Six industry participants recently launched the Open Flash Platform (OFP) initiative, backed by six founding members: Hammerspace, the Linux Foundation, Los Alamos National Laboratory (LANL), ScaleFlux, SK hynix, and Xsight Labs.

The initiative purports to address the challenges presented by the next wave of AI-driven data storage, specifically the limitations of power, heat, and data center space.

OFP advocates for a ground-up redesign of flash infrastructure, shifting away from the traditional storage server model to a network-attached, open standards-based approach that directly exposes flash to the storage fabric.

Technical Overview

The OFP group positions flash not just as a performance tier, as NVMe did a decade ago, but as a capacity tier by eliminating intermediary storage servers and proprietary software stacks.

It builds on open technologies like parallel NFS (pNFS) and standard Linux, pushing for a fully disaggregated model that bypasses the need for traditional array controllers.

The platform’s architecture consists of several core components:

• OFP Cartridges: Each cartridge contains NAND flash (primarily QLC for density), a DPU or IPU for data processing, and a direct network connection. These cartridges act as self-contained storage units optimized for both density and low power consumption.
• OFP Trays: Trays aggregate multiple cartridges, manage power distribution, and provide a physical form factor compatible with modern rack designs. The goal is to maximize rack-level storage density while reducing power and cooling demands.
• IPUs/DPUs: OFP leverages the computational capabilities of DPUs to handle storage services, replacing traditional x86 servers that add cost, complexity, and additional power draw. This move aligns with broader trends in computational storage and data-centric architectures.
• Linux and Stock NFS: Each cartridge runs standard Linux and NFS, enabling direct, standards-based network access without requiring proprietary control layers or custom filesystems.
• Open NAND Sourcing: OFP promotes flexibility in sourcing NAND flash from multiple manufacturers, aiming to reduce single-vendor lock-in and foster supply diversity.

Performance and Efficiency Gains

The OFP group claims that its architecture can deliver significant improvements over traditional all-flash arrays. These include:

• 10x higher storage density, with up to 1 exabyte of flash per rack.
• 60% longer flash lifespan, aligning with the natural longevity of flash compared to x86 processor cycles.
• 90% reduction in power consumption, addressing energy constraints in modern data centers.
• 50% lower total cost of ownership (TCO) by removing licensing costs, proprietary stacks, and excess server hardware.

The group highlights the inefficiencies of current AFA designs, which are tied to server-centric models that often expire on CPU lifecycles, resulting in costly and disruptive upgrades. OFP’s modular, open approach aims to decouple storage infrastructure from these constraints.

Industry Fit

While OFP presents a compelling vision, its market positioning is uncertain. Most established AFA vendors, including Pure Storage, Dell, and VAST Data, already leverage open standards and support capabilities such as RDMA and GPUDirect for AI workloads.

In contrast, OFP’s membership currently lacks broad industry representation. Only one major NAND vendor (SK hynix) and one DPU supplier (Xsight Labs) are participating. The absence of other flash, DPU/IPU, and software-defined storage leaders raises questions about the initiative’s ability to drive a de facto industry standard.

Additionally, OFP’s technical claims, such as the 10x density increase and 90% power reductions, have not yet been validated with public benchmarks or detailed specifications.

Analyst’s Take

OFP’s open, modular approach aligns with a broader industry trend toward data-centric architectures for AI, where storage must scale in both performance and capacity while minimizing power consumption.

If successful, OFP could pressure AFA vendors to further reduce proprietary layers and adopt more modular, networked designs. However, the initiative faces several challenges:

• Limited Ecosystem Support: Without broader participation from NAND, DPU, and storage software vendors, OFP risks being viewed as a niche or vendor-driven effort rather than a community standard.
• Overlap with Existing Standards: OFP’s vision overlaps with initiatives like the Open Compute Project (OCP), which already addresses disaggregated storage and networking at scale.
• Mature Competitors: AFA vendors have years of experience in enterprise software, proven reliability, and advanced features that OFP must match or exceed to gain traction.

In its current form, OFP appears more like a proof-of-concept or advocacy group than an industry-shaping standard. The concept of directly networked flash cartridges with DPUs is technically attractive, especially for AI and HPC environments; however, adoption will depend on the availability of tangible products, broad ecosystem buy-in, and demonstrated performance gains.

For now, AFA vendors face minimal competitive threat, though OFP’s ideas could influence the direction of future disaggregated flash architectures.