Enakta Storage Platform — Generally Available

NVMe at full speed. On hardware you own.

Production-grade NVMe storage built on the open-source DAOS engine. 270+ GB/s per client. Managed deployment from bare metal. Linear scaling to 1,000+ nodes. No proprietary appliances. No per-TB fees. After 3 years, the software is yours to keep.

Configure a Cluster → Talk to Us
#1 #2 #3
IO500 10-Node Production (SC25)
DAOS core holds all 3 — Enakta #3 on TCP
270+
GB/s per client
commodity NVMe
1,000+
Nodes proven
in production
3yr → yours
Perpetual licence
included
GENERALLY AVAILABLE

Why teams switch from legacy storage.

DAOS is the open-source engine behind the top IO500 results globally. The Enakta Platform is what makes it production-ready: our own SMB, NFS, and S3 gateways, managed deployment, real-time monitoring, and self-healing automation. None of these exist in the open-source project. Traditional NAS was designed for spinning disks. We bypass the kernel, eliminate metadata bottlenecks, and let NVMe hardware run at full speed.

The result is storage that runs on standard NVMe servers with standard Ethernet networking, with no custom hardware and no vendor lock-in. Your applications connect over SMB, NFS, or S3, and they work without changes.

For many organisations, the real cost of storage isn't the hardware — it's the integration. When storage is one vendor, GPU provisioning is another, and orchestration is a third, every layer boundary becomes a failure point. The Enakta Platform eliminates that boundary at the storage layer, with managed deployment that takes you from bare metal to running cluster in a single workflow — and an expanding AI layer that extends the same platform into GPU provisioning and model serving.

AppsPOSIX · PyTorch Native · Block · Object
EnaktaSMB · NFS · S3 · Management · Monitoring
DAOS CoreUser-space I/O · Distributed metadata · Self-healing
HardwareNVMe SSDs · Standard Linux · Ethernet / RDMA

Everything you need to deploy, operate, and scale.

Managed deployment, self-healing, protocol access, and linear scaling on standard hardware.

None of these exist in the open-source DAOS project. They're ours.

Managed Deployment

A dedicated HA management block handles the entire lifecycle.

  • PXE provisioning from bare metal in a single workflow
  • 3+1 spare management nodes for high availability
  • One management chassis serves multiple storage clusters
  • Built-in monitoring, logging, and alerting

Self-Healing

When hardware fails, the cluster repairs itself automatically.

  • Sub-10-minute rebuild after drive failure
  • Handles drive, node, and chassis failures
  • No manual intervention, no 18-hour repair windows
  • Data protection via erasure coding or replication

Protocol Access

Your applications connect over standard protocols without special drivers.

  • SMB for Windows and macOS workstations
  • NFSv4 for Linux environments
  • S3 for object storage workflows
  • Native PyTorch integration for AI training

AI & HPC

Storage fast enough that your GPU cluster never idles waiting on I/O.

  • Native PyTorch integration (developed for Google, open sourced)
  • Checkpoint and dataset storage at full NVMe speed
  • Parallel workloads at any scale
  • Direct support for GPU-accelerated workflows

Linear Scaling

Add nodes, get proportionally more speed and capacity.

  • Proven from 4 to over 1,000 nodes
  • Single rack to full data centre deployments
  • No performance cliff as clusters grow
  • Capacity and throughput scale together

Hardware Freedom

No vendor-specific components. Your choice of hardware, your choice of vendor.

  • Any NVMe SSDs from any manufacturer
  • Standard Ethernet or RDMA fabrics
  • Commodity servers or purpose-built clusters
  • No vendor lock-in at the hardware layer
Protocol Access

One Storage Pool. Every Protocol.

Applications connect over SMB, NFS, S3, or native PyTorch — no modifications needed. All protocols hit the same unified DAOS storage pool.

💻 Windows Editing, VFX, Office 🖥 macOS DaVinci, Premiere 🐧 Linux Servers HPC, rendering, NAS Applications Object storage, backups ⬡⬡ GPU Cluster AI training & inference ENAKTA GATEWAY SMB Windows · macOS NFSv4 Linux mounts S3 Object API PyTorch Native DataLoader POSIX DFuse mount UNIFIED DAOS STORAGE POOL User-space I/O · Distributed metadata · Self-healing
🔌
No application changes needed
Connect over the protocol you already use. Same tools, same workflows, dramatically faster storage.

Performance that speaks for itself.

Independently validated benchmarks in production at some of the world's largest deployments.

#1 #2 #3
IO500 10-Node Production (SC25)
DAOS core sweeps the top 3 globally.
Our Core42 deployment: #3 on TCP, no RDMA.
270+ GB/s
Reads and writes
from a single client server
over 8×400GbE RDMA
~320 GB/s
Aggregate reads
per 2U chassis (4 nodes)
on commodity hardware

Reference Architecture Performance

Our next-generation reference architecture pairs KIOXIA CD9P-R and Micron 7600 Pro NVMe drives with Intel Xeon 6 processors and ConnectX-7 400GbE networking. The result is up to ~320 GB/s reads per 2U chassis with proven linear scaling to thousands of nodes.

Full specifications, test methodology, and results are covered in our reference architecture blog post.

IO500 Results

At SC25, DAOS-based systems swept the top 3 positions in the IO500 10-Node Production list. Our Core42 collaboration achieved the #3 position running over TCP, proving that high performance is achievable even without RDMA networking.

See the IO500 SC25 results and read about our Core42 work in this blog post.

DAOS Foundation founding members

What a deployment looks like.

From a compact starter cluster to a full half-rack. All commodity hardware, all standard networking.

Throughput scales with both node count and network bandwidth. The half-rack uses 400GbE per node (4× the starter's 100GbE), which accounts for the throughput jump.

1U 2U 3U 5U 7U 9U TOR SWITCH 1 TOR SWITCH 2 HA MANAGEMENT NODE 1 NODE 2 NODE 3 SPARE STORAGE 1 — SYS-212GT-HNR STORAGE 2 — SYS-212GT-HNR NETWORK 2× 100GbE TOR switches Redundant fabric, dual-homed nodes MANAGEMENT 1× 2U HA chassis (3 active + 1 spare) PXE provisioning, web UI, monitoring Shared across multiple storage clusters STORAGE 2× Supermicro SYS-212GT-HNR (2U) 4 single-socket nodes per chassis 6× NVMe drives per node (KIOXIA / Micron) ConnectX-6 Dx: 2×100GbE per node Intel Xeon 6 per node 8 nodes · 48 drives · 737 TB raw 8 NODES 48 NVMe ~200 GB/s 737 TB RAW 9U RACK 1.6 Tbps
1U 2U 3U 5U 7U 9U 11U 13U 15U TOR SWITCH 1 TOR SWITCH 2 HA MANAGEMENT NODE 1 NODE 2 NODE 3 SPARE STORAGE 1 — SYS-212GT-HNR STORAGE 2 — SYS-212GT-HNR STORAGE 3 — SYS-212GT-HNR STORAGE 4 — SYS-212GT-HNR STORAGE 5 — SYS-212GT-HNR NETWORK 2× Arista 400GbE TOR switches Redundant fabric, 32 ports each Dual-homed: every node to both TORs MANAGEMENT 1× 2U HA chassis (3 active + 1 spare) PXE provisioning, web UI, monitoring Logging, alerting, cluster orchestration Shared across multiple storage clusters STORAGE 5× Supermicro SYS-212GT-HNR (2U) 4 single-socket nodes per chassis 6× NVMe drives per node (KIOXIA / Micron) ConnectX-7: 2×400GbE per node Intel Xeon 6 per node 20 nodes · 120 drives · 1.8 PB raw 20 NODES 120 NVMe ~1.6 TB/s 1.8 PB RAW 15U RACK 16 Tbps

Example configurations with 15.36 TB drives. Actual deployments scale from 2 to hundreds of chassis. Design yours.

Full component specs, test methodology, and benchmark results for this reference architecture.

Reference Architecture →

Built for teams that can't afford slow storage.

Find your workload. If it's on this list, we built the platform for you.

Media & Entertainment

High-bandwidth storage for video editing, VFX, and post-production. SMB access means your editing tools just work on Windows and macOS. Handles everything from HD dailies to 8K deliverables.

AI Training & Inference

Feed your GPUs without bottlenecks. Checkpoint, dataset, and model storage fast enough that your GPU cluster never idles waiting on I/O.

HPC & Research

The same storage engine running the world's top supercomputers, available on your hardware. Simulation output, scientific datasets, and parallel workloads at any scale.

Enterprise File Services

Replace ageing NAS with shared storage that gets faster as it grows. SMB and NFS access for everyday business use, on a platform that scales linearly.

Sovereign & Regulated

Your hardware. Your data centre. Your jurisdiction. Full data residency control for government, financial services, healthcare, and sovereign cloud deployments.

Cloud & Neocloud

White-label and OEM storage for cloud providers and neoclouds. Simple node-based licensing, perpetual or subscription. Built for CSPs and MSPs.


What open-source DAOS gives you. What the Enakta Platform adds.

The DAOS engine is open source — Apache 2.0, developed in the open, proven at exascale on the Aurora supercomputer. That's the foundation: auditable, community-driven, battle-tested. What Enakta builds on top is what makes it deployable: SMB and NFS gateways, S3 integration, managed provisioning, monitoring, automation, and engineering support. These layers are proprietary Enakta IP — commercially developed and fully supported. You get the transparency of open source where it matters most (the storage engine your data lives on) and the reliability of a commercial product where it matters most (the operations layer you depend on daily).

Feature Open-source DAOS Enakta Platform
Storage engine✓ (DAOS core)
POSIX / PyTorch
SMB (Windows/macOS)
Managed NFS gateway
S3 gateway
Automated provisioning
Monitoring & alerting
Self-healing orchestrationManualAutomatic
Validated hardware configs
Configurator tool
Engineering supportCommunityDirect

Simple, transparent pricing.

Node-based licensing with no hidden fees. You know exactly what you're paying for.

Node-Based

Licensed per storage node. No per-TB charges, no per-feature upsells, no surprises.

After 3 Years, It's Yours

Perpetual licence included. After 3 years of support, keep running your version indefinitely. Renew support for updates, patches, and new hardware enablement.

Engineer-Level Support

Direct access to the engineers who build the platform. Included with every licence.

No Hidden Fees

No per-TB, no per-feature, no capacity tiers. One price, full platform. You know what you're paying before you sign.

Talk to Us About Pricing →

Ready to size your deployment?

Use the configurator to design your cluster, or get in touch and we'll help you scope the right solution for your workloads.