NetApp plans to buy PEAK:AIO to scale ONTAP metadata separately from data

The deal targets one specific way storage fails. According to NetApp, metadata tracks information about files and where they reside. At AI scale, coordinating access to vast numbers of files can become a bottleneck even when the storage hardware can deliver data quickly ^[5]. In that case, adding drives or bandwidth leaves the bottleneck where it was.

PEAK:AIO's answer is to run metadata services separately from data services, so the metadata side can scale independently ^[4]. Its parallel NFS approach spreads that coordination work across servers and keeps standard NFS access ^[6]. For an ONTAP operator, I think the NFS compatibility matters most. In principle, existing clients could reach a new metadata tier without a new driver.

The design itself is familiar. Runtimewire noted that separating metadata and data paths and offering parallel access are established approaches in AI and HPC storage, where WEKA, VAST Data and DDN already sell systems ^[11]. PEAK:AIO's case therefore comes down to its particular implementation and its NFS compatibility, plus NetApp's route to existing customers ^[12].

NetApp says the combined system is designed for trillions of files and multi-exabyte deployments ^[7]. Those are targets. The announcement does not include customer benchmarks, a comparison with those competitors, or a timetable for bringing PEAK:AIO's technology into ONTAP ^[8]. Several things have to hold before the figure means anything for a running ONTAP estate. The metadata tier has to hold up at that file count in production. It has to keep the resilience, security and operational characteristics that NetApp says the combination is meant to preserve ^[9]. And it has to arrive as the evolution path NetApp has promised existing customers ^[1]. None of that has been independently validated ^[2].

The scale-out work is recent. PEAK:AIO's funding release described its next step as moving from single-node systems toward scaled-out AI infrastructure ^[14]. That round raised more than $6.8 million in seed money in October 2025, and Pembroke VCT led it with a reported $5 million ^[10]. That leaves a little over $1.8 million from Praetura Ventures and an unnamed Silicon Valley investor ^[1].

Klarzynski has sold storage technology to larger vendors before. He formed Computer Design Group in 1991, and Tandberg Data bought it in 2005 for its BakStor technology. Fusion-io bought two of his later companies, ID7 and SCST, in 2013 ^[15]. According to the announcement, he said the two companies share a belief that AI requires a new approach to data infrastructure ^[16]. PEAK:AIO follows DataPelago, which NetApp announced it was acquiring in July. How the pieces will be packaged and sold has not been specified ^[13].