud documentation](https://docs.oracle.com/en-us/iaas/Content/database-at-gcp/home.htm); Google Cloud account teams can also arrange an architecture briefing.
Source: Kiran Shenoy · cloud.google.com

Comments

Sign in to comment.
No comments yet. Be the first to comment.

Related Articles

Powering the agentic enterprise with latest innovations i...