NetApp used its Insight conference in Las Vegas to unveil Novus, a storage architecture it claims can scale to up to 100TBps of throughput to keep artificial intelligence (AI) factory graphics processing units (GPUs) fed.
Meanwhile, NetApp chief executive George Kurian (pictured) argued that data – not compute – now forms the main constraint on enterprise AI, and presented Novus alongside new European data sovereignty controls, an expanded AI Data Engine and a fresh Oracle partnership.
NetApp said neoclouds and GPU-as-a-service providers build AI factories at a scale legacy storage was never designed for, and that GPU utilisation can fall below 30% when storage cannot feed enough data to the processors.
Novus, he said, separates metadata from the data path so performance, capacity and concurrency scale independently under a single NFS namespace, combining Novus Data Director metadata software on Supermicro hardware with OnTap data services on AFF A90 systems. NetApp said the architecture is available to order now and provides a path to software-defined deployments.
“Data is becoming the constraint to productively scale AI,” said Kurian. “93% of organisations report that at least one form of data challenge is blocking their AI ambitions.”
Kurian put the scale of that challenge in numbers, noting that large language model scale has grown by 5,000 times in four years, and that about 85% of a typical enterprise’s data sits in unstructured form – engineering documentation, research data, media and decades of institutional knowledge scattered across siloed systems. “Petabytes of your unstructured data are dark, with hidden knowledge ignored,” he added.
Syam Nair, NetApp’s chief product officer, framed the launch as an economics problem. “AI factories struggle and GPU economics collapse when data can’t keep up,” he said. “HPC [high-performance computing]-era storage alone cannot serve AI factory scale – the industry needs a true architecture built for it from the ground up.”
Meanwhile, NetApp Keystone Sovereign adds controls beyond data residency, including European-based support and escalation paths, European-controlled access management and clearer documentation of telemetry and data flows, with initial pilots planned for Germany and France.
“Fragmented data, dark data, uneven protection, facing rising attacks and hardening sovereignty rules – this is the state of most customers’ data estates,” said Kurian.
Kurian framed the announcements around what he called the agentic enterprise, in which autonomous software agents act on an organisation’s behalf. Those agents need “memory and context”, he argued, and turning data into trusted, governed context is the job he outlined for the AI Data Engine.
The expanded AI Data Engine lets organisations discover, govern and activate data where it lives without creating copies, with metadata capability that now spans heterogeneous sources beyond NetApp storage. NetApp has made the AI Data Engine free for the first six months, while an expanded Commvault integration aims to detect threats earlier and recover clean data in minutes rather than days or weeks. And a newly expanded partnership makes OnTap storage available on Oracle Cloud Infrastructure.
Simon Robinson, chief analyst for storage and data infrastructure at Omdia, said data sovereignty is becoming a defining requirement in regulated markets. “Solutions that can provide that clarity while preserving the flexibility of as-a-service infrastructure are well positioned to build trust and expand adoption in sovereignty-sensitive regions,” he said.
