
What happens when a startup builds something so unconventional that Nvidia's Jensen Huang sends his team to investigate?
That's precisely what happened to Graid Technology after an article about their GPU-powered RAID solution reached Nvidia's executive floor. What followed was a quiet but growing endorsement of a bold idea: traditional RAID controllers aren't built for a world of NVMe, GPU-driven compute, and AI data demands.
I caught up with the team during IT Press Tour #62, where they unpacked how a bootstrapped prototype has turned into a global deployment engine backed by fresh funding, real-world customer stories, and high-profile partnerships.

From Stealth to Scale
Graid didn't follow the typical enterprise storage playbook. Instead of building another card with a chip on it, they utilized Nvidia's off-the-shelf GPUs and developed RAID software specifically designed to run directly on them.
"When we go to market on an off-the-shelf Nvidia GPU, that means this product is available worldwide," one team member said. "We didn't have to worry about regulatory overhead. Nvidia already handled that."
Since emerging from stealth, the company has scaled up rapidly.
Headcount has grown by over 40 percent in the Americas and more than 80 percent in EMEA. They now have around 80 employees and are expanding into new verticals, including government, AI infrastructure, and HPC. Their reseller network isn't just growing in size; it's maturing—delivering sales in large batches rather than isolated transactions.
Built for Performance, Without the Bottlenecks
SupremeRAID doesn't try to speed things up. It removes the slowdown. That alone sets it apart in a market full of products promising to accelerate, turbocharge, or enhance throughput.
Their newest version, SupremeRAID AE, supports GPU Direct Storage, allowing it to coexist with other workloads in a GPU server and protect NVMe drives without interrupting the primary compute task. "We only use six SMs on a GPU with 144. If there's no I/O, we go to sleep and let the GPU focus on computing," the team explained.
The goal isn't to take over the system, it's to stay out of the way while delivering data protection that doesn't penalize performance.
Why Customers Are Paying Attention
At the University of Southampton, a medical imaging team saw a fourfold performance improvement after replacing Broadcom RAID cards with SupremeRAID. Their 3D rendering workflow had been stuck. Now, it moves in real-time.
In another case, a major credit card company subjected the technology to rigorous testing in a lab before deploying it in production. The result? RAID 6 protection at RAID 10 performance. This resulted in improved fault tolerance with lower infrastructure costs. And then there's the US Army. At Aberdeen Proving Grounds, Graid is now deployed across 60 edge systems. The military demanded what they called "military-grade journaling." Graid delivered. Their journaling system was accepted without changes.
From media production in Europe to federal contracts in the United States, these stories paint a clear picture: Graid is no longer pitching theoretical performance. They have thousands of real customers and growing momentum.
It's easy to get lost in the alphabet soup of RAID, SSD, PCIe, and GDS. What Graid does differently is strip away the complexity for the end user. Install the card or license the software. Plug it into existing infrastructure. That's it.
One of their more forward-looking offerings, the AE edition, even lets users bring their own GPU. "Imagine you've got four GPU servers in a cluster each with eight NVMe drives. You only need one AE license to protect them all," they said.
That simplicity is helping them scale into sectors previously dominated by legacy vendors. Whether in hyperscale, financial services, or defense, the common theme is the same: traditional RAID is holding systems back. Graid removes that obstacle.

A Seat at the Table with Nvidia
At Nvidia's headquarters, Graid was invited to join the Storage Next initiative. The room included representatives from Dell, Microsoft, Samsung, Seagate, Hitachi, and a few other companies. Graid was not just a guest. They were given five minutes to present to Nvidia's top storage researchers. One of them told the team: "We think what you're doing is going to be very important to where we're heading."
That sentiment was echoed when A10 Networks released a public graphic showing the future of AI infrastructure. In the "storage" section, Graid was the only company named. That level of recognition doesn't come from marketing. It comes from cleverly solving a real problem.
Consolidating the Lineup, Preparing for What's Next
Graid is now moving toward a more streamlined product catalog. The new single-slot Ampere-based cards deliver the same performance as older double-slot models but take up less space. That change alone has already increased adoption in Europe, where rack space is tighter.
There's also work underway on a new HPC-focused edition designed in collaboration with Supermicro. This version includes improved failover and node-to-node data migratio features designed to replace software like Weka in high-density computing environments.
Further down the roadmap is support for larger RAID groups and new release builds for both Linux and Windows. Eventually, Graid expects to support RAID over fabrics, enabling a single instance to manage storage across multiple servers or JBODs.
What Makes This Different
During our conversation, I asked what made Graid stand apart in a sea of storage vendors. Their answer was simple: "We don't try to make your system go faster. We stop slowing it down."
It's not just about PCIe Gen 4 or Gen 5 support though they saturate both. It's about getting out of the way, whether the customer is running AI inference, medical scans, or Splunk queries.
Graid has added a REST API, a new GUI, journaling, retry logic, and a bunch of practical features often overlooked in flashy product pitches. For developers working within OEMs, the API is fully open. "We've had some of our biggest partners shaping what the API looks like today," they said.
Storage isn't going away, but it is changing. The systems we're building now are faster, denser, and more expensive to keep idle. The last thing anyone wants is a RAID controller causing downtime in a half-million-dollar GPU server.
Graid's approach is fresh but grounded. They didn't try to invent a new format or push people into proprietary hardware. They utilized something many already had Nvidia GPUs and transformed it into a more innovative approach to data protection.
What does this mean for the future of enterprise infrastructure? Could we be witnessing a broader shift where the GPU assumes more system-level roles beyond computing? Or is Graid a rare case of software cleverly adapting to new hardware trends?
I'd love to hear what you think. Does GPU-powered RAID belong in your infrastructure plan?
I will be speaking with the team at Graid on the Tech Talks Daily Podcast in the next few weeks. If you have any questions you would like me to ask, please let me know, and you can also be a part of the conversation.