Analyst(s): Nick Patience, Mitch Ashley, Russ Fellows
Publication Date: September 30, 2026
NetApp introduced NetApp Novus, an ONTAP-based storage architecture for AI factories, and announced its intent to acquire PEAK:AIO to strengthen the metadata technology behind it. Together with a cross-vendor AI Data Engine, agentic operations in the NetApp Console, and Keystone Sovereign, the news shows NetApp pursuing AI factory builders and enterprise agentic AI buyers at the same time.
What Is Covered in This Article:
- NetApp Novus, an ONTAP-based architecture that uses parallel NFS to separate metadata from data, targeting 100 TB/s of bandwidth across a multi-exabyte single namespace.
- NetApp’s planned acquisition of PEAK:AIO, a UK developer of metadata and parallel file system software, and how it relates to Novus.
- A new AI Data Engine release that indexes and governs metadata across both NetApp and non-NetApp storage.
- Autonomous operations and bring-your-own-LLM support in the NetApp Console, plus Keystone Sovereign for the European Economic Area.
- What the news means for NetApp’s strategy across hybrid multicloud, the agentic enterprise, and AI factories.
The News: NetApp announced a set of product updates at its Insight customer conference this week organized around three markets: hybrid multicloud, the agentic enterprise, and AI factories. The most significant is NetApp Novus, a new storage architecture for AI factories and neoclouds that runs ONTAP on dense data nodes and moves metadata into a separately scalable layer using the Flex Files capability of parallel NFS (pNFS) in NFS 4.2. NetApp says Novus targets 100 TB/s of aggregate bandwidth, enough to sustain about 2 GB/s per GPU across 50,000 GPUs, in a multi-exabyte single namespace with secure multi-tenancy, and GPU servers need only the standard Linux NFS client. Initial systems use a bandwidth-optimized variant of the AFF A90, software-defined versions will follow, and NetApp expects early customers to receive Novus within one to two months.
On the Friday before Insight began, NetApp announced its intent to acquire PEAK:AIO, a Manchester, UK-based developer of software-defined AI storage, and plans to combine its metadata and parallel namespace technology with ONTAP. NetApp did not disclose terms of the deal.
NetApp also released a version of its AI Data Engine that indexes and enriches metadata across any NFS, SMB, or S3 storage, added autonomous operations and a bring-your-own-LLM chat interface backed by an MCP server to the NetApp Console, and made a locally installed Console generally available. Keystone Sovereign, a sovereignty entitlement for NetApp’s storage-as-a-service offering, starts in Germany and France in November 2026 before expanding across the European Economic Area (EEA), while deeper Commvault integration and general availability of the Nutanix Cloud Infrastructure integration complete the list.
NetApp Novus, PEAK:AIO and NetApp’s Two-Market AI Strategy
Analyst Take: Spending on AI infrastructure is splitting into two markets with very different buyers – neoclouds and model builders running GPU clusters in the tens of thousands, and enterprises trying to get agents into production on data they already own. NetApp Novus and the planned acquisition of PEAK:AIO are NetApp’s bid for the first market, while the AI Data Engine and an agent-accessible NetApp Console target the second. We think NetApp is right to pursue both, although doing so stretches a company whose sales motion and brand were built around enterprise IT. However, the greater risk lies in the AI factory segment, where NetApp is a late entrant against specialists that have been winning neocloud deals for several years.
NetApp’s opportunity with enterprise buyers could ultimately prove larger than its AI factory ambitions, and it depends on whether the AI Data Engine earns trust by delivering policy-aware data context across diverse customer estates. If it does, NetApp can play a strategic role as enterprises operationalize AI.
NetApp Novus Tackles a Real GPU Utilization Problem
NetApp’s case for NetApp Novus rests on a mismatch between compute and storage, as a single modern GPU can consume memory bandwidth in the terabytes per second while conventional storage delivers perhaps 40 to 80 GB/s to an entire eight-GPU node. NetApp cites real-world GPU utilization as low as 5% in poorly fed clusters, and while that figure sits at the extreme end, the underlying point holds, since an idle accelerator is the most expensive asset in any data center and storage throughput feeds directly into cost per token.
Novus borrows the logic of parallel file systems such as Lustre, separating the metadata path from the data path so every GPU server can read directly from every storage node, while avoiding the proprietary client software that has made those systems hard to run outside HPC teams. Building on pNFS and the upstream Linux kernel is the right design choice, in our view, and running ONTAP on each data node gives NetApp a credible story on snapshots, replication, and multi-tenancy that neocloud operators will value. NetApp is not alone here, however, since Hammerspace has built its business on the same pNFS Flex Files approach, and VAST Data, WEKA, DDN, Pure Storage, and Dell are all selling into the same accounts. NetApp has published design targets rather than benchmark results so far, and AI factory buyers choose suppliers on measured throughput per rack unit and per kilowatt.
PEAK:AIO Buys NetApp Time on the Hardest Part
PEAK:AIO develops software-defined AI storage that runs on industry-standard hardware, with metadata technology that grew out of work with Los Alamos National Laboratory and Carnegie Mellon University, and its deployments include Los Alamos, NHS AIDE, and the Oxford Robotics Institute. NetApp describes the combined architecture as one that separates metadata from data, scales metadata services independently, and supports trillions of files through standards-based parallel NFS, which closely matches Novus even though the acquisition announcement does not name it.
We read the deal as NetApp buying proven metadata engineering for the hardest part of the Novus design, because metadata is where parallel file systems tend to struggle at AI scale, particularly with the huge numbers of small files common in training and retrieval workloads. PEAK:AIO’s software-defined heritage also matters, since many neoclouds specify their own servers and will want a version of Novus that runs on them, so the acquisition should shorten the path from NetApp’s AFF-based first release to that software-defined option. NetApp has yet to say how PEAK:AIO technology relates to the initial Novus systems, or whether PEAK:AIO’s product will continue to be sold on its own. The two companies didn’t have any existing partnerships prior to the deal announcement, so integration starts from that point.
The AI Data Engine Is NetApp’s Agentic Enterprise Play
The AI Data Engine release is the most strategically important announcement for enterprise AI buyers, because it moves NetApp beyond its own storage. Early customers found a metadata catalog covering only NetApp systems of limited use, according to NetApp, so the new release scans, extracts entities from, and enriches metadata across any NFS, SMB, or S3 store, enforces existing access controls on search and retrieval, and exposes the catalog to partner query engines such as Starburst. That puts NetApp in contention for a control point that data platform vendors, hyperscalers, and governance specialists all want, namely the governed index of what data exists, where it lives, and which agents may use it.
A cross-vendor catalog is necessary but not sufficient, since enterprise buyers also expect provenance, quality signals, sensitivity labels, data classification, and retention rules that meet compliance requirements, and NetApp has a sizable opening if the AI Data Engine becomes the context layer that governs data consistently across storage domains.
Freshness is the main technical limitation, since metadata on ONTAP and StorageGRID updates in near real time, whereas other vendors’ storage must be rescanned, so the zero-copy, always-current promise is weakest precisely where estates are most mixed, and NetApp has not given a general availability date or pricing. The NetApp Console changes support the same agentic strategy, with intent-based provisioning that stays within customer-defined storage classes, a human approval step for changes, and an MCP server that lets customers drive storage operations from their own LLMs and agent frameworks. Opening the control plane to any agent is a better approach than a closed copilot, and a locally installed Console that runs air-gapped and replaces Active IQ Unified Manager extends it to regulated and public sector sites.
NetApp faces the challenge shared by all vendors that provide access to critical data and infrastructure via MCP: how to make agent access governable at enterprise scale. CIOs and CISOs will ask the tough questions about identity binding to agents, least privilege access controls, workflow approvals, policy enforcement, and immutable audit logs. Having answers to these types of questions is the difference between “we’ll get back to you” and answers that will move deals forward with the C-suite.
Keystone Sovereign Tackles Residency, but Jurisdiction Is Harder
Keystone Sovereign keeps customer data, logs, backups, and telemetry inside the EEA, along with support and escalation staff, access and management processes, and the contracts that govern them, and NetApp positions it as addressing jurisdiction as well as geography. European demand for these assurances is growing fast, and NetApp is right to offer controls that customers can audit. The harder question, however, is jurisdictional reach, since NetApp remains a US-headquartered company, and in-region staffing and contracts do not by themselves settle whether a foreign government could compel access, so we expect buyers to press NetApp on how its contracts handle that exposure.
What the News Means for NetApp’s Strategy
NetApp’s strategic argument is that one storage operating system, ONTAP, can now run from branch-office systems through the disaggregated AFX platform to GPU clusters with thousands of storage nodes, and it enters this push from a position of strength, with NetApp’s Q1 FY 2027 results showing net revenue of $2.03 billion, up 30% year over year. That consistency will appeal to enterprise buyers who want one operating model across data center, cloud, sovereign, and AI environments, and the Commvault, Nutanix, and JetStream moves help keep that installed base loyal as customers reassess VMware. AI factory buyers care far less about operational consistency than about throughput per watt, however, so NetApp Novus will succeed or fail on published performance and the speed of the PEAK:AIO integration, while the AI Data Engine will be judged on how open it proves to be in practice.
What to Watch:
- Whether NetApp publishes performance results for NetApp Novus at GPU-cluster scale, and how they compare with established AI storage specialists on throughput per rack unit and per kilowatt, sustained read/write bandwidth, latency, and time to deploy.
- How quickly PEAK:AIO technology reaches a software-defined NetApp Novus release once the deal closes, and whether PEAK:AIO’s research and healthcare customers keep a supported standalone product.
- How NVIDIA certifies and positions Novus within its AI factory reference designs, which heavily influence neocloud purchasing.
- Whether partners beyond Starburst adopt the AI Data Engine catalog, and how NetApp prices cross-vendor indexing.
- How NetApp answers questions on jurisdictional exposure as Keystone Sovereign expands beyond Germany and France.
See the complete press release on NetApp’s September 2026 product announcements, along with NetApp’s announcement of its intent to acquire PEAK:AIO, on the NetApp website.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Other Insights From Futurum:
NetApp Q1 FY 2027: AI-Ready Storage Drives Enterprise Momentum
NetApp StorageGRID 12.1 Scales Object Storage for AI Factories
Does the NetApp-Commvault Partnership Signal a Paradigm Shift for Backup?
Author Information
Nick Patience is VP and Practice Lead for AI Platforms at The Futurum Group. Nick is a thought leader on AI development, deployment, and adoption - an area he has researched for 25 years. Before Futurum, Nick was a Managing Analyst with S&P Global Market Intelligence, responsible for 451 Research’s coverage of Data, AI, Analytics, Information Security, and Risk. Nick became part of S&P Global through its 2019 acquisition of 451 Research, a pioneering analyst firm that Nick co-founded in 1999. He is a sought-after speaker and advisor, known for his expertise in the drivers of AI adoption, industry use cases, and the infrastructure behind its development and deployment. Nick also spent three years as a product marketing lead at Recommind (now part of OpenText), a machine learning-driven eDiscovery software company. Nick is based in London.
Mitch Ashley is VP and Practice Lead for the CIO & Technology Buyers and Software Lifecycle Engineering practices at The Futurum Group. A multi-time CIO and CTO with 30+ years leading technical organizations, Mitch built and operated production systems spanning cybersecurity for the U.S. Department of Defense, PKI services for the broadband and 5G industries, SaaS platforms, large-scale telecom and banking systems, and a national broadband network. His work with AI began early, developing expert systems that diagnosed and repaired complex mainframe environments. That operator foundation grounds his analysis in operational consequence, covering the technology buyer's world of software engineering, cybersecurity, DevOps, cloud, and AI.
