IBM hardware · Storage for AI and analytics
Data at 340 GB/s for your AI and HPC.* Your price in one business day.
For the teams who keep AI, HPC and analytics fed with data: IBM Storage Scale System, sized to your workloads by our engineering team and priced before you commit.
One of our engineers replies within one business day.
Your next storage can feed the GPUs.
- It sends data straight to GPU memory with NVIDIA GPUDirect Storage.
- It puts file and object data from many sites in one namespace.
- It keeps immutable Safeguarded Copy snapshots to recover from after a cyberattack.
The storage you run today may be serving you well. This is what Storage Scale System adds.
We size it with you. You see your price first, within one business day.
One data platform, file and object.
IBM Storage Scale System is IBM Storage Scale software on NVMe flash, as one system. IBM says active file management brings data from other storage, public cloud included, into a single global namespace, read over POSIX, NFS, SMB, S3 and more.
A straight path to the GPU.
With NVIDIA GPUDirect Storage, data moves between NVMe storage and GPU memory without passing through the host CPU and DRAM, as IBM describes it.
Recover from a copy nothing can change.
Safeguarded Copy takes isolated, immutable point-in-time snapshots on a schedule, and Storage Scale RAID erasure coding mitigates hardware failures, as IBM describes it.
Source: IBM’s Storage Scale System page, opened October 23, 2026.
Three of IBM’s own figures.
340 GB/s read, 155 GB/s write.
IBM’s throughput figures for one Storage Scale System 6000. IBM’s data sheet states the 340 GB/s for NVMe over Fabrics configurations.* IBM product page, read October 23, 2026.
4.7 PB raw in one 4U node.
IBM’s maximum raw capacity for the Storage Scale System 6000, with 12 × 30.72 TB TLC and 36 × 122.88 TB QLC NVMe drives. IBM product page, read October 23, 2026.
More than 42 PB in one rack.
IBM’s figure for a Storage Scale System 6000 with All-Flash Expansion Enclosures and 128 TB QLC NVMe drives, in a single 42U rack.
* IBM’s condition, from its Storage Scale System 6000 data sheet (December 2025): performance metrics were obtained by running sample programs in a controlled environment with standard hardware and software configurations and testing procedures. Since performance varies with configuration, program characteristics, and other installation and environment factors, IBM says results obtained in other operating environments may vary.
Source: IBM’s Storage Scale System page, opened October 23, 2026.
Pick the size that fits.
Two systems. The Storage Scale System 3500 is 2U, up to 737.28 TB raw and 126 GB/s, IBM’s entry and mid-level system. The Storage Scale System 6000 is 4U, up to 4.7 PB raw and 340 GB/s, and takes up to three All-Flash Expansion Enclosures of more than 3 PB raw in 2U each.
Source: IBM’s Storage Scale System page, opened October 23, 2026.
What you get.
If you run infrastructure.
You get a system sized to your capacity and throughput, from one 2U Storage Scale System 3500 to a 42U rack of flash. Erasure coding and immutable snapshots built in. Our engineering team stays with you through cutover.
If you run data and AI.
You get file and object data in one namespace, read over S3, NFS, SMB or POSIX, and a straight path to GPU memory. IBM measured up to 340 GB/s read.* Our engineering team sizes it to your pipeline.
If you are the CIO.
You get one storage platform for AI, HPC and analytics, from 2U to a full rack, and the number before you commit. Pay only for what you need. See your price first, within one business day.
You have real choices. Here is each one, plainly.
Keep the storage you have.
Nothing to buy. Nothing to move. What Storage Scale System adds will still be here when you are ready.
Buy from IBM itself.
IBM sells directly to fewer than about 100 companies and organizations in the United States. Everyone else buys through an IBM Business Partner. So the real choice is which partner.
Buy through another IBM Business Partner.
Same IBM system. Same IBM programs. Ask three things: who sizes it, how soon you see the price, and who is with you when the cluster goes live.
Choose a different file and object platform.
Right when your team already runs it everywhere. Compare three things: throughput per rack unit, one namespace across sites and clouds, and a straight path to GPU memory.
Run it in a public cloud.
Right when the data and the GPUs already live in one cloud and will stay there.
Buy it through us.
You get the same IBM system. And you get our engineering team: we size it with you, show you your price first, and stay with you through cutover.
IBM builds the system. We size it with you.
You see your price first.
Tell us what you’re after. One business day later you have a price, sized to fit, with nothing in it you don’t need.
Our engineering team owns it.
The engineers who size it set it up and stay through support. IBM Business Partner since 2008, 29 of 29 IBM proficiencies, one team the whole way.
It is our pleasure to meet you.
230 briefing centers in 113 cities. Pick one; one of our engineers brings the coffee. Or meet us online.
Our engineering team sizes the configuration with you, shows you your price before you commit, and stays through installation and support.
It would be our pleasure to size yours.
How our engineering team delivers
Get it running.
Our engineering team, with IBM Expert Labs (formerly IBM Lab Services)
Build.
We design it, install it and move your workloads across, then hand it over working. Delivered remotely.
Stay connected.
After go-live, IBM’s specialists stay on call beside our engineering team.
One statement of work and one invoice, from us.
One of our engineers replies within one business day.
The proof.
IBM’s published results
Source: IBM’s Storage Scale System page, opened October 23, 2026.
Our trusted and valued customers are shown on our home page. It is our privilege to work for them. See our customers
We would rather tell you now. Storage Scale System is not the fit if your data is mostly block storage for transaction databases, if your file data is modest and grows slowly, or if everything you run already lives in a public cloud. For the first, ask us about FlashSystem.
Pay only for what you need. See your price first, within one business day.
One of our engineers replies within one business day.
AI, automation, data and security, made simple by First Light.
One team sizes it, licenses it, sets it up and supports it.
IBM muscle. First Light hustle.
Enterprise technology, with your quote in one business day.
Give AI the night shift. You take the sunrise.
Build your first IBM watsonx Orchestrate agent on a free 30-day trial.
Trial by sunrise, not by fire.
Most IBM software starts with a 30-day trial and an engineer at your side.
Let Bob write the code. You take the sunrise.
IBM Bob is your AI engineering partner. Try it free for 30 days.