NetApp and Nvidia are collaborating on a new storage architecture called Novus, designed to meet the demands of AI factories. The system separates data and metadata management, allowing each to scale independently based on specific workload requirements.
The architectural shift addresses the inefficiency of traditional storage, where transactional metadata operations often queued behind large data transfers. By decoupling these functions, the system prevents bottlenecks that lead to GPU underutilization and wasted power.
The companies are focusing on handling concurrent workloads, such as thousands of AI agents accessing data simultaneously. This requires higher concurrency capabilities alongside traditional throughput, as agentic workflows increase the volume of metadata operations.
NetApp is developing a new control plane for Novus that prioritizes API-driven consumption. This approach aims to simplify infrastructure management for AI teams, allowing them to interact with storage via APIs and SDKs without needing deep storage engineering expertise.
