Oracle’s Exadata platform is an engineered database system designed to run mission-critical workloads for BFSI and government agencies globally. It was a challenge for small enterprises to invest into its dedicated hardware until a year ago. But now the coin is flipped. Recently, Oracle announced the general availability of Oracle Exadata Database Service on Exascale infrastructure inside its Oracle AI Database@AWS platform, parallel to a renewed long-term collaboration agreement with Amazon Web Services. The platform has grown up from two AWS regions during the launch to 22 regions across Asia, Pacific, Europe, and the Americas.
If you think that this is a region count, my insight is it’s at the architectural level that will be the purpose of this article.
Exascale Pooled-Storage Architecture
Exascale pooled-storage architecture is built on RDMA (Remote Direct Memory Access) networking and a loosely coupled compute-storage design. These two have eliminated the mid-market Oracle customers where they have struggled between paying for capacity they couldn’t afford or walking away from Exadata’s engineering entirely in favor of a hyperscaler-native database that couldn’t match its performance ceiling.
Exascale breaks this sweat by decoupling storage management off the database servers onto a shared pool of intelligent storage nodes every tenant on the system can use. The database cluster need not manage its own storage directly. Rather Oracle AI Database 26ai sends data requests with the help of RDMA for OLTP workloads and Smart Scan for analytics, directly to the pooled storage servers, skipping the coordination layers where latency is expected.
The upshot is almost counterintuitive. It’s about provisioning a modest handful of compute cores, and your queries can still draw on the combined I/O muscle of hundreds of storage-server CPUs, because Exascale spreads every database’s file across the entire pool, regardless of its size.
Oracle’s own benchmarks put raw I/O latency at 17 microseconds — roughly 50 times faster than comparable storage paths on AWS or Azure. A separate figure circulating from the announcement, 165 microseconds, measures something different as full application-to-database round-trip latency over AWS’s high-performance networking layer, a longer journey that includes the network hop from application to database. Worth knowing the difference before quoting either number in a design review.
Cost Efficiency Promise with Same Reliability, Speed, and Performance
The cost consequence is the part that will actually move budgets. Entry-level Exadata capacity that ran roughly €10,000 a month under the old dedicated-quarterly model now starts near €330 a month pooled, reported industry estimates put the savings at around 70% versus the equivalent dedicated setup for smaller workloads.
The second architectural payoff is cloning as it got nearly free. Exascale uses Redirect-On-Write instead of Copy-On-Write, so a fresh clone references the source database’s existing blocks rather than duplicating them. A 10TB clone doesn’t cost 10TB until it diverges from production. For any organization running a dozen dev and test environments in parallel, this can be a different cost model entirely.
AWS Data Centers with OCI Child Sites
The natural question from anyone evaluating comes to, “Is Oracle just hosting on top of AWS?”
The architecture is its own thing and understanding it matters for anyone drawing a resilience diagram.
Oracle places what it calls OCI “child sites” physically inside AWS Availability Zones. The Exadata hardware sits in AWS’s data center, but it’s logically wired into an OCI region. A private, isolated network, the ODB network hosts the Exadata VM clusters inside that AZ, and application servers in AWS VPCs reach the database over a direct peered connection that never touches the public internet. OCI manages the link back to its parent region; AWS manages the link from the application VPC to the ODB network. From the customer’s seat, Exadata shows up as something provisioned in the AWS console with AWS Marketplace billing while the database itself runs on Oracle-managed Exadata iron at AWS-adjacent latency.
However, it isn’t Amazon RDS for Oracle, which only supports Standard Edition. Mission-critical workloads almost always need Enterprise Edition features like Real Application Clusters, and RAC is only available here or through bring-your-own-license on raw EC2, neither of the alternatives comes with the full Exadata hardware stack attached.
Strategic Advantage Through Commitment Delivered at Enterprise Scale
Alongside the Exascale rollout, Oracle and AWS signed a fresh long-term strategic agreement. The integration surface has filled in considerably over the past year. Zero-ETL connections now let Amazon Bedrock, QuickSight, and SageMaker query Oracle data directly, removing the pipeline-building step that used to sit between the database and AWS’s analytics stack. Oracle AI Vector Search, semantic search over meaning rather than keyword matching is now available on Exadata for AWS, positioning it as infrastructure for agentic and generative applications that need to reach live operational data.
The demand for multicloud database capacity is outrunning supply, and the Microsoft, Google, and AWS partnerships have unlocked pent-up demand from customers who wanted Oracle databases running on clouds other than Oracle’s own.
Multicloud Potential Redefined
This AWS release is one chapter of a three-cloud strategy that now spans AWS, Azure, and Google Cloud under the Oracle Database@ naming family. The pitch to enterprises is to run Oracle’s core database wherever your other workloads already sit, without re-architecting the applications built around it.
The old assumption in enterprise cloud strategy was eventual consolidation onto one hyperscaler. Oracle databases are heavy load bearing inside enterprise operations to migrate away from, while the AI and analytics layer feeding off that data increasingly lives on someone else’s infrastructure. Oracle and the hyperscalers have united to capitalize on their unique capabilities by leveraging one another. Exascale’s arrival on AWS is what makes that market reachable for the enterprises that, a year ago, couldn’t afford to be in it. And this is a true multicloud potential that any enterprise can bet on, to digitally transform or centralize the cloud infrastructure for promising outcomes and value impact.