NetApp’s Cloud Volumes ONTAP has entered public preview with a tightly scoped integration for Microsoft OneLake, announced March 10, 2026, that promises to let enterprises surface their existing NAS-held file datasets into Microsoft Fabric’s lake-centric analytics environment via S3-compatible object access—
without mandatory large-scale data migrations. The move aims to remove a common choke point that stalls AI and analytics projects: locked-up unstructured data. If it delivers as advertised, the integration could shorten time-to-insight for Azure-first organizations, reduce duplication and storage costs through ONTAP’s efficiency features, and preserve existing governance models while enabling cloud-native AI workflows to read directly from enterprise file systems.
Background
Why this matters now
AI projects are starving for high-quality, high-volume data. Yet a large portion of enterprise data—estimates repeatedly show upward of 70–90%—remains in file systems (NFS/SMB) or behind legacy application silos. Extracting value often requires time-consuming re‑architecting: copying or converting file shares into object stores, rebuilding access controls, and reconfiguring auditing and classification policies. Those steps add cost, complexity, and governance risk.
Microsoft’s OneLake and Fabric were designed to be a unified data lake and analytics platform that supports multiple engines and formats. But until recently, OneLake’s zero‑ETL promise still required data to be accessible in object or table formats. NetApp is trying to close that gap with Cloud Volumes ONTAP, enabling OneLake—or Fabric engines that rely on OneLake—to treat NAS-resident data as S3-accessible objects in place. The key pitch is simple: keep data where it lives, avoid a costly “migration tax,” and let AI pipelines reach production-ready datasets much faster.
What NetApp and Microsoft each bring
- NetApp Cloud Volumes ONTAP: the cloud-deployed form of ONTAP that brings decades of enterprise file-services features—Snapshot copies, SnapMirror replication, thin provisioning, deduplication, compression, FabricPool-like tiering, and FlexCache—for hybrid cloud and multi‑cloud deployments.
- Microsoft OneLake / Fabric: a lake-centric, multi-engine analytics platform that centralizes governance and enables analytics, warehouse, and AI engines to work from a single logical data lake.
Together, the integration is positioned as a hybrid solution: OneLake can access file datasets through an S3-compatible interface provided by Cloud Volumes ONTAP, while ONTAP continues to manage data protection, storage efficiencies, and access control.
What the integration actually does
In-place S3 access to NAS datasets
The core capability is
object access endpoints for Cloud Volumes ONTAP volumes so that OneLake can create
shortcuts or direct object references to data using S3 semantics. Instead of copying file shares into an object store, OneLake sees the ONTAP‑exposed endpoint as an object source. This virtualization approach keeps the single source of truth inside ONTAP while making that data consumable by Fabric engines and Azure AI/ML services.
Shortcuts and virtualization
OneLake’s shortcut feature provides a virtual pointer to external data sources. When a OneLake workspace creates a shortcut to an ONTAP object endpoint, compute engines in Fabric can query or read the data in place. That eliminates duplication and preserves ONTAP’s governance posture—permissions, snapshots, and retention policies remain enforced at the origin.
Performance acceleration with FlexCache
To address latency for interactive analytics or high-throughput AI training, Cloud Volumes ONTAP supports
FlexCache caches. FlexCache creates read caches closer to compute—reducing round-trip times and minimizing the impact of network latency on large-scale read operations. For AI training that repeatedly reads the same training corpus, caching can materially reduce runtime.
Enterprise protection and governance remain on ONTAP
ONTAP’s data management features continue to apply:
- Snapshot copies give fast point-in-time rollback and support consistent reads for data pipelines.
- SnapMirror can replicate volumes for disaster recovery or to stand up local read reservoirs.
- Encryption, role-based access controls, and auditing ensure compliance requirements can be maintained at the storage layer rather than re-implemented in the lake.
Storage efficiency and tiering
Cloud Volumes ONTAP’s efficiency stack (deduplication, compression, thin provisioning) remains active when exposing data. Cold snapshot data can still be tiered to object storage to reduce cost, and those efficiency gains apply to the amount of data OneLake actually reads or indexes.
Commercial note: Azure Marketplace and MACC
NetApp notes that Cloud Volumes ONTAP purchases made through the Azure Marketplace are eligible to contribute toward Microsoft Azure Consumption Commitments (MACC) under standard marketplace buying rules—meaning procurement through Marketplace can map to existing Azure consumption commitments for organizations that use that purchasing model. Practical eligibility depends on the offer type and your enterprise billing contract.
Technical and operational implications
Protocol translation, semantics, and consistency
Exposing NFS/SMB file data via S3 is inherently a translation of semantics. Files and objects differ in metadata models, directory semantics, and consistency guarantees. NetApp’s object endpoint implements a mapping but operational teams must validate how:
- File metadata (permissions, ACLs) is represented to object APIs.
- File system features like sparse files, extended attributes, and symbolic links surface via an object view.
- Concurrent read/write operations from Fabric and file protocol clients are reconciled—especially when workloads perform writes through both object and file APIs.
Expect careful testing: read-after-write semantics and metadata fidelity matter for training pipelines and for reproducibility in analytics.
Networking, latency, and throughput
OneLake compute engines will connect over Azure network paths to Cloud Volumes ONTAP instances. Key considerations:
- Ensure low-latency network paths (VNet peering, ExpressRoute/Private Link patterns) where possible to avoid egress or public internet hops.
- Use FlexCache for read-heavy, latency-sensitive AI workflows—place caches close to compute (same region/AZ).
- Watch throughput caps on instances and cloud storage tiers; provisioning must match AI training bandwidth demands.
Identity, authentication, and permissions
Access must be tightly controlled:
- S3 endpoints need credentials and policies mapped to Azure identities; integration points can require managed identities or service principals for OneLake to authenticate.
- ONTAP’s native ACLs must be reconciled with OneLake/Fabric access models so data governance remains central and auditable.
- Data access paths used by Fabric engines should be listed in your data access governance policy and monitored for unusual activity.
Data protection interplay
Snapshot and replication tools still run in ONTAP. But there are interactions to be mindful of:
- When volumes are tiered or SnapMirror is used, consider how OneLake shortcuts reference data that might be in a different lifecycle state (tiered, archived).
- SnapMirror-based DR copies may require separate shortcut management or re-registration with OneLake to ensure recovery workflows remain valid.
- Test restores through the OneLake view to ensure the Fabric engines see restored states as expected.
Cost and TCO considerations
Where the savings come from
- Eliminated or reduced migration costs: no bulk copy into object stores; avoids double storage and migration pipelines.
- Reduced pipeline engineering and maintenance: shortcuts remove ETL complexity for certain workloads.
- ONTAP storage efficiencies lower cloud object storage bills by reducing the logical footprint before tiering or before Fabric reads data.
Where costs remain or can rise
- Network egress and data transfer: depending on topology and whether compute sits in a different subscription/region, egress or cross‑region transfer can add cost.
- Cache and compute sizing: adding FlexCache instances and allocating high‑performance compute for AI jobs influences spend.
- Object tier fees and access charges for data Azure bills when data is read frequently from tiered storage.
- Marketplace procurement: while Marketplace purchases may contribute to MACC, organizations must validate offer eligibility and billing path to ensure it counts against their commitments.
License and procurement nuances
Eligibility for MACC contribution generally requires an offer to be transactable through the Azure Marketplace on an eligible subscription and billing arrangement. Customers should confirm with their Microsoft account team and NetApp sales representatives that the specific Cloud Volumes ONTAP SKU and procurement route will decrement their MACC prior to final purchase.
Security, governance, and compliance
Preserving governance at the source
A major advantage of the approach is that governance remains primarily enforced by ONTAP:
- Retention and deletion policies, when managed in ONTAP, carry through so OneLake access respects lifecycle rules.
- Snapshots and immutable protection can be used to secure training datasets against inadvertent deletion or tampering.
Auditability and lineage
Data lineage matters for AI explainability and regulatory audits. Keep these controls in place:
- Ensure ONTAP and BlueXP logging feed into your SIEM to capture who accessed data through OneLake.
- Track which OneLake workspace or Fabric engine consumed which shortcuted object to maintain provable lineage.
Attack surface and exposure risk
Externalizing a file system as an S3 endpoint expands the attack surface:
- Harden credentials and apply least-privilege access between OneLake and Cloud Volumes ONTAP.
- Use private network paths (VNet integration, Private Link) rather than public endpoints where possible.
- Employ encryption in transit and at rest—verify that the object endpoint enforces TLS and that keys are managed to meet compliance regimes.
Practical guidance: getting started (recommended pilot checklist)
- Inventory datasets:
- Identify candidate file shares and volumes that are stable, read-heavy, and valuable for analytics or model training.
- Validate governance policies:
- Ensure retention, classification, and access control policies on those volumes are documented and tested.
- Deploy a small Cloud Volumes ONTAP instance in the target Azure region:
- Configure object endpoints and secure network access.
- Create a OneLake shortcut to the ONTAP object endpoint:
- Use a non-production workspace initially and validate read paths from Fabric engines.
- Add FlexCache for latency-sensitive workloads:
- Place caches in the same region and run representative read tests.
- Exercise data protection scenarios:
- Test snapshot-based restores and SnapMirror replication impacts on OneLake visibility.
- Measure performance and cost:
- Run a cost model that includes compute, networking, cache, and estimated object access charges.
- Expand gradually:
- Move additional datasets into the shortcut model only after governance, performance, and cost are acceptable.
Strengths and opportunities
- Faster time to insight: Avoiding mass data migrations and building shortcuts can compress the timeline from data identification to model training or analytics by weeks or months in many cases.
- Preserved governance: Because the source remains in ONTAP, existing enterprise policies and protections continue to apply—reducing duplication of security controls.
- Storage efficiency: NetApp’s deduplication, compression, and thin provisioning can reduce the billable footprint before data ever reaches longer‑term object storage.
- Flexible hybrid model: The approach supports hybrid scenarios where some datasets must remain on-premises while still being consumable by cloud analytics engines.
- Commercial alignment: Procurement via Azure Marketplace may align with corporate cloud consumption commitments, potentially simplifying purchasing and cost allocation.
Risks, limitations, and where to be cautious
- Semantic mismatches between file and object APIs: Not every file-system feature maps neatly to S3. Some workloads that rely on POSIX semantics or filesystem attributes may behave differently when accessed via object endpoints.
- Hidden performance traps: AI training is extremely sensitive to sustained throughput. Objectized access over network paths can bottleneck training unless caches and network topology are correctly provisioned.
- Governance drift if not carefully managed: While ONTAP can maintain policies, it is possible for teams to create parallel access paths or copies within OneLake that bypass source governance. Enforce organizational policies to avoid shadow data copies.
- Procurement caveats: MACC contribution depends on marketplace eligibility and billing routes. Confirm the exact SKU and checkout path with your procurement and Microsoft contacts.
- Operational complexity at scale: Managing many shortcuts, caches, and cross-region access paths can become operationally heavy. Invest in automation and monitoring (BlueXP or equivalent) early.
- Preview limitations: As a public preview feature, expect limitations, changes, and potential missing features versus GA. Don’t assume production-grade SLAs or full support parity with existing ONTAP features until GA.
How this compares to alternatives
- Azure NetApp Files object REST API: Azure NetApp Files has previously offered object REST API methods that enable OneLake shortcuts. The CVO integration gives similar outcomes but is targeted at customers who already run ONTAP in the cloud or on-prem and want to reuse ONTAP workflows and efficiencies.
- Full data migration to object storage: Moving data into native object stores remains the simplest path for pure cloud-native workflows, but it carries a migration tax—time, duplicate storage, and governance re-implementation.
- Cloud-native file systems and third-party connectors: Solutions like cloud file gateways or third-party connectors can also virtualize file data for analytics, but they may not provide the same integrated storage efficiency, snapshot, and replication features as ONTAP.
Recommendations for IT leaders and architects
- Treat the preview as a pilot opportunity, not a wholesale migration strategy. Validate governance, performance, and cost on representative workloads.
- Design network topology with low-latency, high-throughput paths to ONTAP endpoints; plan caches adjacent to compute where possible.
- Update data catalog and lineage tools to include the OneLake shortcut layer so data consumers and auditors can trace consumption back to the ONTAP origin.
- Work with procurement and Microsoft account teams to confirm Marketplace SKU eligibility for MACC and understand billing flows.
- Automate policies for shortcut creation, cache lifecycle, and Snapshot retention to avoid operational drift and shadow copies.
- Prioritize datasets that are read-heavy, stable in schema, and low in write concurrency for the first wave—these will yield the fastest wins.
Final assessment
NetApp’s Cloud Volumes ONTAP integration with Microsoft OneLake represents a pragmatic, hybrid-first correction to a very real pain point: how to enable modern AI and analytics on the vast volumes of enterprise file data that organizations simply cannot or should not move. By exposing ONTAP volumes as S3-accessible endpoints and letting OneLake create virtual shortcuts, the solution can materially shorten project timelines, reduce duplicate storage, and keep governance where it already works.
That said, the benefits are conditional. Real-world gains will depend on careful engineering: network architecture, cache strategy, identity and access mapping, and ongoing operational discipline. The preview status means organizations should plan pilots to validate semantics (file-to-object mapping), performance under load, and procurement implications for MACC before declaring production adoption.
For Azure-first enterprises wrestling with unstructured file stores that block analytics and AI initiatives, this integration offers a compelling path forward—
if IT teams treat it as an architecture change that requires testing, monitoring, and governance controls, not simply a drop-in shortcut to faster outcomes.
Source: NetApp
https://www.netapp.com/blog/optimize-ai-analytics-microsoft-onelake/