One DDN storage platform for the AI cluster you are scaling.
How a Bengaluru GPU cloud operator stopped bottlenecking H100 nodes on shared NFS once EXAScaler took over the training data path. Story below.
DDN AI & HPC Storage at a glance
Why DDN anchors most Indian AI and HPC storage estates.
- OEM
- DataDirect Networks, Inc., founded in 1998 in the US. Founder-led by Alex Bouzari and Paul Bloch. India operations run through DataDirect Networks India Private Limited. Took its first outside investment in January 2025, USD 300 million from Blackstone Tactical Opportunities at a USD 5 billion valuation.
- Platform
- EXAScaler, a Lustre-based parallel file system built for large GPU clusters, packaged inside AI400X2 and AI400X3 appliances. Infinia 2.x is DDN’s newer software-defined AI data platform for mixed pipelines, not just pure HPC scratch.
- Track record
- DDN says its systems power more than 500,000 NVIDIA GPUs worldwide, including xAI’s Colossus cluster of 100,000 H100s, and that 85 of the Fortune 500 run AI or HPC workloads on DDN. It is the world’s largest privately held data storage company, with 11,000-plus customers.
- India footprint
- Around 500 R&D staff based in Pune and Bengaluru. DDN has manufactured storage in India under Make in India since 2021, the first storage vendor to do so. Co-founder Paul Bloch has stated that 90 percent of India’s AI supercomputers, including PARAM Siddhi and Yotta’s Shakti Cloud, run on DDN.
- Indicative price
- DDN has no published India price list. Entry AI400X2 configurations typically start above roughly Rs 25 lakh, and the final number depends on GPU count, throughput target and redundancy level.
- Sirius Star’s role
- We are an authorized reseller, not DDN itself. We size the array against your GPU roadmap, coordinate the rack-and-stack, and stay the one number you call once support tickets start.
- Best fit
- AI training clusters, HPC research labs and GPU cloud operators who cannot afford storage throttling back a GPU fleet they already paid for.
The DDN AI & HPC Storage ranges Sirius Star supplies
Pick the range that matches the use case. We size the mix in the free 30-minute review.
AI400X2 / AI400X3 appliances
Packaged parallel storage built to plug straight into your GPU fabric and keep up with training jobs, not fall behind them.
- Validated for NVIDIA DGX SuperPOD and GB200 reference architectures
- Scales in blocks, add capacity or performance without a forklift upgrade
- Sizing and per-node throughput confirmed on the review call, no published India list price
EXAScaler parallel file system
The Lustre-based file system underneath the appliances, proven at some of the largest AI clusters in the world.
- Runs extreme-scale deployments including xAI’s 100,000-GPU Colossus cluster
- Node-by-node upgrades need planning, we flag this honestly below
- Not multi-protocol by default, NFS/SMB access needs a gateway layer
Infinia 2.x
DDN’s newer software-defined data platform, built for mixed AI pipelines rather than pure HPC scratch space.
- Object, file and Kubernetes-native access from one platform
- Positioned for inference and RAG pipelines, not only training runs
- Newer than EXAScaler, so its India reference base is smaller so far
A3I for NVIDIA DGX
Pre-validated storage-and-compute pairing for teams standardising on DGX or HGX platforms.
- Removes the guesswork of matching storage throughput to GPU count
- Fits teams buying compute and storage on the same PO
- We run the throughput math before you commit rack space
DDN AI & HPC Storage vs VAST, WEKA, Dell
All four are honest choices. Most Indian buyers land on the first option for service depth and ecosystem fit.
| Brand | Where it wins | Best fit |
|---|---|---|
| DDN (via Sirius Star) | The deepest NVIDIA reference-architecture alignment and the largest proven scale, 500,000-plus GPUs including xAI’s Colossus. | AI training clusters and HPC labs where storage throughput cannot bottleneck an expensive GPU fleet. |
| VAST Data | A genuinely unified, multi-protocol platform out of the box, no separate gateway needed for NFS or SMB. | Teams that want one simple namespace across file, object and block without managing Lustre. |
| WEKA | Software-only flexibility, runs on hardware or cloud of your choice. | Cloud-first or hybrid AI teams that do not want to commit to a fixed on-prem appliance. |
| Dell PowerScale | One vendor relationship if your compute and servers are already Dell, plus a wider India service network. | Mixed workloads where extreme HPC scale is not the primary requirement. |
How a Sirius Star DDN procurement runs
Free 30 minute review first. Then a written quote in 24 working hours.
GPU and workload review
Free, 30 minutes. We map your GPU count, training-versus-inference mix and target throughput.
Array sizing
AI400X2, AI400X3, EXAScaler or Infinia sized against your actual node count, not a generic tier.
Rack, stack and validate
Physical install, cabling to your GPU fabric, and a validated A3I reference build before go-live.
Support and renewal
We stay on as your local contact for firmware, tickets and renewals. Written quote lands in 24 working hours of the review.
Buying DDN storage for an AI cluster in India, a field guide
The deeper read for infrastructure leads sizing storage against a GPU roadmap, not just picking an appliance.
- AI400X2 vs AI400X3 vs Infinia, which one actually fits your pipeline
- How to size throughput against GPU count before you commit rack space
- The node-by-node upgrade planning EXAScaler buyers get caught out by
- 5 questions to ask before you sign, from real buyer complaints
DDN AI & HPC Storage India FAQ
Common questions about this brand for Indian buyers. Hover any underlined term for a plain-English definition.
What does DDN storage actually do differently from a normal enterprise SAN?
DDN builds storage specifically for AI and HPC throughput, not general enterprise workloads. Its EXAScaler and Infinia platforms are built to keep pace with dozens or hundreds of GPUs pulling data at once, something a standard SAN was never designed for. DDN says its systems power more than 500,000 NVIDIA GPUs worldwide, including xAI’s 100,000-H100 Colossus cluster.
Is DDN actually used in India, or is this a US-only platform?
DDN runs roughly 500 R&D staff out of Pune and Bengaluru, has manufactured storage in India since 2021 under Make in India, the first storage vendor to do so, and co-founder Paul Bloch has said 90 percent of India’s AI supercomputers, including PARAM Siddhi and Yotta’s Shakti Cloud, run on DDN. India is one of its largest markets after the US and Japan.
EXAScaler sounds complex to manage. Is that a fair concern?
Yes, and we will not pretend otherwise. Lustre-based systems like EXAScaler need node-by-node planning for upgrades, and some Gartner Peer Insights and PeerSpot reviewers flag that the management tooling lags what buyers expect at this price point. We plan upgrade windows with you upfront so this does not become a surprise mid-project.
How does DDN compare to VAST Data or WEKA for a new AI cluster?
VAST Data ships a genuinely unified multi-protocol platform out of the box, simpler if you do not want to manage a separate NFS or SMB gateway. WEKA is software-only and runs on hardware or cloud of your choice, useful for hybrid teams. DDN’s edge is scale and NVIDIA reference-architecture depth. Most Indian buyers land on DDN once GPU count and training-job size get large enough that the other two need to prove they can keep up.
What does a DDN AI400X2 or X3 deployment cost in India?
DDN does not publish an India price list. Entry AI400X2 configurations typically start above roughly Rs 25 lakh, depending on capacity, throughput target and redundancy level. Sirius Star sizes the exact configuration against your GPU count and confirms pricing before you sign anything.
We are only running a handful of GPUs. Is DDN overkill for us?
Possibly. DDN’s appliances are built for clusters that actually stress storage, typically eight or more GPUs training continuously. If you are running a small inference workload or a handful of GPUs without sustained throughput pressure, we will say so on the review call rather than size you into hardware you do not need yet.
Ready for a written DDN storage quote?
Tell us your GPU count, training-versus-inference mix and target throughput. A sized quote lands in 24 working hours.
More topics
Side-by-side comparisons Sirius Star buyers look at next.
Pair this on one PO
What buyers typically add to a Sirius Star order. Each link is a live page on the Sirius Star site.
Related reading from the Sirius Star blog
Long-form context from our team. Each link is a live post on siriusstar.in.