Blog

Oracle AI World 2026: What Database Teams Need to Know

Written by Neil Barton | Nov 20, 2026, 1:44:59 AM


Unsurprisingly, AI dominated the agenda. Agentic applications and workflows were everywhere. But beneath those headlines were some important developments for database teams, particularly around availability, maintenance, disaster recovery and the RTO and RPO targets Oracle is now designing towards.

For DBAs who couldn't make it to Las Vegas, these were the developments that stood out to me.

Oracle AI Database 26ai moves to the centre

One of the clearest themes was Oracle positioning Oracle AI Database 26ai as the data foundation for its wider AI strategy. Capabilities such as AI Vector Search, Select AI and agentic AI increasingly bring AI to operational enterprise data rather than requiring that data to be moved into separate platforms.

For DBAs, the significance goes beyond AI functionality. If more applications and autonomous agents depend directly on operational data, the underlying database needs to remain available, current and trusted. Availability and recoverability become even more important as more automated processes rely on that data.

Exadata pushes further towards maintenance without downtime

Availability during planned maintenance was another important theme. Oracle continues to expand Exadata Live Update, allowing supported software and operating-system updates to be performed while database services remain online. Exadata System Software 26ai also extends this approach with capabilities such as online Linux kernel security updates using Ksplice and Live Update for RoCE Switches.

The direction is clear: Oracle wants more routine maintenance to happen with little or no visible disruption to applications. There was also an increasing emphasis on agentic AI for database and infrastructure management, which I expect to grow significantly as Oracle applies agents to more operational activities.

More options for local and regional standby protection

Oracle also highlighted greater flexibility in standby architecture: Exadata Database Service on Exascale Infrastructure now supports multiple Data Guard standby databases associated with a primary, making it possible to combine local and remote standby strategies.

That enables different standbys to address different risks. A local standby can protect against infrastructure or availability-domain failures, while a remote standby provides protection against a wider regional outage. Rather than treating DR as a single secondary environment, the architecture increasingly allows protection to be designed around individual failure scenarios.

MAA Diamond raises availability expectations again

One of the most interesting developments was Oracle's Diamond tier, introduced earlier this year, within Maximum Availability Architecture. Diamond sits above Bronze, Silver, Gold and Platinum and is aimed at what Oracle calls Extreme Availability.

Oracle's latest MAA architecture sets extremely aggressive targets around both local high availability and regional disaster recovery, using technologies including Oracle AI Database 26ai, RAC, Exadata, Active Data Guard, GoldenGate and Globally Distributed AI Database.


Most organisations will never need an architecture at this level. But the wider significance is important. Oracle is continuing to push expectations towards:

> recovery measured in seconds rather than minutes;
> zero or near-zero data loss;
> maintenance without application interruption;
> and resilience across multiple failure domains.

Those expectations inevitably influence the wider Oracle community.

The business does not necessarily care which Oracle edition or infrastructure is underneath an application when it asks: “If this database fails, how much data will we lose and how quickly can we recover?”

Recovery is also becoming a security conversation

Another noticeable theme was the overlap between disaster recovery and cyber resilience. Oracle's recovery architecture increasingly considers ransomware alongside traditional infrastructure and regional failures. Immutable backups, protected recovery environments and cross-region recovery are becoming part of the wider resilience discussion.

OCI Full Stack Disaster Recovery takes that a step further by orchestrating recovery across the wider application stack rather than treating database recovery as an isolated process. That reinforces an important distinction: Having a backup is not the same as being recoverable. Similarly, having a standby does not tell you how current that standby is, how quickly it can take over, or whether the wider application will successfully recover around it.

DR wasn't centre stage,  but resilience clearly was

For all the investment Oracle is making in availability and resilience, I was struck by how little dedicated disaster recovery content there was at AI World. I could only find two sessions specifically focused on DR, and one of those was a smaller theatre session on the show floor. That feels consistent with Oracle events over the last few years. The focus has shifted heavily towards OCI, Autonomous Database and AI, while traditional database and DR topics take up less of the agenda.

But DR itself has certainly not become less important. The announcements around lower RTOs and RPOs, online maintenance, multiple standby configurations and cyber recovery all point in the same direction. The conversation is moving from: “Do I have DR?” to: “How resilient is the service, and how much data could I actually lose?”

What does this mean for Oracle Standard Edition?

This is where I think things become particularly interesting. Many of Oracle's most advanced availability architectures depend on Enterprise Edition technologies and platforms such as Exadata, RAC, Active Data Guard and GoldenGate. Those are powerful options, but they are not appropriate or necessary for every Oracle workload.

A large number of organisations continue to run Oracle Standard Edition because it provides the functionality their applications need without the additional complexity and licensing economics of Enterprise Edition. Historically, disaster recovery has been one area where Standard Edition users have had to accept a compromise. Oracle Standard Edition does not include Data Guard, so physical standby solutions such as StandbyMP have traditionally relied on archive-log replication.

That is proven and reliable, but there is an unavoidable timing gap. Transactions committed to the primary are not protected on the standby until the corresponding redo has been archived, transferred and applied. For many environments, an RPO measured in minutes is completely acceptable. But what if it isn't?

Dbvisit unveils sub-second RPO for Oracle Standard Edition

Against that backdrop, Dbvisit unveiled StandbyMP Zero Data Loss (ZDL) at Oracle AI World.

For the first time, organisations running Oracle Standard Edition can target a sub-second RPO using StandbyMP, without moving to Enterprise Edition purely to access Data Guard. That is a significant change for the Standard Edition community.

ZDL adds continuous redo streaming alongside StandbyMP's existing archivelog replication. Archivelog files continue to be transferred as normal, maintaining the proven replication foundation. At the same time, ZDL monitors the active online redo and continuously synchronises those changes to the standby.

If the primary fails, that streamed redo can be used during activation to close the recovery-point gap beyond the last completed archive log. In suitable environments, the result is a move from an RPO measured in minutes to less than a second.

Importantly, ZDL does not turn Standard Edition into Oracle's Diamond MAA architecture, and it does not mean zero downtime. It addresses a specific issue: How much committed data is potentially lost if the primary fails?

RTO remains a separate consideration. The standby still needs to be activated and the application brought back into service. But for Standard Edition users, the change is significant.

Until now, organisations requiring this level of RPO could find themselves considering Enterprise Edition and Data Guard even when the workload did not otherwise require Enterprise Edition functionality. Sub-second RPO on Oracle Standard Edition gives them another option.

My biggest takeaway from AI World

For me, the most important database story from Oracle AI World wasn't any single announcement. It was the direction of travel Oracle is continuing to push availability and recovery expectations towards seconds, zero or near-zero data loss, automated operations and resilience across the entire service.

Not every database needs the architecture required to achieve Oracle's highest MAA tiers. But every organisation should understand its own requirements.

> How much downtime can the application tolerate?
> How much committed data can the business afford to lose?
> And does the current architecture actually meet those expectations?

For Oracle Standard Edition users, an RPO measured in minutes has historically been one of the compromises they had to consider.

With the introduction of StandbyMP Zero Data Loss at Oracle AI World, that compromise is no longer automatic.

➡️ Learn more about Zero Data Loss for Oracle SE