What Microsoft announced
The announcement says Next Generation keeps what Elastic SAN customers already use. That means a single SAN resource hierarchy that consolidates block-storage workloads and shares performance dynamically. It also keeps iSCSI connectivity for SAN migrations such as datacenter exits and hardware refreshes.
The new generation adds support for mission-critical workloads. It also adds thin provisioning and autoscaling, which Microsoft says will help customers optimize storage costs further.
Microsoft's Azure Storage documentation also lists the service as a preview. It describes centrally managed block storage with Business Critical and General Purpose QoS tiers, shared capacity and performance, thin provisioning, and autoscaling.
The QoS tiers
The main change is quality of service (QoS) set at the volume-group level. Different workloads can share one SAN with what Microsoft calls noisy-neighbor protection. The announcement lists three tiers:
| Tier | Performance model | Latency | Per-volume maximum | Intended for |
|---|---|---|---|---|
| Business Critical | Reserved | Low sub-millisecond | 400,000 IOPS, 10,000 MBps | SAP HANA, Epic, I/O-intensive OLTP |
| General Purpose – Direct Attach | Shared | Sub-millisecond | 80,000 IOPS, 2,000 MBps | Vector and OSS databases |
| General Purpose – iSCSI | Shared | Low single-digit millisecond | 80,000 IOPS, 1,280 MBps | Azure VMware Solution, iSCSI-preferring customers |
These are announced ceilings, not guaranteed application-level results. The announcement doesn't say how they were measured.
Capacity and cost features
Microsoft says customers can provision for growth without paying for the full logical capacity of volumes upfront. Capacity and performance scale independently.
- Thin provisioning: Volumes are sized for anticipated growth. Physical capacity is consumed as data is written, up to 5x the SAN's provisioned capacity.
- Autoscaling: SAN capacity is added automatically based on customer-defined thresholds.
- Independent provisioning: Capacity, IOPS, and throughput are set separately at the SAN level.
- Crash-consistent snapshots: These give point-in-time protection for volumes.
- Thin clones: These are space-efficient, mountable dev/test copies of Business Critical volumes. They are marked "coming soon", so they are not part of the preview at announcement time.
A Next Generation SAN can scale to 1 PiB, 4 million IOPS, and 160 GBps. Those are SAN-wide figures, not per-volume ones. Don't confuse the 4 million IOPS SAN ceiling with the 400,000 IOPS Business Critical per-volume limit.
How this differs from today's Elastic SAN
Microsoft's current Elastic SAN documentation describes the original design. It is a fully integrated way to deploy, scale, manage, and configure a SAN, with built-in high availability. Volumes are carved from a SAN's capacity, and SAN performance is shared across them. Volume groups manage volumes at scale, for example by applying virtual network rules that the volumes inherit.
The documentation says the current service's volumes can each scale to 80,000 IOPS. That is the same figure as the new General Purpose tiers, so the new headline is the Business Critical tier's reserved performance. The documentation also describes iSCSI as the connection method for existing volumes. The Direct Attach tier is therefore a distinguishing feature of the new generation.
Elastic SAN already has some of these capabilities as separate features. The related-updates list shows capacity autoscaling and single-volume snapshots reached general availability in 2026. Next Generation folds thin provisioning and independent performance provisioning into the model.
What the announcement doesn't say
Several details that matter for planning are missing from the announcement:
- Pricing and billing: Nothing is stated, so don't assume thin provisioning or autoscaling lowers your bill.
- Regions, redundancy, and networking: The announcement doesn't list preview regions, LRS/ZRS choices, or network requirements. The existing service documents LRS and ZRS options, but don't assume they carry over unchanged.
- iSCSI constraints: The current documentation lists limitations for the existing service, including no CHAP and one LUN per target. Confirm whether any apply to Next Generation.
- Autoscaling controls: No example thresholds, limits, or billing behavior are given.
- Access: The announcement points to a sign-up for the preview, so access may be gated.
Microsoft's own product page also quotes a different, higher figure for a low-latency performance mode. That figure is up to 1 million IOPS and 30,000 MB/s per volume. It differs from the November 17 tier numbers. Treat the announcement figures as announcement-time claims, because limits may change or refer to a different configuration.
A sensible evaluation plan
- Sign up for the preview and use a non-production subscription only.
- Replay a representative workload against each tier. Check whether the per-volume numbers hold, and test Business Critical isolation while a General Purpose volume group is busy.
- Find out what triggers autoscaling, and how thin provisioning behaves as you approach the 5x ratio.
- Confirm regions, redundancy, networking, and preview billing terms before building any plan around the service.
- If you run Azure VMware Solution, compare the General Purpose iSCSI tier against your current Elastic SAN setup.
Bottom line
Elastic SAN Next Generation tries to make Azure's SAN offering credible for demanding databases, while keeping the shared-pool economics that define the product. The tier numbers are impressive on paper. The missing details on pricing, regions, and limits mean it is a good candidate for testing now, not for production commitments.
Update: Microsoft confirms preview regions, billing basis, and recovery features (November 18, 2026)
Microsoft’s Azure Storage blog now confirms that the public preview is initially available with locally redundant storage in France Central and Southeast Asia. UK South, East US, and West US 2 are listed as coming soon. This narrows the earlier uncertainty around regional availability and redundancy: LRS is confirmed for the first wave, but ZRS is not announced.
Microsoft also clarifies thin-provisioning billing. Customers can assign up to five times the SAN’s provisioned capacity as logical volume capacity, but capacity charges remain based on provisioned SAN capacity—not the logical volume total. Autoscaling can then expand that provisioned capacity up to a configured maximum, increasing charges as it does.
The company additionally sets a preview-scale figure of up to 20,000 volumes per SAN. For protection, it says incremental crash-consistent snapshots are instantly accessible and restorable for five hours before being automatically hardened to lower-cost storage; hardened snapshots survive deletion of the source volume. Commvault and Druva integrations are cited for application-consistent backup and recovery, while Veeam support remains on the roadmap.
Microsoft’s 43% savings claim is a modeled comparison against an unnamed competing cloud provider, based on specific assumptions about shared workload peaks and initial utilization. It should not be treated as a general price guarantee.
References
- (In preview) Pulic Preview: Azure Elastic SAN Next Generation Azure Updates · 2026-11-17T22:54:18Z
- Azure Elastic SAN Next Generation public preview: Mission-critical performance & cloud-native scale Azure Storage Blog · 2026-11-18T04:56:46+00:00
- Introduction to Azure Elastic SAN | Microsoft Learn learn.microsoft.com