Graid Technology Inc.

10/15/2026

🇯🇵 Bringing GPU-accelerated storage to Japan IT Week Autumn 2026!

Join Graid Technology at the OEC booth to explore storage solutions for demanding AI and HPC workloads.

🔹 SupremeRAID™ AE
Using existing GPUs, SupremeRAID™ AE accelerates NVMe storage with enterprise-grade data protection—without extra RAID hardware, cables, or PCIe slots. Explore AE with OEC’s server solutions for demanding AI workloads.

🔹 SupremeStore
Combining SupremeRAID™ HE and BeeGFS, SupremeStore delivers high-performance cluster storage with scalable capacity and cross-node high availability—supporting the growing storage demands of AI and HPC environments.

📅 October 21–23, 2026
📍 Makuhari Messe, Japan
🤝 OEC Booth A7-5
🌟 Learn More:https://zurl.co/XTp3d

Meet us to discuss storage solutions for your next AI or HPC deployment.

10/13/2026

"You've got a four-lane highway, and to get to your storage, you have to go down to a one-lane bridge."

That's how Garrett McKibben, Senior Director of Technical Marketing at Graid Technology, described the storage bottleneck choking AI inference today, on TechArena's Data Insights podcast with Allyson Klein and Jeniece Wnorowski (Solidigm).

The topic: KV cache, the working memory an inference model holds for every active request. As context length and concurrent users grow, teams run out of room fast. McKibben's math: a Blackwell GPU tops out around 180GB of HBM at $50 to $80 per gig, and concurrency collapses after three or four users. Buying another GPU for every few users isn't a real answer.

Graid Technology's approach moves that overflow to NVMe and accelerates the IO path on the GPU itself, using a small slice of GPU resources before releasing them back. The result, per published testing: faster time to first token than recomputing the context from scratch.

Watch the full episode: https://zurl.co/jP3LJ

10/08/2026

Ships, submarines, aircraft, and remote edge sites need servers that protect and process mission data locally while withstanding shock, vibration, and temperature extremes, all within tight space and power budgets. Systems built the TRS14-18.5 for that environment: an ultra-rugged, short-depth 1U server on Intel® Xeon® 6, rated from -40°C to 50°C, and fully designed and manufactured in the USA.

Trenton Systems uses both Graid Technology storage platforms in the same server:

➡️ VROC™ by Graid Technology provides CPU-native, bootable NVMe RAID for boot and OS volumes, such as a mirrored RAID 1 boot volume, without a dedicated RAID card.
➡️ SupremeRAID™ provides GPU-accelerated RAID for a separate, performance-critical NVMe data array where sustained throughput and low latency are mission requirements.

Each storage tier gets the RAID architecture best suited to its role. Trenton Systems has also demonstrated a configuration pairing SupremeRAID™ with PCIe Gen5 NVMe storage and dual 400GbE networking, designed to scale to nearly 1 PB of raw storage in a short-depth 1U footprint.

Read the blog: https://zurl.co/NnLBY

Attending AUSA in Washington, DC, Oct 12-14? Connect with Mike Flender of Trenton Systems to set up a meeting and explore a configuration built for your mission.

10/06/2026

AI infrastructure doesn't fail because the GPUs are slow. It fails when the storage behind them can't keep up.

Graid Technology CEO Leander Yu and VP of Product Development Alven Yen sat down to explain why the company rebuilt RAID from the ground up, moving core computation onto the GPU to eliminate the bottleneck traditional controllers create, without giving up the redundancy and data protection enterprises depend on.

The result is two complementary platforms built for the AI era: SupremeRAID™, our GPU Accelerated Storage flagship, and VROC™ by Graid Technology, extending that same protection to CPU-native environments from the data center out to the edge.

Full interview and breakdown here: https://zurl.co/xzYQf

10/01/2026

Can your storage keep up with AI?

Join Graid Technology at Cypher 2026 in Bengaluru, October 7–9 -- BOOTH 34 -- to explore storage built for AI workloads.

We’ll spotlight SupremeRAID™ AE, which uses the GPU already in your system to deliver high-performance NVMe storage with data protection, without adding a dedicated RAID card.

At the edge, see how Supreme JBOF extends storage for NVIDIA GB10-class systems. Connected over 100GbE with NVMe-oF over RDMA, it enables protected external KV cache offloading for AI inference.

From local AI workloads to edge deployments, let’s talk about keeping data ready when your models need it. See you there!

-- Cypher 2026--
📍 KTPO Whitefield, Bengaluru
📍 October 7–9, 2026
📍 Booth 34

Learn More:https://zurl.co/SFgQS

09/30/2026

Hardware RAID caps your NVMe array at 12–18% of line rate. Software RAID burns 18–28% of your CPU. With enterprise NVMe up ~257% since Q2 2025, the storage tax has never been more expensive.
Here's how to stop paying it 👇
https://zurl.co/kA7QR

09/28/2026

Offloading KV cache to storage only helps if fetching it is actually faster than recomputing it. That's not always the case — and it's worth checking before you assume you're saving latency.

In a Q&A with TechArena ahead of AI Infra Summit, Graid Technology Field CTO Randy Kreiser breaks down why memory — not FLOPS — now gates inference economics, what a storage tier needs to keep up with GPU memory, and how rack-scale system design is reshaping where storage fits.

👉 Full interview: https://zurl.co/AREai

👋 Headed to 2026? Connect with Kelley Osburn, Graid Technology's Business Development Director, to find out more!

TechArena's Q&A with Graid Technology's Randy Kreiser: why memory now gates inference economics, and how GPU-driven storage keeps up with agentic AI.

09/22/2026

AI infrastructure doesn't fail because the GPUs are slow. It fails when the storage behind them can't keep up.

Graid Technology CEO Leander Yu and VP of Product Development Alven Yen sat down to explain why the company rebuilt RAID from the ground up, moving core computation onto the GPU to eliminate the bottleneck traditional controllers create, without giving up the redundancy and data protection enterprises depend on.

The result is two complementary platforms built for the AI era: SupremeRAID™, our GPU Accelerated Storage flagship, and VROC™ by Graid Technology, extending that same protection to CPU-native environments from the data center out to the edge.

Full interview and breakdown here: https://zurl.co/3sUEf

09/21/2026

We’re proud to share that our partner, Superphi Technology, received the Innovation Solution Award at NVIDIA NPN Bootcamp 2026 China for our jointly developed high-speed local hot data pool solution for quantitative finance.

In financial backtesting, model training, and large-scale market data analysis, GPU performance is only part of the equation. When storage cannot supply data fast enough, valuable GPU resources may be left waiting.

The solution combines:
• 8× Solidigm PS1010 7.68 TB NVMe SSDs

• 8× NVIDIA RTX PRO 6000D Blackwell GPUs

• NVIDIA GPUDirect Storage

• Graid Technology SupremeRAID™ AE

• XFS high-performance file system

In this deployment, SupremeRAID™ AE works with NVIDIA GDS and aggregates eight NVMe SSDs into an approximately 49 TB high-speed data pool, while providing RAID 5 data protection and reducing the RAID workload on the CPU.

Real-world testing delivered:
✅ 27.02 GB/s single-GPU GDS read throughput

✅ 75 GB/s aggregate throughput across eight GPUs

✅ More than 5× higher overall read performance

By combining enterprise NVMe, SupremeRAID™ AE, and NVIDIA GDS compatibility, the solution delivers the storage performance, capacity, and data protection required for demanding quantitative finance workloads.

Congratulations to the Superphi Technology team on this well-deserved recognition. We look forward to continuing our collaboration and bringing more innovative storage solutions to AI, financial services, and accelerated computing.

09/17/2026

"You've got a four-lane highway, and to get to your storage, you have to go down to a one-lane bridge."

That's how Garrett McKibben, Senior Director of Technical Marketing at Graid Technology, described the storage bottleneck choking AI inference today, on TechArena's Data Insights podcast with Allyson Klein and Jeniece Wnorowski (Solidigm).

The topic: KV cache, the working memory an inference model holds for every active request. As context length and concurrent users grow, teams run out of room fast. McKibben's math: a Blackwell GPU tops out around 180GB of HBM at $50 to $80 per gig, and concurrency collapses after three or four users. Buying another GPU for every few users isn't a real answer.

Graid Technology's approach moves that overflow to NVMe and accelerates the IO path on the GPU itself, using a small slice of GPU resources before releasing them back. The result, per published testing: faster time to first token than recomputing the context from scratch.

Watch the full episode: https://zurl.co/ctV0I