NetApp Novus: Inside NetApp’s Biggest Platform Bet — 100 TB/s Storage, Four Clouds and Agentic Data

Recently, NetApp shipped the densest product slate of its recent history at INSIGHT 2026 in Las Vegas this week. NetApp Novus is a new storage architecture for AI factories. An Oracle partnership puts NetApp’s ONTAP software natively into a fourth hyperscale cloud. An expanded AI Data Engine reaches into rivals’ storage to make data agent-ready. And a sovereign service tier targets European regulators.

Chief Product Officer Syam Nair called Novus the fastest storage on the planet. Company framing at the event called it the most important launch since Data Fabric in 2013. The first claim does not survive scrutiny. The second is plausible.

What NetApp actually announced

Four releases landed together. Novus is a disaggregated architecture for neoclouds and GPU-as-a-service providers. The Oracle deal brings a fully managed, OCI-native NetApp storage service. AI Data Engine (AIDE) now discovers and governs data in place across ONTAP, StorageGRID and non-NetApp storage. A deeper Commvault integration links ransomware detection to clean recovery.

The through-line is one data platform stretching from GPU-factory throughput to agent-facing governance. That is the strategy. The details are where the story gets interesting.

The Novus architecture, minus the marketing

Novus separates metadata from the data path, so performance, capacity and concurrency scale independently under a single namespace. The initial release pairs two components. Novus Data Director, the metadata software, runs on qualified Supermicro servers. ONTAP data services arrive through AFF A90 all-flash systems.

Everything speaks standard NFS and pNFS. It requires no proprietary client. That standards choice matters: AI infrastructure buyers have watched storage vendors lock them into custom protocols before. NetApp is betting on familiarity — ONTAP’s enterprise data services, carried over plain NFS, attract customers who already run its stack.

The ONTAP foundation also means no new silo. A software-defined version is planned for flexibility in deployment. Context: days before the conference, NetApp moved to acquire Peak:AIO, a parallel-NFS specialist. That technology maps directly onto Novus’s metadata layer.

Is it really the fastest storage on the planet?

Here the claims need careful weighting. The 100 TB/s figure is an architecture target. Omdia audited demonstrations showing near-linear performance scaling as ONTAP clusters were added to a single global namespace. Its modeling then projects the architecture reaching 100 TB/s of sequential read throughput, with dozens of exabytes of capacity. NetApp says tests achieved the figure and that it is sustainable.

Verified today: near-linear scaling. Projected: the headline number.

Rivals contest the performance crown in any case. Pure Storage’s FlashBlade//EXA topped an MLPerf benchmark earlier this month. WEKA’s WEKApod appliance delivers 10.2 TB/s per rack and 1.1 exabytes of effective capacity in a single rack. Different metrics — per-rack density versus federation-wide aggregate. The field is crowded regardless: DDN, VAST Data, WEKA and Pure Storage all chase the same AI-factory workloads.

The scale math explains the design point. Clusters are moving from roughly 10,000 GPUs toward 50,000 and beyond. At about 2 GB/s per GPU, a 50,000-GPU cluster needs around 100 TB/s of cumulative bandwidth. Novus was drawn to that number deliberately.

One more figure needs its label: NetApp says GPU utilization “can drop below 30 percent” when storage starves compute. That is the company’s own number. Independent reporting suggests worst cases run far lower. Either way, idle GPUs are the economic problem Novus exists to solve.

The Oracle deal: four clouds, one asterisk

With OCI NetApp Storage Service, NetApp becomes the only storage vendor with first-party offerings in the world’s four largest clouds. The precedents are real. Azure NetApp Files arrived in 2017 as the industry’s first hyperscaler-native storage service. Amazon FSx for NetApp ONTAP and Google Cloud NetApp Volumes followed; Google says Volumes now reaches 200 GB/s.

The asterisk: the OCI service is planned for general availability within 12 months. It is announced, not shipping. Both companies wrapped the launch in future-product disclaimers that expressly warn against purchasing decisions based on it. Today the claim is three clouds plus a promise — and Oracle customers gain nothing concrete until it lands.

When it does, the value is straightforward. ONTAP workloads move to OCI without application redesign: databases, EDA, HPC, regulated applications. For Oracle, native enterprise file storage widens OCI’s lift-and-shift appeal. For NetApp, it removes the last gap in its cloud coverage story.

The agentic layer: AIDE and Commvault

The quieter announcement may matter most. AIDE now indexes metadata across NFS, SMB and S3 repositories — including storage NetApp does not sell. Data is activated in place, zero-copy, rather than being moved through pipelines into AI systems. Previewed next: analytics integrations with Starburst and Onehouse.AI, and agentic services that turn governed data into AI context.

The Commvault integration applies the same logic to ransomware. NetApp’s detection signals on primary storage now connect to Commvault’s recovery orchestration. The package adds cleanroom validation and a metric the companies call mean time to clean recovery.

The strategic bet: agents need trusted, current, governed data at machine speed. And the storage layer — where the data already lives — is the cheapest place to provide it. Rivals building copy-first AI pipelines pay in latency, cost and staleness.

Sovereignty and autonomous operations

Keystone Sovereign gives qualified EEA customers data-residency and regional controls on a storage-as-a-service model. NetApp Console gains Autonomous Operations, an outcome-based automation layer that runs in air-gapped environments. Fleet Management and an AI ChatOps interface round it out. ChatOps works through an open LLM gateway: customers bring their own model, with policy guardrails on every action. A Nutanix collaboration goes generally available this fall.

What this means for infrastructure buyers

Three takeaways. First, treat 100 TB/s as a design target and demand sustained-throughput evidence for your own workload before sizing. The audited fact is near-linear scaling; the rest is modeling. Second, discount the four-cloud claim accordingly — the OCI service is a roadmap item until it ships.

Third, watch AIDE. Its heterogeneous, zero-copy, storage-layer governance reaches beyond NetApp’s own estate. That reach — turning every vendor’s storage into agent-ready data — is what could actually change procurement.

Editor’s Note

This article draws on NetApp‘s four INSIGHT 2026 press releases (29–30 September 2026) and on conference reporting by Blocks & Files, Techzine, IT Pro and Investing.com. Further sources: Omdia’s audit statement, NetApp’s INSIGHT event page, and Blocks & Files’ July 2026 coverage of WEKA.

Vendor statements: the “fastest storage” claim, the 100 TB/s figure, the GPU-utilization figure and the Commvault recovery times. Omdia’s characterization is a projection from observed near-linear scaling, not a sustained audited benchmark. The OCI service is announced but not generally available. The interpretation of the AIDE procurement impact is TechRecast’s own.