The three announcements at a glance
| Announcement | Problem it targets | Stated status |
|---|---|---|
| ANQ Frozen | Long-term retention of file data without losing visibility | Controlled preview; GA date reported inconsistently |
| Blob Accelerator | Latency and repeat-read API charges on Azure Blob data | Controlled preview; read/write and orderability expected Q1 2027 |
| Cloud Native Qumulo on Cobalt | Scarce or costly x86 compute for the data layer | Available now, per Qumulo |
ANQ Frozen: an archive tier that still has a file system on top
Azure Native Qumulo (ANQ) has a new cold data tier, ANQ Frozen, which stores data at the cost of the Azure Blob Archive tier. ANQ is Qumulo's scale-out file system software running natively in Azure. Qumulo says the new tier keeps the full namespace online and file metadata searchable through Qumulo NeuralSearch without rehydrating data.
According to Blocks & Files and Qumulo's release, the workflow looks like this:
- Data lands in ANQ as usual.
- Administrators move it to Frozen as whole files, using manual or API commands. Metadata, path and identity and access attributes are retained.
- Users keep browsing directories over SMB and NFS, even for frozen data.
- NeuralSearch indexes file-system metadata and application metadata from file headers, such as well names or project IDs. Teams can find candidate files before paying for retrieval.
- Retrieval is approval-based. Administrators can sign off on requests, and Qumulo says that keeps retrieval costs predictable.
- Each system uses its own dedicated Azure storage account. Qumulo says the data sits there as native objects, readable with standard Azure tools and no Qumulo software in the path.
- Existing ANQ Hot and Cold customers can add Frozen. Large archives can get a dedicated Frozen system.
Blocks & Files also says the product relies on "Frozen bridges", with freeze jobs moving data to Archive and thaw jobs rehydrating it.
Searchable is not the same as readable
A browsable namespace can mislead. Microsoft's documentation says an archived blob is offline and can't be read or modified until it is rehydrated to an online tier. Qumulo's pitch is that you can locate files cheaply. It does not say you can open them instantly.
Microsoft's rehydration documentation adds some limits that matter for planning:
- Standard-priority rehydration might take up to 15 hours for objects under 10 GB. High priority can finish in under an hour for such objects, but it costs more and has no guaranteed completion time.
- The rehydration limit is 10 GiB per hour per storage account with priority retrieval. Microsoft warns that bulk timelines do not scale linearly.
- Archive blobs carry a 180-day minimum retention period. Moving or deleting them earlier incurs an early-deletion charge.
- Archived blob metadata stays readable but is read-only. Its storage is billed at cool-tier rates.
Qumulo describes a retrieval window of typically 2 to 12 hours. Treat that as a vendor description of typical behavior. A large legal-hold or model-training pull could run longer, especially on a single storage account.
Tape or not tape?
Blocks & Files speculates that the Archive tier is characteristic of offline magnetic media such as tape libraries. It also says Microsoft has not confirmed this. That is the publication's inference, and I haven't seen Microsoft confirm the media type. For buyers the practical point is the access model: hours of latency, higher retrieval cost and rehydration before use.
Availability: a date discrepancy
Qumulo's own release says ANQ Frozen is in controlled preview, with general availability planned for early 2027. Blocks & Files reports the narrower target of May 2027. Treat May as the publication's more specific reporting, not a confirmed Qumulo commitment. Preview access goes through a Qumulo account team.
Qumulo's release also quotes Microsoft's Aung Oo, VP of Azure Storage. He says Frozen extends the data lifecycle from hot to cold to Archive tier storage. He also notes that Qumulo was a launch partner for Azure Smart Tier earlier this year.
Blob Accelerator: file access and caching for data already in Blob
Qumulo says the Blob Accelerator attaches directly to an Azure Blob container a customer already owns, which remains the single, authoritative copy of the data. File-oriented applications reach it over NFS and SMB. Applications using the Blob APIs keep working against the same container.
Qumulo's performance claims:
- Object latency of 10–150 ms drops to 1–5 ms.
- NeuralCache serves more than 90% of repeat reads from NVMe and DRAM, in under a millisecond.
- There are no additional transaction costs after the first read of an object, which would reduce repeat API charges.
- The same container can be read from another Azure region, another cloud, on-premises or at the edge, with "strict consistency" and no copies.
The release gives no test conditions, workload mix, cache-hit definition or pricing assumptions for these numbers. The "no API charges" claim applies after the first read, so cold, one-pass scans won't see the benefit. Workloads that reread the same data, such as training epochs or repeated analytics, are the best fit. The consistency and cross-cloud claims are also unproven until detailed documentation or independent testing appears.
Existing ANQ customers can add the accelerator to their cluster and connect existing storage accounts. The release says it adds cache and performance when a job needs it and scales down afterward. Qumulo lists AI training and inference, lakehouse and observability analytics, media delivery, genomics and seismic imaging as early use cases.
Availability is controlled preview beginning now, with read/write support and general orderability expected in the first quarter of 2027. That implies the preview is read-oriented for now.
Azure Blob Access Points
Qumulo also says it is a launch partner for Microsoft's new Azure Blob Access Points. The feature lets services such as Microsoft Fabric, Azure AI Foundry and Azure OpenAI reach unstructured file data on Qumulo through Blob Storage APIs, with no copying or custom integration. The source for this is Qumulo's announcement. Expect to still configure access, permissions and supported interfaces. "No custom integration" doesn't mean zero setup.
CNQ on Azure Cobalt: cheaper, easier compute
Cloud Native Qumulo is the customer-managed version. The customer deploys it into their own Azure subscription, typically with Terraform, and controls the VMs, networking and scaling. Qumulo says every CNQ tier, plus Qumulo AI Accelerators and the Blob Accelerator, now runs on Cobalt-based instances. Qumulo says these cost 9–48% less than comparable x86 instances, with software, features and the single namespace unchanged. It says CNQ on Cobalt is available today through the Azure Marketplace and directly from Qumulo.
The release doesn't name the SKUs, regions or baseline behind the 9–48% range, so it's a vendor estimate. Microsoft's own Cobalt news from Ignite gives some context. It announced general availability of Cobalt 200 VMs, with up to 50% better performance than Cobalt 100 and availability in 14 regions, expanding to 23 by the end of 2026. Those figures describe the platform. They don't validate Qumulo's savings or show which Cobalt generation Qumulo uses.
What admins should take from this
- Frozen suits archives you must keep but rarely read. Model retrieval costs, the 180-day minimum and the per-account rehydration ceiling before moving petabytes.
- Blob Accelerator suits repeat-read AI and analytics on data that already lives in Blob. Ask for workload-specific benchmarks and the preview's read/write limits.
- CNQ on Cobalt is worth a pilot if x86 capacity or cost is a constraint. Confirm the exact instance sizes and regions you'd use.
Everything here is vendor-announced and mostly in preview. The ideas are credible, particularly searchable metadata over an offline tier. The headline numbers for latency, cache hits and savings should be validated in your own tenant first.
References
- Qumulo on Azure embraces the data Ice Age - Blocks & Files Blocks & Files · 2026-11-17T19:00:00+00:00
- Your Privacy Choices Opt-Out Icon azure.microsoft.com
- Qumulo Brings High Performance to Azure Blob Data qumulo.com