NetApp has agreed to acquire PEAK:AIO, the Manchester, UK software company whose metadata and parallel NFS work NetApp says will accelerate its AI infrastructure roadmap. NetApp plans to combine PEAK:AIO’s metadata services and parallel namespace technology with ONTAP, aiming at shared storage that scales alongside growing GPU clusters, in an architecture it says is intended to support trillions of files and multi-exabyte deployments. Terms weren’t disclosed, and the deal is subject to customary closing conditions and regulatory approvals.
“AI clouds need high-performance shared storage that can scale alongside growing GPU clusters while maintaining the resilience, security, and operational simplicity organizations depend on,” said George Kurian, CEO of NetApp. “With PEAK:AIO, we are delivering on our vision for a new generation of AI infrastructure that combines scalable metadata services with the proven foundation of ONTAP to help customers maximize infrastructure efficiency and support AI at hyperscale cloud.”
Metadata That Scales on Its Own
NetApp describes an architecture that disaggregates metadata from data so metadata services can scale independently of capacity. PEAK:AIO brings metadata scale, a global namespace, and parallel NFS access for massively parallel workloads, and NetApp brings ONTAP as the data foundation, with the resilience, security, and operational maturity its customers already run. NetApp says the combination is meant to reduce data-related GPU stalls and give existing ONTAP customers a clear evolution path.
The most important piece of PEAK:AIO’s metadata work is pNFS Lattice, an open-source, scale-out metadata server for NFSv4.2 and pNFS Flex Files that PEAK:AIO initiated with Los Alamos National Laboratory and Carnegie Mellon University and published with support from the Linux Foundation. Lattice splits metadata authority, protocol coordination, and data placement into separate layers: multiple user-space metadata server daemons share a RonDB metadata store, clients reach the right metadata server through standard NFSv4 referrals, and the data servers are unpatched Linux knfsd instances serving flex-file layouts over TCP or RDMA. The code is on GitHub under an MIT license, with one GPL-2.0 file where it links to RonDB. PEAK:AIO labels it an early release for research, development, and testing, not yet recommended for production. LANL and PEAK:AIO are scheduled to present Lattice at the SNIA Developer Conference in Santa Clara September 28 to 30, including a Birds of a Feather session led by LANL’s Gary Grider.
“NetApp and PEAK:AIO connected through the shared belief that the AI era demands a fundamentally new approach to data infrastructure,” said Mark Klarzynski, CTO and founder of PEAK:AIO. “What excites us most about joining NetApp is the opportunity to pair our culture of innovation with the reach, scale, and customer trust of a global leader.”
What We’ve Seen PEAK:AIO Do
We’ve followed PEAK:AIO closely for years, and the common thread is software that gets a lot out of standard servers. In 2023, PEAK:AIO’s AI Data Server pushed 80GB/s from a single storage node. In 2024, we visited London Zoo, where the 런던 동물원 학회 runs two NVIDIA DGX systems against 1.2PB of PEAK:AIO storage for wildlife monitoring, and we covered its work with MONAI and Solidigm to keep medical AI on premises inside hospitals. In 2025, we tested a 2U AI Data Server with 1.5PB and 120GB/s on Dell hardware and Solidigm 61TB SSDs, followed by a CXL-based Token Memory Platform for KV cache reuse.
우리의 Dell PowerEdge R7725xd review, the server sustained over 300 GB/s across its 24 Gen5 NVMe drives locally, and PEAK:AIO served 160 GB/s over NVMe-oF RDMA to two clients on four 200Gb links, with random read bandwidth tracking sequential from 32K blocks up. We also used PEAK:AIO to build the NVMe-oF target for our NVIDIA DGX Spark review. When Brian sat down with Klarzynski at SC25 for 팟 캐스트 #144, he talked about PEAK:AIO staying independent instead of becoming a feature inside a large vendor’s product line, and about a model where customers start with storage on a GPU server and add nodes to a growing namespace as their AI projects prove out. NetApp’s release centers on that same metadata and namespace work.
PEAK:AIO CEO Roger Cummings put the company’s approach plainly in his LinkedIn post announcing the deal: “We didn’t start with technology and go looking for a problem to solve. We started with the problems customers were experiencing and built around them.”
Where NetApp Takes It
NetApp launched the AFX AI portfolio last October, pairing a disaggregated storage system that has a native parallel architecture with the AI Data Engine, and in July it bought 데이터펠라고 to run GPU data processing where the data already lives, which makes PEAK:AIO its second AI-focused acquisition in about two months. NetApp hasn’t said where PEAK:AIO’s technology will integrate first, but independently scaling metadata services, a global namespace, and parallel NFS access in front of ONTAP seem like a natural fit for AFX’s disaggregated design.
As the deal closes, we’ll be watching whether NetApp keeps Lattice developing in the open; it’s a young, MIT-licensed project with LANL engineers presenting it at SDC this month. PEAK:AIO also built a following among hospitals, universities, and research labs that couldn’t justify a large array behind a single GPU server, and NetApp’s release is written around AI clouds at hyperscale, so it’ll be worth seeing how both ends of that spectrum play out. We like what PEAK:AIO has built, and NetApp’s sales reach and ONTAP installed base give that work a path into far more data centers than PEAK:AIO could on its own.
아마존