PEAK:AIO

PEAK:AIO is a software-defined storage platform for feeding data to GPU clusters used in AI, machine learning, and high-performance computing workloads. The software is deployed on NVMe-equipped servers from hardware partners such as Dell, HPE, and PNY, creating what the vendor calls an AI Data Server. PEAK:AIO supplies the storage layer; it does not provide GPU compute, model training software, or an inference runtime.

The platform presents shared NVMe capacity through NFS 3 and 4 over TCP or RDMA, allowing it to operate as high-performance network attached storage without proprietary client software. It can also expose individual high-performance volumes through NVMe over Fabrics, supporting block storage for single-host, scratch-space, database, and non-shared workloads. Administrators can begin with a directly attached storage server and introduce Ethernet or InfiniBand switching as additional GPU systems are added.

PEAK:AIO supports NVIDIA GPUDirect Storage, enabling data transfers between storage and GPU memory with less CPU involvement and fewer intermediate memory copies. Supported configurations use NVIDIA ConnectX network interfaces and NVMe-oF or NFS over RDMA to reduce latency and increase throughput. The vendor’s 2024 data-server specifications list configurations ranging from approximately 80 TB to 3 PB of raw NVMe capacity in a typical 2U server, with up to 80 GB/s using two ConnectX-7 interfaces; actual capacity and throughput depend on server, drive, network, protocol, and workload configuration.

Data protection is provided through PEAK:PROTECT, with RAID 0, 10, 5, and 6/N+2 options. Storage can be expanded by adding drives or additional server nodes, giving the platform scale-out storage characteristics alongside its single-node enterprise flash array storage capabilities. A browser-based interface provides remote administration.

For larger AI and HPC environments, the related PEAK Open pNFS architecture uses standard Linux pNFS Flex Files clients and separates metadata operations from direct data-node access. Clients can read from the appropriate NVMe data server over RDMA or TCP, allowing capacity and aggregate throughput to grow without installing a proprietary parallel-file-system client. PEAK Open pNFS and the separately marketed PEAK:ARCHIVE product should not be assumed to be included with every AI Data Server deployment.

PEAK:AIO is a Manchester, UK-based software-defined AI storage company that NetApp announced plans to acquire on September 25, 2026. NetApp intends to combine PEAK:AIO’s scalable metadata services, global namespace, and parallel NFS architecture with ONTAP to support shared storage for large GPU clusters.