HPE Alletra X10000 R4: 16 Nodes, NFS+Object in 2026

Release 4 lands: what's actually shipping now

HPE Alletra Storage MP X10000 Release 4 has shipped, making good on the scale-out roadmap HPE first showed off at its GreenLake update back in May. A single cluster can now stretch to 16 nodes and 16 JBOFs — doubling the single-cluster scaling for performance and capacity compared with the previous release — with a single system topping out at roughly 23PB of raw capacity.

Since the HPE Alletra Storage MP platform keeps compute nodes and JBOF capacity separate, teams can grow either independently — adding nodes to handle more concurrent workloads, or shelves to expand the data lake — reducing the need for data migration. HPE argues that AI training, inference, analytics and sprawling unstructured repositories can now converge onto a single platform at this scale, instead of being scattered across dedicated silos.

Native NFS beside object: the consolidation argument

The standout architectural shift here is native-namespace NFS operating side by side with the platform's existing object storage in the same release. According to HPE, this works with no translation layer or second storage silo. Because most enterprise environments still depend on file-based pipelines for daily operations, Release 4 means those workloads no longer require a standalone filer sitting next to the object store.

For buyers assembling solutions for file services and unstructured data, this is the detail that reframes the RFP: a vendor list that previously needed a file specialist and an object specialist can now include a single platform that claims both as first-class protocols on a shared data foundation.

RDMA for file, GPUDirect Storage and the AI performance case

With Release 4, HPE broadens the platform's RDMA acceleration to cover file in addition to object, enables NVIDIA GPUDirect Storage, and says the update positions the platform to pursue NVIDIA AI Enterprise certifications. The X10000 had previously become the first object storage platform to secure NVIDIA-Certified Storage validation.

HPE points to third-party benchmarking of the RDMA-accelerated X10000 handling KV cache offload, which recorded time-to-first-token gains of up to 20x and effective inference throughput improvements of up to 17x. The in-array Data Intelligence layer also picks up NVIDIA NIM-based multimodal AI features, the ability to run across multiple nodes, and support for AI functions that customers build themselves, letting a customer's own enrichment or classification logic execute directly against the stored data. Buyers weighing AI, analytics, and large unstructured repositories against GPU-heavy roadmaps will want these figures independently validated against their own model and pipeline mix before treating them as guaranteed outcomes.

Security, sovereignty and day-2 operations

Release 4 introduces Key Management Interoperability Protocol (KMIP) support for external key management, upgrades to TLS 1.3, and further compliance and security refinements HPE has not fully itemized, alongside broader support for disconnected and air-gapped environments. HPE frames these additions as the credentials needed to compete for government and regulated-industry contracts, where storing encryption keys outside the array itself is frequently a hard requirement.

When combined with HPE Data Fabric Software, the X10000 functions as the system of record for stored data, while the fabric layer supplies a federated global namespace, lineage tracking, geofencing and sovereignty controls spanning edge, core and cloud — a combination HPE markets as its Unified AI Data Foundation. On the day-2 side, the release brings orchestrated software updates, automated firmware handling, and improved supportability.

• •KMIP external key management and TLS 1.3 enhancements
• •Expanded disconnected/air-gapped deployment support
• •Federated namespace, lineage and geofencing via HPE Data Fabric Software

Procurement mechanics: CapEx, Flex and longer SaaS terms

The system retains the 100% data availability guarantee HPE rolled out in May for file and object workloads, available under both CapEx and Flex purchasing models. HPE has also introduced SaaS subscription lengths of one, six and seven years on top of its existing terms.

Sizing runs from a minimum of three nodes up to the new 16-node ceiling, which supports staged deployment — start small, scale the cluster as demand grows rather than over-provisioning upfront. That flexibility matters when comparing IT finance options for storage across a multi-year contract, since the availability guarantee and term length both affect total cost of ownership calculations.

What this means for UK storage RFPs

For UK enterprises currently evaluating scale-out arrays, Release 4 shifts the calculus in two ways. First, the doubled ceiling means fewer clusters are needed to hit a given capacity or concurrency target, which simplifies licensing, support and data governance boundaries. Second, native NFS parity with object means a single platform can now credibly replace what used to be a two-vendor file-plus-object bid.

That doesn't remove the need for due diligence: buyers should validate the AI performance claims against their own workloads, confirm KMIP integration with existing key management infrastructure, and model the SaaS versus CapEx trade-off across the full contract term before shortlisting. Teams starting from scratch with our storage solution finder can use the 16-node ceiling and 23PB figure as a sizing anchor when comparing bids.