# NetApp Debuts Zettabyte-Scale Storage, Partners with Oracle Cloud > Source: > Published: 2026-09-30 17:00:20+00:00 TL;DR — Key Takeaways – NetApp unveiled Novus, a disaggregated storage architecture designed to supply data to AI factories running tens of thousands of GPUs. – Novus separates metadata management from the data path, allowing each to scale independently while maintaining a single namespace. – NetApp says the architecture can scale beyond 100 TB/s of aggregate throughput and support environments with hundreds of thousands of GPUs. NetApp has unveiled Novus, a new storage architecture created to deliver the massive data throughput required by AI factories operating tens of thousands of GPUs. NetApp touts Novus as the world’s fastest storage system, with an architecture that can exceed 100 TB per second of aggregate throughput, though that claim hasn’t been verified with an industry-wide benchmark. GPUs can process data only as fast as storage infrastructure can deliver it. NetApp says GPU utilization in large AI environments can fall below 30% when storage can’t provide data quickly enough. Given the costs of GPU clusters, idle capacity gets expensive quickly. This bottleneck gets more difficult as data centers expand capacity. While earlier clusters might contain around 10,000 GPUs, newer deployments are moving toward 50,000 or more. NetApp says a 50,000-GPU cluster could require 100 Tbps of cumulative bandwidth. Novus takes a disaggregated approach to this problem, separating metadata management from the data path. This allows bandwidth and metadata resources to expand independently while data remains accessible through a single namespace using NFS. The initial Novus implementation combines Novus Data Director metadata software running on qualified Supermicro servers with NetApp AFF A90 systems providing ONTAP data services. The company says the architecture can support high utilization across hundreds of thousands of GPUs. Technology research firm Omdia observed testing that showed near-linear performance gains as additional ONTAP clusters were connected to a single global namespace. Based on those tests, Omdia modeled the architecture scaling to 100 TB per second of sequential read throughput with dozens of exabytes of effective capacity, with a path toward zettabyte-scale deployments. Novus is part of NetApp’s effort to expand beyond traditional enterprise storage into the specialized infrastructure needed by neoclouds and GPU-as-a-Service providers. These operators need to keep costly GPU resources working at high utilization while handling enormous AI training and inference datasets. ### **Alliance with Oracle** NetApp is also extending its storage platform deeper into the public cloud. NetApp and Oracle announced OCI NetApp Storage Service, a fully managed native Oracle Cloud Infrastructure offering based on NetApp ONTAP. The planned service will allow customers to move ONTAP-based workloads into OCI without completely redesigning their storage operations. It will support databases, virtualized environments, regulated applications, HPC workloads and AI data pipelines. Admins will be able to manage the service through the OCI Console and SDKs as well as ONTAP APIs. General availability is planned within the next 12 months. The Oracle agreement gives NetApp a native ONTAP offering in another major public cloud, complementing its existing services for AWS, Microsoft Azure and Google Cloud. Bottom line: Novus is focused on feeding data to extremely large GPU deployments, while the OCI service extends the company’s existing data management environment to another cloud platform.