Exadata Exascale on Google Cloud: What's New and Why It Matters

On this page

1. 01 · Why Google Cloud Teams Should Pay Attention
2. 02 · Oracle AI Database@Google Cloud in One Page
3. 03 · The Architecture in a Google Cloud Region
4. 04 · What's New: The Milestones So Far
5. 05 · Exascale With ExaDB-D: What the August 2026 Change Means
6. 06 · Before and After: Why It Matters
7. 07 · Integrations With BigQuery and Vertex AI
8. 08 · Getting Started on Google Cloud
9. 09 · Key Takeaways
10. 10 · Frequently Asked Questions

01 · Why Google Cloud Teams Should Pay Attention

Update brief · Exascale on Oracle AI Database@Google Cloud

"Exadata on Google Cloud was for the big ERP. What about everything else?"

When Oracle Database@Google Cloud reached general availability in September 2024, it brought Exadata into Google Cloud regions for the first time. The launch services were built on dedicated Exadata infrastructure, which made sense for large consolidated estates, but left smaller Oracle databases where they were: on Compute Engine VMs, on premises, or in another cloud.

That has changed. Exadata Database Service on Exascale Infrastructure (ExaDB-XS) is now part of the service, and in August 2026 Oracle also brought Exascale capabilities to Exadata Dedicated Infrastructure customers there. For a Google Cloud architect, the result is a range of Exadata entry points, from a small pay-per-use VM cluster to a full dedicated rack, all inside the same region as GKE, BigQuery and Vertex AI.

02 · Oracle AI Database@Google Cloud in One Page

Answer: Oracle AI Database@Google Cloud is the Oracle and Google service that places Oracle-operated Exadata infrastructure inside Google Cloud data centers. You order it through Google Cloud Marketplace, manage it from the Google Cloud console and APIs, and connect it to your VPC network. It was called Oracle Database@Google Cloud until October 2025.

The model is the same one Oracle uses with Azure and AWS. Exadata racks are installed in the Google Cloud facility and run by OCI as a child site linked to an OCI parent region. Google provides the region, the network fabric, the console experience and billing. Spend goes through your Google Cloud billing account and can count toward Google Cloud commitments, depending on your agreement.

03 · The Architecture in a Google Cloud Region

Answer: In a Google Cloud region, your VPC network provides a client subnet and a backup subnet for the Exadata VM cluster. The VM cluster runs in an Oracle-operated Exadata site in the same data center, either on Exascale pools or on dedicated infrastructure. OCI's parent region provides the control plane.

The figure below lays out the full regional topology. Your applications run in an application subnet on GKE or Compute Engine. The client subnet holds the VNICs and SCAN listener addresses of the VM cluster, which is what your connection strings resolve to. The backup subnet carries backup traffic. Inside the Oracle-operated site, ExaDB-XS and ExaDB-D clusters can sit side by side, both using Exascale storage since August 2026. Google Cloud services sit on the right, and the OCI parent region runs along the bottom.

The blue SQL*Net arrow never leaves the region: application to SCAN listener to VM cluster. The two black arrows into "Exascale storage" are the August 2026 change in one picture. ExaDB-XS always used Exascale storage, and dedicated clusters can now use it as well. The dashed red arrow from the bottom is OCI building and patching everything.

Networking: client and backup subnets

Plan two non-overlapping CIDR ranges per VM cluster, one for clients and one for backups, and make sure neither collides with existing subnets, peered VPCs or on-premises ranges reached through Cloud Interconnect. Firewall rules in the VPC must allow the listener port from the application subnets. If applications live in other projects, use Shared VPC or a hub design so that they still reach the client subnet without leaving the region.

Identity and operations

Google Cloud IAM controls who can create and change Exadata resources in a project. Metrics and logs flow into Cloud Monitoring and Cloud Logging, so the database shows up on the same dashboards and alerting policies as GKE and Compute Engine. Inside the database, users and roles are still Oracle users and roles, managed by your DBAs.

04 · What's New: The Milestones So Far

Answer: The service went GA in September 2024 with dedicated Exadata. In October 2025 it was renamed Oracle AI Database@Google Cloud and Exascale (ExaDB-XS) became generally available. In August 2026 Oracle added Exascale capabilities for ExaDB-D customers. Region and service coverage continue to expand.

The vertical timeline below lists the milestones that matter for architecture decisions. Each card shows what became possible at that point, and the tag on the right marks the change in short form. Read it as a maturity guide: the further down the list a capability appears, the newer it is, and the more likely it is that some regions do not have it yet or that runbooks and Terraform modules are still catching up.

The two Exascale entries are the reason this post exists. October 2025 gave Google Cloud a low-entry Exadata option. August 2026 brought Exascale storage features to dedicated infrastructure, which narrows the functional gap between the two Exadata services on Google Cloud.

05 · Exascale With ExaDB-D: What the August 2026 Change Means

Answer: Exascale with ExaDB-D means a customer on dedicated Exadata infrastructure can use Exascale storage capabilities, such as storage vaults and space-efficient snapshots and clones, without moving to the shared ExaDB-XS service. You keep dedicated hardware and gain Exascale's storage model.

Before this change, the choice on Google Cloud was binary. ExaDB-D gave you isolation and scale with the classic ASM-based storage model. ExaDB-XS gave you Exascale storage and pooled economics but on shared hardware. Customers with large dedicated estates could not easily use Exascale features such as fast thin clones for their test environments.

With Exascale available alongside ExaDB-D, a dedicated estate can adopt the Exascale storage model where it helps most, typically in non-production and in databases that need many copies. The production core can stay on dedicated infrastructure with its existing layout until there is a reason to change it.

Where it helps first

• Test data refreshes. Thin clones from snapshots replace long RMAN duplicates for each test cycle.
• Development sandboxes. Each developer can get a private copy of a large PDB without multiplying storage.
• Reporting copies. Point-in-time copies for month-end reporting without a second full dataset.

06 · Before and After: Why It Matters

Answer: Before Exascale, using Exadata on Google Cloud meant renting a full infrastructure shape and paying for capacity whether or not it was used. With Exascale, you pay for your slice of a shared pool, grow storage in GB, and create thin clones, while keeping Exadata performance features.

The comparison below makes the economic difference visible. Each square is a unit of Exadata capacity. On the left, a dedicated rack is entirely yours: the blue squares are what your databases use, and the red outlined squares are capacity you pay for but leave idle, which is common in the first year of a project. On the right, the same blue workload sits in a shared pool where other tenants use the grey squares, and only your slice appears on your bill. Underneath, each panel lists the practical consequences for planning and operations.

The point is not that dedicated is wrong. The green tick under "Full hardware isolation" on the left is a real requirement for some regulated workloads. The point is that teams no longer have to choose between Exadata and a sensible entry cost.

07 · Integrations With BigQuery and Vertex AI

Answer: Because the database sits in the same Google Cloud region as BigQuery and Vertex AI, data can move to analytics and AI services without crossing clouds. Oracle and Google describe integrations that let Oracle data feed BigQuery analytics and Vertex AI applications, alongside Oracle AI Database 26ai's own in-database AI features.

For many Google Cloud customers this is the real reason to keep Oracle data in the region rather than in another cloud. Analytics teams already work in BigQuery, and AI teams build on Vertex AI. Putting the system of record next to them removes a cross-cloud data pipeline and its egress charges.

Keep governance in mind. Copying Oracle data into analytics and AI services multiplies the places where sensitive data lives. Decide which columns leave the database, mask where needed and record the flows in your data catalogue.

08 · Getting Started on Google Cloud

If you are evaluating Exascale on Oracle AI Database@Google Cloud for the first time, this sequence avoids the most common delays:

1. Confirm the region and service. Check that the Google Cloud region next to your applications offers Oracle AI Database@Google Cloud and the specific service you need: ExaDB-XS, ExaDB-D with Exascale, or Autonomous.
2. Set up the commercial path. Arrange the Google Cloud Marketplace offer, link it to the right billing account, and agree BYOL or license-included terms.
3. Reserve network ranges. Allocate client and backup CIDR ranges that do not overlap with anything you peer or interconnect, and plan firewall rules for the listener port.
4. Grant IAM roles. Give the platform team the roles needed to create Exadata resources in the target project, and nobody else.
5. Build a pilot cluster. Start with a small ExaDB-XS VM cluster, load a representative PDB, and measure latency from the application subnet.
6. Test the Exascale workflows. Scale ECPUs, grow the vault, take a snapshot and create a thin clone, so your runbooks are proven before production.

On Google Cloud, Exascale is less about a new feature and more about a new audience: every Oracle database that was too small for a dedicated rack now has an Exadata option next to its application.

09 · Key Takeaways

1. Exascale is part of the Google Cloud service. ExaDB-XS became generally available on Oracle AI Database@Google Cloud in October 2025.
2. Dedicated customers gained Exascale too. In August 2026 Oracle added Exascale capabilities for ExaDB-D customers on Google Cloud.
3. Same region, same data center. Client and backup subnets in your VPC connect applications to an Oracle-operated Exadata site in the region.
4. Pay for your slice. Exascale replaces idle dedicated capacity with ECPU and GB billing on a shared pool.
5. Verify per region. Services and features roll out region by region; confirm before you design.

10 · Frequently Asked Questions