Oracle AI Database@Google Cloud at Oracle AI World 2026: Everything Announced

On this page

1. 01 · The Short Version
2. 02 · Everything Announced, by Status
3. 03 · For AI Agents: MCP Server and Knowledge Catalog
4. 04 · Gemini and Oracle, in Both Directions
5. 05 · Three Ways to Bring Oracle Data to BigQuery
6. 06 · The Big Picture: How It All Fits Together
7. 07 · Regions, FedRAMP High and How to Buy
8. 08 · How We Got Here, and What Comes Next
9. 09 · What to Do Now
10. 10 · Key Takeaways
11. 11 · Frequently Asked Questions

01 · The Short Version

Oracle AI World, Las Vegas · Google Cloud blog, 28 October 2026

From "Oracle runs in Google Cloud" to "Oracle works with Google's AI"

When Oracle Database@Google Cloud launched in 2024, the story was about location: Exadata hardware inside Google data centers, close to the applications that use it. The updates announced for Oracle AI World 2026 change the story to integration. Google's post, titled "Powering the agentic enterprise with latest innovations in Oracle AI Database@Google Cloud", is about letting AI agents, Gemini and BigQuery work with Oracle data directly, safely and without copying it around.

The partnership is also widening beyond databases. The same week, Oracle and Google announced that Google TPUs will be available on Oracle Cloud Infrastructure from 2027. Google describes the combined goal as connecting Oracle AI Database@Google Cloud with its Agentic Data Cloud and Gemini to "break down traditional data silos".

02 · Everything Announced, by Status

Answer: Generally available or available now: the Remote MCP Server, Oracle in Gemini and the Turin and Queretaro regions. In preview: the Gemini Connector, Knowledge Catalog integration, Oracle access to Google's borderless Lakehouse and BigQuery query federation. Planned: FedRAMP High.

Status matters more than headlines when you plan real projects. A generally available feature can go into production; a preview usually needs a sign-up, may change, and is best used for pilots. The status board below sorts every item in the announcement into four columns.

The left column is what you can build on today. The orange column is where most of the excitement is, but every item there needs a sign-up form or a conversation with your Google Cloud account team. The red column is a plan, not a certification.

03 · For AI Agents: MCP Server and Knowledge Catalog

Answer: The Remote MCP Server gives AI agents a standard, IAM-governed way to run SQL on Oracle databases through the ExecuteSQL Data API, with credentials resolved from Secret Manager. Knowledge Catalog, in preview, adds the metadata and governance agents need to find and understand the right data.

MCP, the Model Context Protocol, is an open standard for connecting AI agents to tools and data. Before this release, letting an agent query Oracle usually meant writing custom natural-language-to-SQL code, running it on a bastion VM and storing a database password somewhere. Google's post names all three as things the MCP Server removes. Access is controlled by Google Cloud IAM, credentials come from Secret Manager, and agents get one standard interface instead of bespoke glue code.

Knowledge Catalog integration complements it by automatically cataloging Oracle schemas and business metadata alongside the rest of Google's Agentic Data Cloud, with consistent access controls and provenance tracking. In practice that means an agent can discover which table holds "customer orders" and what the columns mean before it writes a query. Post 2 in this series walks through a full agent request step by step.

04 · Gemini and Oracle, in Both Directions

Answer: Oracle in Gemini brings Oracle data and 135+ prebuilt skills into Gemini today, including the new Oracle AI Database Agent. The Gemini Connector, in preview, lets users query Oracle databases in plain language inside the Gemini app. In the other direction, Gemini's reasoning is being embedded in Oracle Fusion and NetSuite.

The Gemini announcements run both ways. Oracle in Gemini adds conversational analytics, cross-cloud database connectors and domain-aware agents to Gemini, with multi-agent workflows across JD Edwards, E-Business Suite, NetSuite and Oracle Fusion, using MCP with zero plaintext credentials. The Gemini Connector grounds answers "strictly in the governed Oracle data each user is authorized to access", using Knowledge Catalog to find it.

Gemini in Oracle goes the other way: Gemini frontier reasoning embedded in Oracle Fusion AI Agent Studio, Fusion Claw, Oracle Fusion Applications and NetSuite. A group of partners, including Argano, Automus, Calfus, HCL, IBM, Infosys, LTM, Pythian, TCS and Tessell, is providing prebuilt skills, custom MCP connectors and industry blueprints. Post 3 covers which option fits which team.

05 · Three Ways to Bring Oracle Data to BigQuery

Answer: Google's post describes three options: a bi-directional Iceberg lakehouse where Oracle and BigQuery share the same tables (preview), sub-second change streaming with OCI GoldenGate into BigQuery, and BigQuery query federation with EXTERNAL_QUERY for live, zero-copy lookups (preview).

The lakehouse option uses Apache Iceberg tables in Google Cloud Storage, a REST catalog and Oracle's DBMS_CATALOG package, so Oracle Autonomous AI Database and BigQuery or Managed Spark can both query and write the same data. The GoldenGate option reads Oracle redo logs and streams changes into BigQuery managed tables through the Storage Write API. Federation runs native Oracle SQL from BigQuery through the BigQuery Connection API. Posts 4 and 5 compare them in depth and explain the streaming option step by step.

06 · The Big Picture: How It All Fits Together

Answer: Inside a Google Cloud region, users and agents reach Oracle AI Database through governed paths: the MCP Server for SQL, the Gemini Connector for natural language, Iceberg tables, GoldenGate and federation for analytics, all on Oracle-operated Exadata managed from an OCI parent region.

Taken one by one, the announcements can look like a list of unrelated features. Drawn together, they form a clear architecture: Oracle data stays on Exadata in the Google Cloud region, and every way of using it, by people, agents or analytics, goes through a governed path. The figure below shows that picture, with the status of each path marked.

Blue arrows are governed SQL through the MCP Server, green is natural language through the Gemini Connector, purple dashes are GoldenGate change streams and orange is shared Iceberg storage. The red dashed arrow is OCI's control plane, which still manages the Exadata infrastructure as it has since launch.

07 · Regions, FedRAMP High and How to Buy

Answer: Two new regions are live: Turin, Italy (europe-west12-a) and Queretaro, Mexico (northamerica-south1-b). Oracle plans to pursue FedRAMP High for Exadata Database Service on Dedicated Infrastructure in Google Cloud Assured Workloads. You buy through Google Cloud Marketplace using existing commitments and Oracle Support Rewards.

The new regions join existing hubs that Google's post names, including Milan, Iowa, Sao Paulo, Tokyo, Sydney and Mumbai. Turin and Milan in particular give Italian customers two locations in the same country, which is useful to explore for data residency and disaster recovery designs; confirm the supported pairings with Oracle and Google before you rely on them.

FedRAMP High is a plan, not a certification. When it arrives, US government and public sector organizations could run regulated workloads on Exadata dedicated infrastructure inside Google Cloud Assured Workloads, close to their applications and Gemini, and buy it through Google Cloud Marketplace or certified partners. Commercially, nothing changes for existing customers: Marketplace purchase, existing Google Cloud commitments, unified billing and Oracle Support Rewards. Post 6 covers region choice and buying in detail.

08 · How We Got Here, and What Comes Next

These announcements build on two years of steady expansion. The timeline below shows the path from general availability in 2024 to Oracle AI World 2026, and the items still to come.

The pattern is consistent: features are often shown first at Google Cloud Next, then reach general availability months later. That makes today's previews good candidates for pilots now and production plans later, once Google or Oracle confirms general availability.

09 · What to Do Now

1. DBAs: review which schemas agents should see, create least-privilege roles, and read the MCP Server documentation.
2. Data teams: pick one Oracle to BigQuery option for a pilot; GoldenGate CDC for live dashboards, federation for ad-hoc joins, Iceberg for shared lakehouse data.
3. Business and app teams: try Oracle in Gemini, and sign up for the Gemini Connector preview if natural-language access to Oracle data matters.
4. Architects: check whether Turin or Queretaro improves latency or residency, and plan pilots on preview features separately from production.
5. Public sector: track the FedRAMP High plan and talk to your Google Cloud account team about Assured Workloads timelines.

The 2026 updates are less about where Oracle runs and more about who can safely use its data: people, agents and analytics, all through governed paths.

10 · Key Takeaways

1. MCP Server is GA. AI agents can run governed SQL on Oracle through the ExecuteSQL Data API, with IAM and Secret Manager instead of stored passwords.
2. Gemini works with Oracle both ways. Oracle in Gemini is available today; the Gemini Connector is in preview; Gemini reasoning is coming into Fusion and NetSuite.
3. Three paths to BigQuery. Iceberg lakehouse (preview), GoldenGate CDC (sub-second) and query federation (preview) suit different needs.
4. More regions, a compliance plan. Turin and Queretaro are live; FedRAMP High is planned, not yet certified.
5. Buying stays simple. Google Cloud Marketplace, existing commitments, unified billing and Oracle Support Rewards.

11 · Frequently Asked Questions