NetApp and Nvidia rethink storage for AI factories

Storage systems are being redesigned for artificial intelligence factories, a shift driven by the unique demands of AI workloads. Traditional enterprise storage was built for predictable workloads, but AI introduces heavy data movement alongside transactional metadata activities, often on shared infrastructure. NetApp and Nvidia have been collaborating on this challenge for over a decade, co-engineering AI infrastructure solutions.

NetApp's Novus architecture separates data and metadata functions, allowing each to scale independently based on varying workload demands. This separation prevents smaller, transactional metadata operations from competing with large data transfers for resources, addressing inefficiencies caused by storage delays. By scaling each component independently, GPUs are better utilized and power consumption optimized.

The goal is to create a flexible, scalable system that can adapt to changing demands without requiring a complete system redesign. AI agents, which may rapidly access data with temporary, narrowly scoped permissions, add another layer of complexity. Their concurrent access introduces new concurrency requirements, further redefining storage architecture for the future.

To make this transition smoother, NetApp is focusing on creating an API-driven, agent-friendly control plane. This approach removes complexity, enabling AI teams to consume storage without needing storage expertise. TheCUBE is a paid media partner for NetApp INSIGHT, ensuring coverage of this evolving storage landscape.