A common narrative among cloud database providers is that they are winning lots of migrations from Oracle Corp. By conventional wisdom, Oracle’s market should be shriveling.

Yet Oracle continues to show very strong growth numbers, with an organic annual constant dollar growth rate of 8%, including substantial increases from cloud databases. Furthermore, its Autonomous Database processes more queries per hour than rival services.

What gives? Although conventional wisdom is that enterprises are slowing their growth in cloud consumption, the answer is that the secular trend toward cloud transformation continues to be real and that, over time, the tide is lifting a lot of boats.

In the short term, the unleashing of ChatGPT has thrust artificial intelligence, and especially generative AI, into monopolizing the conversation this year. Beyond the headlines, generative AI is a journey, and we are only at the very beginning — just like with data mesh, which dominated the data conversation last year.

The common thread to all this is that it will take data, and it will also take continuing growth of the cloud footprint. That’s where Oracle’s cloud strategy gets interesting.

Oracle Cloud Infrastructure, did you say? When the conversation is about hyperscalers, Oracle is typically not the first name mentioned. Although Oracle has been an incumbent in the enterprise computing world for 30 to 40 years, it is a challenger in the cloud. Amazon Web Services Inc. and Microsoft Corp.’s Azure, followed by Google Cloud, tend to get mentioned first before OCI.

Oracle typically doesn’t get first mention, not just among hyperscalers, but on its home turf of databases, its Autonomous Database processes many times more queries per hour than Snowflake Inc., which typically tends to get the spotlight. OK, that encompasses both transaction processing and data warehousing, so Oracle casts a wider net. But specifically, that’s more than 9 billion queries per hour. Who knew?

Oracle has picked up on that fact and acknowledges in the cloud that it is playing the role of scrappy challenger. In the old days, never known as the low-cost provider, it claims that its Autonomous Data Warehouse will underprice Amazon Redshift. This one is a street brawl.

Moving to multicloud

Oracle, like fellow enterprise computing incumbents IBM Corp. and SAP SE, has pivoted from a unicentric view, where it was the center, toward a multipolar, or more specifically, multicloud world. Aside from small businesses, for most midsized to large enterprises multicloud is the rule, either by plan or default.

Oracle’s journey to OCI is clearly that of a second-generation starter that learned the lessons from those who came before. Technologically, it has started from, almost, a clean slate. We say “almost” because Oracle did have a first-generation cloud architecture that it subsequently deprecated with a second generation that reinvented cloud infrastructure from the ground up.

It starts with a flattened network topology that eliminates the hops that are inevitable in large, diversified cloud data centers, and exploits innovations such as the Remote Direct Memory Access protocol or RDMA, where distributed nodes on its cloud network can read or write from each other’s cache remotely without having to navigate the overhead of central processing units or operating systems. Though most hyperscalers tout the economies of commodity hardware, RDMA is not possible unless you literally own and specify the entire hardware stack.

From a practical standpoint, it allows the Oracle Database to run across multiple physical instances as if they were one virtual instance. And in conjunction with the flattened, high-bandwidth networking topology, it allows large portions of Oracle cloud data centers to run as if they were one supercomputer. That’s when it comes to compute. As for storage, it partners with WEKAIO Inc. to provide a high-bandwidth fabric for a high-performance computing bare-metal compute instance type.

Another piece of the puzzle is multicloud: Oracle realizes that OCI won’t be the only game in town for its customer base. In a move that we never would have expected, several years ago Oracle inked a deal with Microsoft to build direct network tie-ins between OCI and adjacent Azure data centers. The guiding notion was to bring interoperability through low-latency dedicated connections between the clouds.

At the time, we saw it as being an example of the coming together of the enemy of thy enemy. More to the point, joint customers, which typically use Microsoft 365 and Oracle databases, pushed the two into a marriage.

In most cases, these interconnects go very short distances. Given that all the usual suspects tend to locate their cloud data centers in close proximity to major markets, and therefore to each other, those interconnects are often in the same building or another building a short campus drive away.

A European customer that uses a number of Azure services found that using Oracle Exadata Database Service through the interconnect (where the control plane was in Azure and the Exadata instance in OCI) benchmarked 30% faster IOPS versus working with other Azure services. Right now, out of OCI’s 41 “regions” (we’ll get to that in a moment), 12 have Azure interconnects.

This brings us to reviewing what is Oracle’s multicloud strategy. In isolated cases, such as MySQL HeatWave on AWS, it involves implementing and hosting Oracle database services natively on foreign territory. But given the degree to which Oracle’s database cloud services are optimized for hardware, HeatWave on AWS is going to be the exception that proves the rule. In most cases, Oracle’s multicloud strategy rests with the kindness of strangers. Customers brought Oracle and Azure together, and Oracle expects that they will exert similar pressures with other hyperscalers.

Exhibit A would be AWS. When it comes to third-party cloud services, AWS has a long history of coopetition. By number of SKUs, AWS offers more fit-for-purpose database services than anyone else, with our estimate being 16 at last count, depending on what you count as a database service.

Four years ago, it introduced its MongoDB-killer DocumentDB. Yet, just last year, the two signed a six-year joint go-to-market partnership. The EC2 side of AWS can become a best frenemy when it smells a significant revenue opportunity. You can imagine the sweet nothings that Oracle account reps are likely whispering to their most loyal joint customers.

Regional saturation

The other side of the picture is OCI’s world conquest strategy, which is to saturate the world with OCI “regions.”

There are a couple points to keep in mind when understanding what an OCI region is. We put regions in quotes because, to Oracle, a region is typically a single data center, whereas to AWS, Azure and Google, they are collections of two or more data centers designated to cover a specific geographic area.

Another point to keep in mind: In most cases, Oracle’s regions, or data centers, tend to be smaller than their AWS or Azure counterparts. This is an artifact of OCI’s origins. While Oracle has always sought to build out a multipurpose cloud that envelopes all services, from infrastructure as a service to platform as a service to software as a service, compared with its rivals its initial business was more niche-oriented, catering to Oracle Database and Fusion applications customers.

In other words, classic enterprise workloads rather than born-in-the-cloud web or social network use cases. Consequently, the requirement was focused more on redundancy at the rack, row and data center level rather than seemingly endless scale-out.

The compactness of OCI regions, initially shaped by its original market, has also become a design template. A key part of OCI’s positioning is wicked high performance. Much of that is thanks to innovations such as flat network topology and use of RDMA-based superclusters that bypass layers of compute and operating system.

As Oracle builds out its presence, it is also seeking to locate as close to its end customers as possible to turn the speed-of-light limitation to its advantage. Keeping data centers small and uniform means you can build more of them, faster and cheaper, with a cookie-cutter approach. Oracle Chairman and Chief Technology Officer Larry Ellison (pictured) likens it to Elon Musk’s Starlink, which relies on swarms of close-to-Earth satellites to saturate the planet. Ellison wants to do the same with cloud.

Oracle has a ways to go. Today, OCI has 41 regions, comprised of 49 availability domains (data centers), along with an undisclosed number of secret regions for U.S. national security customers. By comparison, AWS has 31 regions, comprised of 99 availability zones (data centers), and 450 points of presence (the cloud equivalent of classic content delivery network endpoints) in production.

Both also offer prepackaged options for customers that want a piece of AWS or OCI within their own four walls. Additionally, AWS has 33 Local Zones (hyperlocal facilities) and 29 Wavelength zones (data centers colocated with telco providers).

Our take is that Oracle wants to build the equivalent of AWS Local Zones, and lots of them.

Tony Baer is principal at dbInsight LLC, which provides an independent view on the database and analytics technology ecosystem. Baer is an industry expert in extending data management practices, governance and advanced analytics to address the desire of enterprises to generate meaningful value from data-driven transformation. He wrote this article for SiliconANGLE.

Image: Oracle