NetApp intends to acquire PEAK, a specialist in parallel file systems, as it works on storage architecture for growing AI clusters. The deal stage is an announced intention, subject to closing conditions and regulatory approval. No purchase price was disclosed in the sources, and no timetable for a combined product was given.
According to the source, the proposed deal would bring PEAK's metadata technology into a future design built around NetApp ONTAP. NetApp says the proposed architecture would separate metadata services from stored data, allowing each to scale independently. PEAK's parallel NFS approach is meant to give many clients shared access without turning the coordination layer into a choke point.
The source describes the target problem as GPUs left waiting while shared storage locates and delivers training data. A cluster handling an enormous number of small files can spend substantial time on file lookups and coordination, even when its storage media can move data quickly. The source notes that better storage behavior can improve utilization if metadata requests, shared namespace operations, and file delivery are what hold a workload back, but it will not cure bottlenecks in networking, preprocessing, application code, or the way a training job divides work. NetApp has not published a measured reduction in GPU stalls for an integrated NetApp ONTAP and PEAK system.
NetApp describes an eventual architecture capable of supporting trillions of files and deployments measured in multiple exabytes. The source states those figures describe an ambition, not a tested configuration available today.
The source also raises integration questions: how much of PEAK's design survives integration with NetApp ONTAP, whether existing customers get a migration path and compatibility details, and which features work across the combined system.