NetApp used its INSIGHT 2026 conference in Las Vegas to announce NetApp Novus, a disaggregated storage architecture for AI factories that separates metadata from the data path and that the company says is designed to exceed 100 TB/s of aggregate throughput, keep hundreds of thousands of GPUs fed, and grow toward a zettabyte-scale file system. The first configuration pairs NetApp Novus Data Director, the metadata software, running on qualified Supermicro servers with NetApp AFF A90 systems running ONTAP as the data layer, and NetApp says it is orderable today. The announcement comes a week after NetApp agreed to acquire PEAK:AIO for its scale-out pNFS metadata work.
NetApp Novus Separates Metadata and Data Scaling
Novus keeps a single, standards-based NFS namespace while moving metadata services into their own tier. Data Director runs on standard x86 compute and handles the namespace and layout work (NetApp’s architecture diagram lists OPEN, LAYOUTGET, and LAYOUTCOMMIT on the Data Director path), while AFF A90 HA pairs serve as Novus data nodes that carry capacity and bandwidth. Clients are stock Linux pNFS clients on the GPU servers’ front-end NICs, using NFSv4.2 with pNFS Flex Files, nconnect, and GPUDirect Storage per NetApp’s product page, so there’s no proprietary client to install.
The pNFS data path rides a lossless Ethernet client fabric that is separate from the GPU backend fabric used for collectives, and NetApp’s diagram makes a point of saying Novus is not in that backend path. Hosts mount once and data nodes can be added without remounting, which lets performance, capacity, and metadata concurrency scale independently. NetApp positions that for multi-tenant AI factories run by neocloud and GPU-as-a-Service providers.
In a pre-briefing ahead of INSIGHT, Jeff Baxter, NetApp’s VP of Technical Marketing, described the Data Director as answering lookups and layout requests out of band and in memory, so every GPU client can address every data node directly without being proxied through a cluster interconnect, which is the parallel file system behavior NetApp is after without a Lustre-style agent on the client.
He also said the first data nodes are a specific variant of the AFF A90 tuned for sequential bandwidth with additional NVRAM, with a software-defined ONTAP build for a select group of x86 servers to follow, and that Novus slots in above the AFX disaggregated platform in NetApp’s lineup for the point where a disaggregated cluster’s networking becomes the limit.
NetApp frames the problem in utilization terms: “Under traditional architectures, GPU utilization can drop below 30 percent when AI factories cannot feed enough data to their GPUs,” the release says. NetApp is also billing Novus as “the fastest storage on the planet,” a claim that rests on the projections below.
The scaling figures are projections built on a smaller audited test. NetApp says tests audited by Omdia showed near-linear performance scaling as ONTAP clusters were added to a single global namespace, and that Omdia’s modeling projects the architecture can scale to 100 TB/s of sequential read throughput with dozens of exabytes of effective capacity. “We are confident in our projections that the solution can scale to the needs of zettascale deployments,” said Tony Palmer, Chief Analyst, Omdia.
NetApp’s own sizing on the Novus page works out to more than 50,000 GPUs at about 2GB/s each for the 100 TB/s figure, and the company says the design preserves ONTAP’s data services, security, quality-of-service, and multi-tenancy while supporting billions of files across thousands of storage nodes. NetApp also says the initial hardware-bound configuration “provides a path to software-defined deployments over time.”
On timing, Baxter said the intent is to have Novus in the hands of lighthouse customers within a month or two of INSIGHT and to roll it out to other AI factory customers in the months after.
He put the target population on the order of a hundred companies running tens of thousands to hundreds of thousands of GPUs, which told NetApp they need exabyte-class single namespaces that can grow toward zettabytes, tens to hundreds of TB/s of cumulative bandwidth, density per rack unit, multi-tenancy, and provisioning through their own control planes over APIs and SDKs. “Probably the most expensive asset in any data center in the world is an unused GPU in an AI factory,” he said.
“AI factories struggle and GPU economics collapse when data can’t keep up,” said Syam Nair, Chief Product Officer at NetApp.
Supermicro’s role goes beyond hosting Data Director on qualified servers alongside AFF A90 data nodes: the two companies are pairing Supermicro’s rack-scale compute, networking, power, and liquid-cooling work with NetApp’s data management and cyber-resilience software for AI factory builds. They are also jointly supporting NetApp AIPod with Supermicro, a NetApp-validated converged infrastructure architecture based on NVIDIA Reference Architecture guidance for training, inference, and agentic workloads at the dozens-to-hundreds-of-GPUs scale, sold through NetApp’s channel partners.
AI Data Engine Expands Metadata Discovery
NetApp AI Data Engine, or AIDE, now extends metadata discovery across ONTAP, StorageGRID, and non-NetApp storage. NetApp says AIDE can discover and analyze data across NFS, SMB, and S3 repositories in place, so IT teams can find AI-ready datasets, storage optimization opportunities, migration candidates, and security risks without first copying data into a separate AI pipeline.
The company is also previewing AI-powered data understanding, deeper governance intelligence, open analytics integrations with Starburst and Onehouse.AI, and agentic services that turn data into AI context at scale. Those remain in preview, and NetApp hasn’t given a general-availability date.
NetApp has expanded its Commvault integration so that autonomous ransomware protection signals from primary storage feed directly into Commvault recovery workflows. According to NetApp, the joint solution provides near-real-time threat detection, deeper scans for evasive threats, cleanroom validation with AI-based threat detection in an isolated environment, and clean recovery from the Commvault console, which the companies say cuts recovery to minutes, from days or weeks, without reintroducing the infection.
In the pre-briefing, Baxter said the integration also lets Commvault use ONTAP’s native mechanisms where they apply: Snapshot copies in place of streaming backups, SnapMirror for replication, SnapLock for an immutable vault on the storage side, and SnapRestore for recovery when a primary snapshot still exists.
Hybrid Multi-Cloud Operations and Sovereignty Controls
NetApp Console Autonomous Operations replaces manual approval loops with outcome-based automation under human oversight for large hybrid fleets, and NetApp says it runs in air-gapped environments. The company describes an API-first design and agentic orchestration behind predictive analytics, capacity optimization, incident remediation, and policy-driven operations.
NetApp Console Fleet Management adds a second Console delivery model, installed in the customer’s own environment, for management, observability, and AI across the whole storage estate. Baxter said the local install can run fully disconnected for dark sites and is meant to bring the customers who stayed on ActiveIQ Unified Manager for on-premises fleet management onto the same Console as the hosted service.
The new NetApp Console AI ChatOps Interface brings conversational storage operations to ONTAP systems, covering intent-based provisioning, automated remediation, policy guardrails, reporting, and predictive analytics. It runs through an open LLM gateway, so customers can plug in their own model, with every action constrained to defined storage classes, policies, and governance boundaries. Baxter described the chat interface as a front end to an MCP server built into the platform, so the same provisioning and operations can be driven by a customer’s own agents, with the REST API, Terraform, and Ansible paths staying in place.
NetApp Keystone Sovereign extends the storage-as-a-service offering with regional controls for qualified European Economic Area customers. The service keeps applicable customer data, logs, backups, and telemetry, along with support and escalation processes and access and management workflows, within defined European boundaries, and documents the telemetry, monitoring, and data flows involved.
Pilots start in Germany and France, with more countries planned; Baxter put the start in November and said support and escalation will be handled in region by passport holders of that region, the same model NetApp uses for its U.S. public sector offering.
NetApp and Nutanix also plan an integrated offering that combines NetApp ONTAP with Nutanix 云基础设施 for virtualized workloads. NetApp says general availability is scheduled for this fall.
SAP and OCI Storage Integrations Remain Forward-Looking
NetApp and SAP announced an intent to explore deeper cloud-native integration of NetApp capabilities into SAP Cloud Infrastructure, SAP’s IaaS platform. The NetApp Platform already provides file and block storage, data protection, and replication there; the new work covers joint go-to-market programs for migrating and operating SAP and non-SAP workloads and a technical validation of NetApp object storage on SAP Cloud Infrastructure. It is an intended expansion of the partnership with no timeline, and NetApp’s release carries its usual caution that unreleased offerings should not be relied on for purchasing decisions.
NetApp and Oracle announced the OCI NetApp Storage Service, a planned, fully managed, OCI-native storage service that brings ONTAP data management to Oracle Cloud Infrastructure. Customers would manage it through the OCI Console, OCI SDKs, and ONTAP APIs, with NetApp naming AI data pipelines, databases, enterprise applications, virtualized environments, EDA and HPC, and regulated workloads as targets. It is planned for general availability within the next 12 months.
“OCI NetApp Storage Service gives customers a native OCI experience while allowing them to use the ONTAP familiar capabilities for data protection, migration, governance, and multiprotocol access,” said Pravjit Tiwana, Senior Vice President and General Manager, Cloud Storage and Services, NetApp.
Amazon