NetApp has announced its intent to acquire PEAK:AIO, a developer specialising in high-performance parallel file systems. The acquisition aims to solve metadata bottlenecks in AI infrastructure, enabling storage to scale to trillions of files. By integrating PEAK:AIO’s technology into its ONTAP architecture, NetApp seeks to reduce GPU idle time and support multi-exabyte deployments for AI at hyperscale.
The deal targets a specific weakness in how AI infrastructure is built. As cloud providers and enterprises expand GPU clusters, the storage systems behind them have to answer enormous numbers of simultaneous requests, and much of that load falls on metadata, the index recording where each file sits, who can access it and when it last changed. Conventional storage architectures keep metadata and data together, so when thousands of processors request files at once, the index itself can turn into a queue and leave costly GPUs idle while they wait for data.
NetApp said it was building an architecture that separates metadata from data and allows metadata services to scale independently. The company said the design was intended to support trillions of files and multi-exabyte deployments. Those figures are design targets, and NetApp has published no benchmarks or delivery timeline to show how the combined system performs in practice.
The concurrency problem started in national laboratories
PEAK:AIO’s technology originated in collaborations with Los Alamos National Laboratory and Carnegie Mellon University, where scientific simulation and research workloads ran into the limits of shared storage long before commercial AI brought the same pressures to enterprise data centres. NetApp said the technology had been designed for some of the most demanding data-intensive computing problems in the industry.
Klarzynski said the two companies had found common ground in their view of what AI requires from infrastructure. “NetApp and PEAK:AIO connected through the shared belief that the AI era demands a fundamentally new approach to data infrastructure,” he said. “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.”
For PEAK:AIO, the acquisition offers distribution that a specialist vendor would struggle to build alone. For NetApp, it brings in a capability the company has chosen to buy and attach to its existing platform, where a longer internal development cycle might have left it behind the pace of GPU deployment.
ONTAP remains the foundation NetApp is asking customers to trust
NetApp plans to integrate PEAK:AIO’s metadata services and parallel namespace technology into its ONTAP-based data architecture. The company said this would give existing ONTAP customers a clear evolution path, although it has not explained what that migration will involve or whether it will require new hardware or licensing.
Kurian placed the emphasis on keeping ONTAP’s established strengths intact while adding scale. “AI clouds need high-performance shared storage that can scale alongside growing GPU clusters while maintaining the resilience, security, and operational simplicity organisations depend on,” he said. “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 maximise infrastructure efficiency and support AI at hyperscale cloud.”
The combined architecture will rely on three elements: scalable metadata services, a global namespace that presents data across many storage nodes as a single file system, and access through parallel NFS. Parallel NFS is an open extension of the Network File System standard that lets clients read from multiple storage nodes at once, which spares customers from installing a proprietary client on every server that needs the data.
NetApp’s promised gains still lack a baseline
NetApp said the architecture would help reduce data-related GPU stalls, improve infrastructure efficiency and accelerate AI innovation. The company did not say by how much, against which existing systems, or under which workloads, which leaves buyers without a way to compare the claim against what they run today.
The company’s own statement of product direction adds a further note of caution. NetApp said it made no commitment to develop or deliver any products, integrations or features arising from the acquisition, and that the timing of any release remained at its sole discretion.
Customers weighing the deal will be watching for three things once it closes: how quickly PEAK:AIO’s metadata layer ships inside ONTAP, what the upgrade path costs existing users, and whether independent testing confirms the reduction in GPU idle time on which NetApp has built the case for the acquisition.


