NetApp goals Novus storage at AI manufacturing facility bottlenecks
NetApp used its Perception convention in Las Vegas to unveil Novus, a storage structure it claims can scale to as much as 100TBps of throughput to maintain synthetic intelligence (AI) manufacturing facility graphics processing items (GPUs) fed.
In the meantime, NetApp chief government George Kurian (pictured) argued that knowledge – not compute – now varieties the primary constraint on enterprise AI, and introduced Novus alongside new European knowledge sovereignty controls, an expanded AI Knowledge Engine and a recent Oracle partnership.
NetApp mentioned neoclouds and GPU-as-a-service suppliers construct AI factories at a scale legacy storage was by no means designed for, and that GPU utilisation can fall under 30% when storage can not feed sufficient knowledge to the processors.
Novus, he mentioned, separates metadata from the info path so efficiency, capability and concurrency scale independently below a single NFS namespace, combining Novus Knowledge Director metadata software program on Supermicro {hardware} with OnTap knowledge providers on AFF A90 programs. NetApp mentioned the structure is on the market to order now and gives a path to software-defined deployments.
“Knowledge is turning into the constraint to productively scale AI,” mentioned Kurian. “93% of organisations report that at the least one type of knowledge problem is obstructing their AI ambitions.”
Kurian put the dimensions of that problem in numbers, noting that giant language mannequin scale has grown by 5,000 occasions in 4 years, and that about 85% of a typical enterprise’s knowledge sits in unstructured kind – engineering documentation, analysis knowledge, media and many years of institutional data scattered throughout siloed programs. “Petabytes of your unstructured knowledge are darkish, with hidden data ignored,” he added.
Syam Nair, NetApp’s chief product officer, framed the launch as an economics downside. “AI factories battle and GPU economics collapse when knowledge can’t sustain,” he mentioned. “HPC [high-performance computing]-era storage alone can not serve AI manufacturing facility scale – the business wants a real structure constructed for it from the bottom up.”
In the meantime, NetApp Keystone Sovereign provides controls past knowledge residency, together with European-based assist and escalation paths, European-controlled entry administration and clearer documentation of telemetry and knowledge flows, with preliminary pilots deliberate for Germany and France.
“Fragmented knowledge, darkish knowledge, uneven safety, going through rising assaults and hardening sovereignty guidelines – that is the state of most prospects’ knowledge estates,” mentioned Kurian.
Kurian framed the bulletins round what he referred to as the agentic enterprise, by which autonomous software program brokers act on an organisation’s behalf. These brokers want “reminiscence and context”, he argued, and turning knowledge into trusted, ruled context is the job he outlined for the AI Knowledge Engine.
The expanded AI Knowledge Engine lets organisations uncover, govern and activate knowledge the place it lives with out creating copies, with metadata functionality that now spans heterogeneous sources past NetApp storage. NetApp has made the AI Knowledge Engine free for the primary six months, whereas an expanded Commvault integration goals to detect threats earlier and recuperate clear knowledge in minutes relatively than days or even weeks. And a newly expanded partnership makes OnTap storage accessible on Oracle Cloud Infrastructure.
Simon Robinson, chief analyst for storage and knowledge infrastructure at Omdia, mentioned knowledge sovereignty is turning into a defining requirement in regulated markets. “Options that may present that readability whereas preserving the flexibleness of as-a-service infrastructure are effectively positioned to construct belief and broaden adoption in sovereignty-sensitive areas,” he mentioned.
