Menu
Up to 11.8 petabytes of effective capacity in a single 2U enclosure. Here's what that means in practice — and how it stacks up against the competition.
Enterprise storage buyers face a familiar tension: the need to store and serve more data, without consuming more rack space, more power, or more budget. The IBM FlashSystem 9600 is IBM's answer to that challenge at the high end of the market — and its capacity and density figures are worth understanding in detail before any comparison is made.
This article draws only on verified published specifications from IBM and its competitors. Where figures differ between manufacturers in how they calculate effective capacity (some assume 5:1 data reduction, others 6:1), we flag that clearly so you can make a like-for-like assessment.
The FlashSystem 9600 is IBM's flagship all-flash array, available generally from March 2026 as part of IBM's latest-generation FlashSystem portfolio. It sits at the top of a three-model refresh that also introduced the FlashSystem 5600 and 7600.
The 2U form factor is key. The 9600 houses 32 NVMe drive slots in a dual-controller, active-active architecture — both node canisters serve I/O simultaneously, with automatic failover if one fails. Each node canister uses dual 48-core AMD EPYC processors and supports 768 GB or 1.5 TB of DRAM cache.
The headline capacity figure is only possible because of IBM's fifth-generation FlashCore Module (FCM5). Unlike standard industry SSDs, FCMs are designed and manufactured for IBM's FlashSystem arrays exclusively. The 9600 is the only model that supports the 105.6 TB FCM5 — the largest capacity module in the range. Populate all 32 slots with 105.6 TB modules and you have approximately 3.4 PB of raw flash in a single 2U chassis.
The effective capacity figure of 11.8 PBe is reached after IBM's inline data reduction — compression and pattern removal are handled directly on the FCM5 hardware without impacting system performance. IBM's position is that its FCMs use QLC NAND with proprietary IP that enables them to perform like TLC for write-intensive workloads — a meaningful distinction when other vendors' high-capacity drives are limited to read-intensive use cases.
Each FCM also records statistics for every I/O, enabling real-time ransomware detection at the hardware level. IBM says this can identify anomalous encryption patterns within a minute of an attack beginning.
Important note on effective capacity figures: IBM's 11.8 PBe figure is an effective capacity estimate based on data reduction — it is not raw media capacity. IBM's effective-capacity figures assume inline compression and pattern removal via FCM5. Deduplication on the 9600 currently operates in scan-only mode, meaning the system identifies duplicate data but does not yet eliminate it at the storage layer. Teams should model sizing based on confirmed behaviour for their target code level rather than assuming future deduplication gains.
A single 9600 enclosure delivers impressive numbers, but IBM's FlashSystem grid allows up to 32 systems to be federated and managed as a single virtual array. At that scale, the numbers change significantly.
FlashSystem Grid — maximum scaled capacity (32 nodes):
FlashSystem 5600: up to 77 PB | FlashSystem 7600: up to 230 PB | FlashSystem 9600: up to 377 PB
Maximum aggregate bandwidth for the 9600 grid: 2,752 GB/s
This grid capability matters for organisations planning for multi-year data growth, AI workload consolidation, or large-scale virtualisation programmes that would otherwise require multiple disparate arrays with separate management planes.
The FlashSystem 9600 competes in the high-end enterprise all-flash block storage market. The main alternatives UK enterprise buyers typically evaluate are the Dell PowerStore 9200T, Pure Storage FlashArray//XL, and HPE Alletra Storage MP B10000. Here is an honest comparison based on published, verifiable specifications.
Physical rack footprint to reach maximum effective capacity
Effective capacity figures are manufacturer-stated and use different data reduction assumptions — not directly comparable without workload-specific sizing. Pure XL170 R5 effective capacity shown at 5:1 assumption; 1.9 PB raw. *HPE Alletra B10000 U-count is indicative — scale-out architecture differs from dual-controller arrays and scales by adding nodes. Dell max capacity requires base + 3 expansion enclosures.
| Specification | IBM FlashSystem 9600 | Dell PowerStore 9200T | Pure Storage FlashArray //XL170 R5 | HPE Alletra MP B10000 |
|---|---|---|---|---|
| Form factor (base unit) | 2U | 2U base + up to 3× 2U expansion enclosures | 5U (base chassis; expands further) | 2U nodes, scale-out architecture |
| Max effective capacity (single system) | 11.8 PBe (2U) | 5.9 PBe per appliance (assumes 5:1 DR); 23.6 PBe per 4-appliance cluster | Up to ~9.5 PBe (//XL170 R5, 5:1 DR assumed; 1.9 PB raw) | Up to ~5.6 PB capacity (scale-out) |
| Drive slots (base chassis) | 32 NVMe (EDSFF / FCM5) | 25 NVMe drives per base enclosure (2.5") | Proprietary DirectFlash Modules; expands via DirectFlash Shelf | Scale-out nodes; capacity added per node |
| Max individual drive / module capacity | 105.6 TB (FCM5; exclusive to 9600) | 15.36 TB NVMe SSD | 150 TB DirectFlash Module (R5) | Not publicly specified per drive |
| IOPS (peak, single system) | 6.3M (4K read-hit) | Not publicly specified per appliance at this level | 100,000–200,000 IOPS at <1ms latency (32K block size); optimised for real-world enterprise workloads rather than 4K synthetic benchmarks | Rated at millions of IOPS; exact per-node figure not publicly confirmed |
| Controller architecture | Active-active dual controller (48-core AMD EPYC per node) | Active-active dual node (Intel Xeon, 112 cores total in 9200T) | Active-active dual controller | Disaggregated, scale-out, multi-node — no single point of failure |
| Data reduction guarantee | Data reduction handled inline by FCM5 hardware; specific guarantee varies — verify with IBM | 5:1 DRR guarantee (on reducible data types) | 5:1 average assumed; Purity Deep Reduce announced H2 2026 | Not publicly specified as a fixed guarantee |
| Ransomware detection | Hardware-level, inline via FCM5; under 1-minute detection | AI-powered Cyber Detect via software | SafeMode snapshots; software-level detection | HPE Zerto integration; software-level detection |
| Management / AI operations | FlashSystem.ai — agentic AI co-administrator; up to 90% reduction in management overhead (IBM claim) | AI-powered autonomous optimisation via PowerStore Manager | Pure1 and Fusion control plane; cloud operational model | HPE InfoSight / GreenLake cloud management |
| Latency | Under 50 microseconds (28.6% reduction vs prior gen) | Sub-millisecond (NVMe end-to-end) | <1ms at 32K block size (real-world enterprise workload optimised) | Sub-millisecond; exact figure not publicly specified |
| Typical workload fit | Mission-critical ERP, banking, AI, large-scale consolidation | Block, file & container workloads; unified; VMware-heavy environments | Tier-0 databases, SAP/HANA, Oracle, SQL Server, VMware | Mission-critical apps; cloud-native workloads; hybrid/multicloud |
Sources: IBM newsroom, IBM product documentation, Dell PowerStore 9200T product page (dell.com), Dell PowerStore Gen 2 spec sheet, Pure Storage FlashArray//XL R5 datasheet (purestorage.com), Blocks & Files (February 2026, September 2025), HPE Alletra Storage MP B10000 product page (hpe.com), Scribd B10000 overview document. Effective capacity figures are manufacturer-stated and use different data reduction assumptions — not directly comparable without vendor-specific sizing.
The most striking aspect of the 9600's specification is the combination of form factor and capacity. The 9600 delivers up to 11.8 PBe in a single 2U enclosure. The Dell PowerStore 9200T reaches 5.9 PBe per appliance, but that requires scaling up to three additional 2U expansion enclosures — making it a potential 8U footprint to match a fraction of the 9600's capacity. The Pure Storage FlashArray//XL170 R5 starts at a 5U base chassis and expands further via DirectFlash Shelves.
For UK organisations where data centre space is a tangible cost — particularly those in London, where colocation pricing is high — the ratio of usable capacity per rack unit is a real financial variable, not just a technical preference.
Pure Storage's 150 TB DirectFlash Module is the largest commercially available drive in the market and exceeds IBM's 105.6 TB FCM5 in raw per-drive terms. However, IBM's FCM5 is engineered for enterprise write-intensive workloads, whereas high-capacity QLC drives from other vendors are generally positioned for read-heavy use cases. IBM argues its proprietary QLC-with-TLC-performance approach means the capacity figure is genuinely usable for mixed or write-heavy production workloads — though organisations should verify this against their specific application profiles.
The HPE Alletra MP B10000 uses a fundamentally different architecture to the IBM, Dell and Pure products. It is a disaggregated, scale-out platform where compute and storage are separated, allowing each to scale independently. This brings architectural flexibility — particularly for organisations whose performance and capacity growth trajectories are not in step — but makes direct capacity comparisons less meaningful than with the monolithic dual-controller arrays. Published third-party documentation places the B10000's capacity at up to approximately 5.6 PB, with scale achieved by adding nodes.
All four vendors in this comparison have made AI-assisted storage management a central part of their 2025–2026 positioning. IBM's FlashSystem.ai introduces what IBM calls an agentic AI co-administrator — software that monitors, diagnoses, and auto-remediates issues, and can reduce audit and compliance documentation time by half according to IBM. IBM claims this can reduce overall storage management effort by up to 90%. HPE's InfoSight, Dell's autonomous optimisation, and Pure's Pure1/Fusion platform all offer comparable intent — but the implementation and depth of automation varies, and any specific claim of management time reduction should be tested against your environment's actual operational complexity.
IBM states that the new FlashSystem range reduces the required storage footprint by 30–75% depending on model, compared to the previous generation. For the 9600 specifically, IBM positions this as a result of both the increased FCM5 drive density and a 40% improvement in data efficiency versus the prior 9500.
For a UK mid-market organisation currently running an older all-flash or hybrid array, consolidating onto a FlashSystem 9600 may allow multiple enclosures to be replaced by a single 2U system — with implications for power, cooling, rack space, and management complexity. The economics of that consolidation depend heavily on your current installed base, data reduction ratios on your actual data types, and support contract costs.
A note on comparing vendor data reduction claims: Effective capacity figures are not directly comparable across vendors because manufacturers use different assumptions. IBM's 11.8 PBe figure uses IBM's own data reduction model. Dell's 5.9 PBe assumes a 5:1 data reduction ratio and "double drive tolerance." Pure's figures assume 5:1 average data reduction, excluding thin provisioning. Always request a workload-specific sizing analysis from your vendor or reseller before making procurement decisions based on effective capacity alone.
Based on its published specification, the FlashSystem 9600 is best suited to organisations that:
It is less likely to be the first choice for organisations primarily running file or object workloads (block-only architecture), those whose workloads are better served by a unified block-file-container platform like Dell PowerStore, or those whose growth projections do not justify the investment in flagship-tier hardware.
Fortuna Data is an IBM Silver Partner with direct access to IBM's storage portfolio, technical pre-sales support, and specialist knowledge built over decades in UK enterprise IT. We work with mid-market and enterprise organisations across sectors including finance, media, construction and professional services — helping them make the right storage decision for their specific workload, budget and growth plans.
If you are evaluating the IBM FlashSystem 9600 — or comparing it against alternatives from Dell, Pure Storage or HPE — we can provide a no-obligation workload assessment, a like-for-like capacity comparison using your actual data types, and a total cost of ownership model that reflects your real operational environment rather than vendor marketing assumptions.
Sources and accuracy: All specifications cited in this article are drawn from IBM's official newsroom announcements (February 2026), IBM product documentation, Dell PowerStore product pages and specification sheets, Pure Storage FlashArray//XL datasheet and Blocks & Files reporting (September 2025), and HPE Alletra Storage MP B10000 product documentation. Effective capacity figures are manufacturer-stated and use different data reduction assumptions — they are not directly comparable without workload-specific sizing. Specifications are correct at time of publication and subject to change. Last reviewed June 2026.
