VDURA has made its VDURA Data Platform V12 generally available, pitching a 60 percent-plus reduction in total cost of ownership versus competitive architectures at the same GPU feed rate, alongside claims of more than 2x performance per watt. The company says a single VDP system at roughly 20 PB usable capacity spans a 35x performance range, from a capacity-optimized mixed fleet at 2 percent flash drawing 18 kW to an all-flash configuration delivering 1,000 MB/s per TB.

V12, first announced in November 2025, targets multi-tenant GPU cloud operators and AI factories. It is available now for all V5000 class systems and as an upgrade path for V11 customers, and ships as a qualified solution on Supermicro Building Block Solutions, scaling from 8 to 100,000 GPUs on one software stack.

The release marks the continuation of a strategy that began in 2024, when high-performance parallel filesystem supplier Panasas rebranded as VDURA under new CEO Ken Claffey and pivoted from hardware-led sales toward software-defined, enterprise AI storage. The V11 release rebuilt the PanFS code base as the VDURA Data Platform, introducing the VeLO (Velocity Layered Operations) metadata accelerator and the VPOD (Virtualized Protected Object Device) construct, deployed as discrete microservices across thousands of nodes.

Claffey framed V12 as a direct response to neocloud operator requirements. "Neocloud and AI factory operators told us exactly what they need from storage: keep the GPUs fed, isolate the tenants, automate everything, expand capacity for cold data without a second system, and move that data back to flash the moment it warms up for extended context," he said. "V12 is that list, shipped. It is the same mixed-fleet, software-defined model the hyperscalers run inside their own clouds, delivered on Supermicro systems our customers already buy. Every watt and every rack unit we give back is another GPU the operator can put into service."

The HYDRA (High-performance, Yield-optimized, Distributed, Resilient Architecture) design underpins the release. Tenancy controls deliver per-tenant quality of service, namespaces, encryption keys and VLAN isolation on a shared fleet, letting a provider carve one storage pool into hard-walled tenant services with capacity and performance guarantees. Provisioning runs through REST APIs, Kubernetes CSI and infrastructure-as-code pipelines, so storage deploys and bills through the same tooling as the rest of the GPU cloud.

Context-Aware Tiering moves data between media as access patterns shift between training and inference. VDURA says roughly 90 percent of files stay on flash while roughly 90 percent of capacity settles on HDD, with no stub files, rehydration steps or manual tuning. A persistent KV cache that outlives the inference pod lets sessions resume rather than re-prefill, which the company says delivers faster first tokens and more concurrent users per GPU at flash cost rather than recompute cost.

On the data path, RDMA transfers move data directly between GPU and storage with the CPU out of the loop. VDURA says its DirectFlow parallel client takes roughly 191 MB of DRAM and zero cores from the GPU node. The Elastic Metadata Engine, accelerated by VeLO, reaches up to 20x improvement, 225,000 creates and deletes per second per Director, and billions of metadata operations per second in aggregate. File and S3 access share one platform, with an S3 object appearing as a file in the volume rather than a copy, eliminating staging between ingest, training, inference and archive.

The platform also adds instantaneous space-efficient snapshots for checkpointing, SMR drive support that unlocks 25 to 30 percent more capacity per rack, AES-256 encryption at rest and in flight with per-tenant KMIP key management, and self-healing resiliency with failure domains as small as a single VPOD. The VDURA Sentinel support service opens a service request with diagnosis attached before the customer sees a fault.

The hardware configuration is concrete: the V12 Supermicro reference uses the AS-1116CS-TN, a 1U single-socket AMD EPYC 9005 series system with 12 NVMe bays, serving as both VeLO Director node and all-flash F-Node. The AS-2015HS-TNR hybrid storage node and the CSE-947HE2C 4U 90-bay JBOD make up the mixed-fleet data plane. Every node connects via RDMA directly to GPU nodes, with no dedicated back-end storage fabric.

Clusters grow online from three nodes to thousands, and the flash-to-HDD ratio is a continuous setting operators adjust by adding all-flash nodes, hybrid nodes or HDD shelves independently — one control plane and one data plane spanning both media, rather than forking into separate all-flash and capacity systems.

VDURA positions itself against DDN, VAST Data and WEKA in the AI-era data store category, with NetApp also now in that group. The company will demonstrate V12 at the Ai Everything event at ADNEC Centre in Abu Dhabi, 6–7 October, Booth H3-D45. If the TCO and performance-per-watt claims hold in operator deployments, VDURA's bet that neoclouds will trade dedicated storage silos for one tunable mixed fleet becomes a credible challenge to the incumbents.