NetApp Announces Intent to Acquire PEAK:AIO to Advance Scalable AI Infrastructure Architecture

Comments

This news is paramount for NetApp and clearly reshuffles the cards in the high-performance file storage market. For several years, we repeatedly mentioned to George Kurian that NetApp was surprisingly absent from this segment. For a company that has championed NFS for decades, this was difficult to understand.

For years, NetApp product data sheets mentioned pNFS support, but the company never built or promoted a significant offering around it. Support existed, but essentially as a feature rather than the foundation of a dedicated architecture. This created both frustration and disappointment. As long-time market observers, we were convinced that if NetApp decided to seriously address high-performance file storage, NFS and its pNFS extension was the natural direction to follow.

At Insight 2025, NetApp finally made a visible move with AFX, which we consider phase one of its pNFS journey. However, AFX remained targeted at specific use cases and workloads rather than representing a broad answer to the high-performance file storage opportunity.

Until now, NetApp has relied on parallel file systems from partners such as ThinkParQ or IBM Storage Scale, typically coupled with E-Series systems. To be fair, other major vendors have faced similar challenges. Everpure and Dell EMC also lacked comprehensive internally developed solutions for high-performance file storage and relied on third-party technologies. They subsequently reacted with pNFS-based approaches, with FlashBlade EXA at Everpure and pNFS extensions for PowerScale at Dell. Dell also announced its Lightning file system, but little has emerged since then. The most recent Dell Technologies World included only one session dedicated to Lightning, reinforcing our view that the product is not yet ready for broad deployment. We don't mention HPE here, as the company took a different path with its timely acquisition of Cray, which gave it immediate access to established high-performance computing and storage technologies. 

Back to NetApp. The absence of a stronger answer was particularly surprising given the company's deep pool of NFS expertise. But AI has dramatically changed the landscape. The opportunity has become too large to ignore, and remaining outside this market could rapidly translate into lost business. NetApp, Everpure and Dell ultimately had little choice but to react.

What is particularly positive here is NetApp's recognition that time to market is crucial. AI is pushing infrastructure vendors toward new architectural boundaries, but, more importantly, it is forcing them to move quickly. NetApp could have decided to develop a new technology internally, but that would have taken considerable time. A much faster approach was to analyze the market, identify promising technologies, conduct aggressive stress and performance testing, and finally acquire the right company. And if the company is small, it is even better as it will be more easy to digest and integrate. It confirms once again that small companies are more agile to innovate than large ones.

And this is exactly what happened with Peak, a company recognized for both the quality of its technology and the expertise of the team behind it.

Regarding NFS and its pNFS extension, the specifications are public, although implementations can obviously differ significantly from one vendor to another.

As a reminder, pNFS first appeared as part of NFSv4.1, standardized in 2010, 16 years ago, and remains part of NFSv4.2. The history goes much further back. NFSv2 was documented in RFC 1094 in March 1989, NFSv3 in RFC 1813 in June 1995, while NFSv4 was later documented in RFC 7530. NFSv4.1, including pNFS, was defined by RFC 5661 in January 2010, followed by NFSv4.2 with RFC 7862 and subsequent updates. Other relevant specifications include RFC 5667 for RPC-over-RDMA and RFC 8166 for Remote Direct Memory Access Transport for RPC. Several RFCs cover pNFS layouts, with RFC 8435 defining the Flexible File Layout particularly relevant here.

Technically, NFS itself is an open protocol rather than an open-source product, while Peak has placed its implementation, Lattice, in the open-source domain. And again, pNFS like NFS is an access method that helps to decouple layers.

By design, pNFS separates metadata management from data access, while different transport mechanisms can be used between clients and data servers. One particularly simple yet powerful model is to retain NFS between clients and data servers, with access orchestrated by metadata servers, referred to here as pNFS directors.

This disaggregation enables highly scalable architectures because clients, directors and data servers can scale independently according to capacity and performance requirements. It also naturally supports very large single namespaces, or multiple namespaces segmented by tenants or workloads, while delivering very high aggregate performance.

The Peak approach also supports very large files, very large file systems and extremely high unified throughput. Another key advantage is its standards-based model. Since Linux distributions already include NFS clients, the architecture is non-intrusive and does not require proprietary client software to be installed across potentially thousands of compute nodes. Modern implementations must, of course, also take advantage of flash, SSD, NVMe, RDMA and increasingly CXL-related architectures to address emerging HPC and AI requirements.

NetApp is now equipped with this technology. During discussions yesterday at the company's annual conference with CEO George Kurian and CPO Syam Nair, they confirmed that NetApp intends to position it for HPC and AI workloads. And this is clearly a new era for NetApp.

This announcement will obvisouly have an impact on the competitive landscape and brings NetApp into the relatively small group of vendors addressing high-performance file storage with dedicated technology. It will be particularly interesting to see how other pNFS implementers react, including Dell, Everpure, Hammerspace, Red Hat and even Infortrend, which also selected pNFS for its GX series.

For readers interested in the history of this architecture, it is also worth revisiting Tonian Systems, which developed a pNFS director in the early 2010s. The company was acquired by Primary Data in 2013, whose technology and team later contributed to what became Hammerspace in early 2018. Hammerspace represents a software alternative to this joint NetApp - Peak solution.

NetApp's move should generate significant attention from partners and end users looking for scalable, standards-based high-performance file storage architectures, particularly as AI infrastructure increasingly demands massive parallel data access. Looking at partners at the event we anticipate some announcements with some of them to confirm this open standards direction.

We should learn more within hours. George Kurian's keynote will feature Gary Grider from Los Alamos National Laboratory on stage, and this announcement appears positioned as one of the major pieces of news of the week. We therefore expect the NetApp team here in Las Vegas to provide additional technical and product details about this direction.

This strategic move illustrates perfectly how modern infrastructure challenges increasingly require modern architectures. Peak is a particularly relevant example, giving NetApp both the technology and, perhaps even more importantly, the time-to-market advantage it needed to finally become a serious participant in high-performance file storage.