Research Finder
Find by Keyword
Will IBM’s New Storage Scale System Advancements Give Enterprises the AI-Ready Platform They’ve Been Expecting?
New flash tiers, DPU acceleration, and data-aware caching mark IBM’s next chapter in AI-optimized storage
19/11/2025
Key Highlights:
- IBM has introduced a new generation of its Storage Scale System with expanded flash capacity and DPU-accelerated performance.
- 122 TB QLC NVMe drives triple flash capacity while reducing footprint and power.
- All-Flash Expansion Enclosure (JBOF) integrates BlueField-3 DPUs, providing ~100 GB/s per 2U and > 3 PB capacity per unit.
- New bimodal Data Acceleration Tier (DAT) delivers up to 28 million IOPS, optimized for NVIDIA VGX SuperPOD environments.
- Storage Scale System 6000 is NVIDIA-certified, forming the foundation of IBM’s anticipated AI Factory reference architectures.
The News
At Supercomputing 2025 (SC25), IBM announced significant updates to its Storage Scale System portfolio, highlighted by the 7.0.0 software release that introduces multi-flash tiering across NVMe, IBM FlashCore Modules, and QLC flash drives for enhanced storage efficiency. The company also unveiled a new IBM Scale System All-Flash Expansion Enclosure (JBOF) offering controller-less, high-density expansion with QLC NVMe drives and up to four NVIDIA BlueField-3 DPUs per 2U unit. Complementing the hardware is the new Data Acceleration Tier (DAT), a high-performance flash layer engineered to optimize data access for compute resources, supporting up to 28 million IOPS in both centralized and decentralized deployment modes.
Analyst Take
IBM Storage Scale System is a scalable distributed POSIX file system that runs on open operating systems and supports any block storage platform, making it adaptable to diverse AI and data-intensive environments. IBM’s Storage Scale System 6000 update reflects a deliberate evolution from high-performance computing into enterprise-class AI infrastructure that balances scalability, efficiency, and simplicity.
IBM’s positioning of Storage Scale System as a data engine for AI factories came through clearly in the launch materials. By unifying data across edge, core, and cloud through its global namespace and AFM, the system aims to eliminate data silos and reduce GPU idle time. This is increasingly important as enterprises look to start pipelines faster and train on larger, more distributed datasets.
Since early 2025, IBM has added progressively higher-capacity QLC NVMe drives within a multi-tier architecture that also includes TLC and FlashCore Modules. This move has tripled system capacity while cutting both power and physical footprint roughly in half. Where two racks were once required, IBM now achieves the same density in one, resulting in a clear TCO reduction and encouraging broader AI and analytics deployment.
I believe the new JBOF design amplifies that progress. These units can be added to live Storage Scale System 6000 deployments without downtime, expanding capacity seamlessly inside the same global namespace under ILM policy control. The architecture’s simplicity, density, and resilience mark a decisive step toward enterprise scalability.
The updated architecture also enables larger caches and more flexible multitenancy. This matters for service providers and internal AI platform teams managing diverse workloads, as it allows fine-grained isolation, resource governance, and quota enforcement without sacrificing throughput or density.
IBM’s new DAT is perhaps the platform’s most forward-looking element. The tier supports two modes that together create synchronized reliable and performance pools. Data reads from the performance pool for low latency and writes asynchronously to the reliable pool via asymmetric replication. The result is a burst-buffer-like architecture enabling low-latency AI training and analytics I/O. Of note, IBM has also simplified deployment with a recent “one-button deploy” automation and software-defined operations that reduce setup time and management overhead.
In my opinion, these software and hardware enhancements collectively help position the Storage Scale System 6000 as a complete AI-ready platform, offering the flexibility to mix flash tiers, add JBOFs without downtime, and activate data acceleration capabilities that bridge storage and compute.
What Was Announced
The Storage Scale System 7.0.0 software release introduces multi-flash tiering across NVMe, IBM FlashCore Modules, and QLC flash drives, allowing organizations to blend performance-optimized and capacity-optimized media within a single system. Available in 30 TB, 60 TB, and 122 TB configurations, QLC NVMe drives help achieve new levels of efficiency in Storage Scale System environments.
IBM’s new All-Flash Expansion Enclosure (JBOF) extends the Storage Scale System 6000 with controller-less, high-density flash expansion. Each 2U enclosure incorporates 26 dual-ported QLC NVMe drives and up to four 400 Gb/s NVIDIA BlueField-3 DPUs, delivering around 100 GB/s of throughput and over 3 PB of raw capacity. Up to twelve JBOFs can populate a 42U rack, supporting roughly 47 PB of flash per rack and scaling beyond that through multi-rack clustering. Because the JBOFs connect to the Storage Scale System 6000 over NVMe-over-Fabric, they add capacity without the need for additional controller nodes and can be introduced without downtime. All expansion units operate within the same global file-system namespace and are governed through existing ILM policies for tiered placement and data mobility.
Complementing the hardware advances is IBM’s new DAT, a high-performance flash layer engineered to keep data as close as possible to the compute resources that use it. DAT sustains up to 28 million IOPS and offers two deployment modes: a centralized NVMe-over-Fabric configuration residing inside the Storage Scale System 6000 controllers, and a decentralized option where NVMe storage, persistent memory, or DRAM inside compute nodes forms a cooperative, non-redundant persistent cache.
IBM reports that DAT can deliver up to 340 GB/s of throughput and that wider erasure code options (16+2/3P) improve write performance and disk efficiency. Integration with NVIDIA Spectrum-X Ethernet networking may also reduce checkpoint time during foundation model training, which has become a major pain point for GPU-rich AI operations.
The IBM Storage Scale System 7.0.0 software upgrade will be available on December 9, and general availability for the All-Flash Expansion Enclosure begins December 12.
Looking Ahead
I believe IBM is positioning the Storage Scale System as the foundation for an integrated architecture that unites compute, networking, and storage for scalable AI training and inference. Both the 3500 and 6000 versions are now NVIDIA-certified, and future iterations may extend BlueField offload and automation to further enhance throughput, efficiency, and data-pipeline consistency across large GPU estates.
We will be watching for IBM’s AI Factory reference architecture and corresponding GPU-utilization benchmarks. It will also be important for IBM to quantify advantages such as checkpoint-time reductions and power-per-PB improvements relative to competitors. Clients should be asking how IBM plans to align Fusion, watsonx, and Storage Scale into a unified AI data-platform narrative, including future integration points with Red Hat OpenShift AI.
In my view, IBM’s announcements at SC25 demonstrate that IBM sees Storage Scale System as central to the company’s AI infrastructure value proposition. The next step will be aggressively extending capabilities such as the Data Acceleration Tier deeper into commercial deployments beyond the platform’s HPC origins.
Don Gentile | Analyst-in-Residence -- Storage & Data Resiliency
Don Gentile brings three decades of experience turning complex enterprise technologies into clear, differentiated narratives that drive competitive relevance and market leadership. He has helped shape iconic infrastructure platforms including IBM z16 and z17 mainframes, HPE ProLiant servers, and HPE GreenLake — guiding strategies that connect technology innovation with customer needs and fast-moving market dynamics.
His current focus spans flash storage, storage area networking, hyperconverged infrastructure (HCI), software-defined storage (SDS), hybrid cloud storage, Ceph/open source, cyber resiliency, and emerging models for integrating AI workloads across storage and compute. By applying deep knowledge of infrastructure technologies with proven skills in positioning, content strategy, and thought leadership, Don helps vendors sharpen their story, differentiate their offerings, and achieve stronger competitive standing across business, media, and technical audiences.