From the NetApp edition of October 29, 2026
StorageReview: NetApp Novus 100 TB/s AI-factory storage claim rests on Omdia modeling of a smaller audited test
StorageReview (trade press) reporting on NetApp · Strategic partner: AI data infrastructure · Global (announced at NetApp INSIGHT 2026, Las Vegas)
- Publisher
- StorageReview
- Original publication
- October 1, 2026
- Source retrieved
- 2026-10-30
- Event date
- 2026-09-29
What happened
StorageReview reports NetApp Novus separates pNFS metadata (Novus Data Director on qualified Supermicro servers) from AFF A90 data nodes running ONTAP, uses stock NFSv4.2 pNFS clients, and is orderable now, with lighthouse customers expected within a month or two of INSIGHT. The article stresses that the 100 TB/s figure is a projection: Omdia audited near-linear scaling on a smaller test and modeled the larger result. It also notes AIDE advanced features remain preview with no GA date, and that NetApp's planned EEA sovereign Keystone support uses the same in-region passport-holder model NetApp uses for its U.S. public sector offering.
Why it matters
Limited: Novus targets roughly a hundred AI-factory operators with tens of thousands of GPUs, far beyond typical state, local or K-12 scale. Possible relevance to large research universities and state research computing consortia; most SLED buyers would see AFX or AIPod-class systems instead.
Evidence and measured results
Independent trade reporting based on vendor briefing and release; no independent benchmark was run by StorageReview. Omdia test report, configuration and raw results are not linked.
Limitations and uncertainty
Vendor-sourced figures relayed by trade press; projection not measurement; scale far exceeds most SLED environments.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-30; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Large research institutions need storage that keeps GPU clusters fed without proprietary clients.
- Stakeholders
- Research computing director
- HPC architect
- CIO
- Sponsored research office
- Discovery questions
- What GPU count and throughput do your planned workloads actually need?
- Is a proprietary parallel file system client a problem today?
- What metadata scale (file counts) do you expect?
- Value hypothesis
- For very large GPU clusters, a standards-based pNFS design may reduce client management overhead, pending independent validation.
- Potential engagement
- Workload characterization and benchmark planning session for a research computing team.
- Unsupported claims
- 100 TB/s as a measured result
- Relevance to typical agency or K-12 workloads
- Existing public-sector reference deployments
Pre-sales engineering
Role takeaway
- Fit
- Only very large research GPU clusters.
- Architecture and integration
- Novus Data Director metadata tier on qualified servers plus AFF A90 data nodes; stock pNFS clients on GPU servers over a separate lossless Ethernet fabric.
- Prerequisites
- Lossless Ethernet client fabric
- Linux clients with NFSv4.2 pNFS Flex Files
- Qualified hardware
- Constraints
- Initial release hardware-bound
- Scale targets are modeled
- Early availability timing
- Security
- Vendor states ONTAP data services, multi-tenancy and cyber-resilience are preserved; verify tenant isolation in shared research clusters.
- Proof of value
- Obtain Omdia audit methodology
- Run mixed small-file and checkpoint workloads
- Measure metadata operations at expected file counts
Delivery
Role takeaway
- Work
- Bounded proof of concept and fabric readiness assessment.
- Dependencies
- Network team for lossless fabric
- Hardware lead times
- NetApp lighthouse program
- Ownership
- Research computing / HPC operations.
- Skills and adoption
- Storage and HPC staff need pNFS and fabric tuning skills.
- Governance checkpoints
- Benchmark plan approval
- Fabric design review
- Go/no-go after proof of concept
- Acceptance criteria
- Proposed: sustained throughput meets institution-defined target on its workloads
- Proposed: metadata latency within agreed bound at target file count
- Proposed: no client-side proprietary software required
- Risks
- Overestimating need relative to scale
- Early-release maturity
- Network fabric cost
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Standards-based pNFS client avoids proprietary agents; requires a lossless Ethernet client fabric separate from the GPU backend fabric; initial release is hardware-bound to AFF A90 variant and qualified servers.
Governance
Who approves, reviews and stays accountable for outcomes?
Buyers should treat headline throughput as a modeled target and require workload-specific acceptance tests in contracts.
Security and privacy
What data, permissions and controls need testing?
Retains ONTAP data services, multi-tenancy and cyber-resilience per vendor; sovereign support models are relevant for regulated data but EEA-specific here.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Staff skills in storage, identity and security operations are the main workforce dependency; no accessibility findings in the source.
Procurement
What should contracts, pricing and exit terms secure?
Buy against documented, testable requirements and acceptance tests rather than vendor maxima.
Operating model
Which teams own the service once it runs?
Ongoing operations sit with central or departmental IT storage teams; no change in operating model is reported by the source.
Publication history
- 2026-10-29NetApp · Issue 016 resources
Stable resource ID: storagereview-netapp-novus-pnfs-ai-factory-20261001