NetApp and Nvidia rethink storage for AI factories
SiliconANGLE Chad Wilson ● Covered by 5 sources
NetApp and Nvidia are redesigning storage for AI factories. They want data and metadata to scale separately so GPUs don’t sit idle.
Based on reporting by SiliconANGLE, Chad Wilson — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
NetApp and Nvidia are taking another run at the storage layer, this time with AI factories in mind. The two companies have worked together for more than a decade, and that partnership now includes co-engineering around AI infrastructure.
The core idea behind NetApp’s Novus architecture is simple to say and harder to build: separate data and metadata so they don’t fight over the same resources. Arindam Banerjee, NetApp’s chief platform and technology officer, said traditional systems were often good at either transactional work or big sequential transfers, but not both at once. AI workloads, with checkpointing and metadata activity happening alongside heavy data movement, are forcing both to happen together.
That matters because storage delays can leave GPUs waiting around while still burning power. Banerjee said Novus is designed so data and metadata can scale independently, each on different axes, instead of being trapped on the same media and network. The goal is to keep the GPUs fed and make the whole setup more efficient.
Jason Hardy, Nvidia’s vice president of storage technology, said the pressure is moving AI systems from experiments into mainstream production. Fine-tuning, inference, capacity and AI-ready data all have to fit into the same evolving infrastructure as use cases change. And AI agents are adding a new problem: thousands of them may hit storage at once.
Banerjee said that kind of concurrency changes the game as much as raw throughput. The permissions may be temporary and narrow, but the metadata traffic piles up fast. His answer is to make storage easier to consume through APIs and SDKs, because AI teams are not storage engineers. NetApp is also building a new control plane for Novus that is meant to be API-driven and friendly to agents.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI that gets less glossy than the demo videos: storage is where the bill comes due. Everyone loves talking about GPUs, but the real fight is making data and metadata stop tripping over each other like interns in a server room. If AI factories are the pitch, then boring old plumbing just became the main event.
Read more about this at: SiliconANGLE