Rubrik now backs up Azure Cosmos DB, the database behind AI apps
Rubrik has added immutable backup and cross-tenant recovery for Azure Cosmos DB, the vector database whose revenue grew 50% year on year on AI workloads, with retention beyond Azure's 35-day cap.

Why it matters
- Rubrik added Azure Cosmos DB protection on 19 November 2026, covering databases up to 10 TB with daily immutable backups within a 24-hour window.
- Azure Cosmos DB revenue rose 50% year on year, driven by AI application workloads, per Microsoft's figures.
- The offering supports retention of months or years, beyond Azure Cosmos DB's native 35-day continuous backup limit.
- Recovery works across subscriptions, regions and tenants from an air-gapped copy in a separate trust domain.
- Rubrik CPO Anneka Gupta and Microsoft Cosmos DB VP Kirill Gavrylyuk both framed the launch around resilience for AI applications.
Rubrik has added Azure Cosmos DB to its cloud data protection portfolio, targeting the database that Microsoft says generated 50% year-on-year revenue growth on the back of AI application workloads.
The announcement, made on 19 November 2026, extends Rubrik's protection to a service that has quietly become a load-bearing wall in enterprise AI infrastructure. Companies building AI systems use Cosmos DB — Microsoft's managed NoSQL and vector database — to store the vectors, memory and application state those systems depend on. Until now, that layer had limited options for independent, immutable backup.
What does the product actually do?
At its core, the offering gives customers a copy of their Cosmos DB data that lives outside the main Azure environment. That matters because the failure modes companies worry about are not just outages. If data gets deleted, corrupted or encrypted in an attack, the application built on top of it can stop behaving as intended even while the wider system looks online.
The technical specifics tell you who this is for:
- Daily immutable backups for Azure Cosmos DB for NoSQL databases up to 10 TB, completed within a 24-hour window.
- Application-consistent recovery, meaning restores reflect the exact committed state of the system, including deletions — not just an older copy of the data.
- Container-level restores to existing or new accounts, preserving partition keys, indexing policies and consistency settings.
- Cross-subscription, cross-region and cross-tenant recovery.
The restore fidelity point is worth pausing on. Rubrik keeps the database's configuration intact when it brings data back, so a recovered container doesn't arrive with different indexing behavior or consistency settings than production had. For an application that depends on those settings, a restore that loses them is a partial failure dressed up as success.
The short version: this is recovery designed for databases where the details of state matter as much as the bytes.
Why does a separate trust domain matter?
The service is also built to maintain a trust domain separate from the production Azure identity boundary. That's a direct answer to one of the uglier scenarios in modern ransomware attacks: compromised credentials.
If an attacker holds valid production identities, they can often reach the backups too — unless those backups sit behind a different authentication perimeter. Rubrik's air-gapped immutable copy in a separate trust domain is designed to survive exactly that situation.
Rubrik's Chief Product Officer, Anneka Gupta, framed the launch around the gap between availability and recovery.
"Azure Cosmos DB is where AI applications keep the data they depend on: the vectors, the memory, the state," Gupta said.
"Availability keeps that data online, but it doesn't bring the business back after a disruption, whether that's an operational failure or an attack. By extending Rubrik Cloud Data Protection to Cosmos DB, Azure teams will be able to recover at the level they need, at scale, from an air-gapped immutable copy in a separate trust domain, so the data behind their enterprise AI applications stays recoverable even if production identities are compromised."
That's the argument in one breath: uptime is not the same as recoverability, and AI applications have raised the stakes on the difference.
How does it go beyond Azure's native limits?
Retention is the other half of the story. Azure Cosmos DB natively retains continuous backup data for 35 days. Rubrik lets organisations apply retention policies lasting months or years, which is the kind of timeline regulators and auditors tend to demand.
For financial services, healthcare and retail — the sectors Rubrik points to — that gap between 35 days and multi-year retention is often a compliance requirement, not a nice-to-have. A bank that must reconstruct records from 18 months ago cannot do it from a 35-day window.
The database also now sits under the same policy engine Rubrik customers use for their other cloud applications. For larger organisations, that's a consolidation play: backup, retention and recovery rules for multiple services managed through one control framework, instead of a patchwork of service-specific tools.
The takeaway: Rubrik is positioning Cosmos DB protection as one tile in an existing compliance and recovery mosaic, not a standalone product.
Why is AI changing the backup problem?
The launch lands at a moment when AI workloads are moving from pilots into production, and that shift changes what recovery means.
In the sectors Rubrik names, the database layer increasingly holds more than transactional records. It holds the state and memory used by AI-driven services — the accumulated context that makes those services behave the way they do. Lose that, and you're not just restoring an application; you're rebuilding an application that no longer knows what it knew.
That reframes the recovery challenge in a concrete way. Restoring an AI application now means recovering data in a way that reflects the exact committed state of the system, including deletions, rather than simply bringing back a previous copy. A restore that reintroduces data the system had already committed as deleted can corrupt application logic in ways that are hard to detect and harder to debug.
Microsoft's own numbers underline why this database in particular has become a protection priority. Azure Cosmos DB revenue rose 50% year on year, driven by AI application workloads, according to Microsoft's figures. Growth at that pace means more business-critical state is landing in a service whose native recovery options were built for a different era of workload.
Kirill Gavrylyuk, Vice President of Cosmos DB at Microsoft, framed the partnership as part of a broader resilience strategy for the platform's customers.
"Customers are building their most important AI applications on Azure Cosmos DB," Gavrylyuk said. "Working with Rubrik gives them another way to help protect and recover the data for their applications, as part of a resilient cloud strategy."
Note the posture: Microsoft isn't claiming native tools are insufficient, but it is endorsing a third-party safety net for the workloads its customers describe as their most important.
What's the bigger picture?
The move reflects a broader shift in cloud data protection. As AI systems move into production, the industry is discovering that its backup assumptions were built for stateless applications sitting on top of relational databases. Vector stores, memory layers and application state break that model — they're constantly changing, tightly coupled to application behavior, and increasingly the target of attacks aimed at the data layer itself.
Rubrik's bet is that enterprises will want the same discipline around AI data infrastructure that they spent two decades building around conventional databases: immutable copies, separate trust domains, long retention and policy-driven recovery. Whether Cosmos DB protection becomes table stakes across the backup market or stays a Rubrik differentiator, the launch signals that AI-era data protection is now a product category rather than a theoretical concern.
Source: itbrief.com.au
More from Sophie Lindqvist
Show full bio
Staff writer covering regulation, policy and their impact on artificial intelligence and machine learning at AI In Context.
777 articles
Related articles
- Cohesity Expands Backup, Recovery and AI Data Access Across Microsoft Stack
- Rubrik brings BigQuery backup to general availability, extends cover to Cloud Storage
- OpenAI's Habitat storage layer hits 20M QPS on Cosmos DB
- OpenAI's Habitat Storage Layer Hits 70 Million Requests a Second
- Commvault and Databricks turn backed-up data into governed AI fuel