The startup that took a 1980s idea - RAID - and rebuilt it around the graphics processor, unlocking the full speed of NVMe flash for the AI era.
RAID - the decades-old trick of spreading data across multiple drives for speed and safety - was designed for spinning disks. When NVMe flash arrived, drives got so fast that the traditional RAID controller became the bottleneck. Graid Technology's answer was not a faster controller. It was a different chip entirely.
Founded in 2019, Graid Technology builds SupremeRAID, marketed as the world's first GPU-based RAID for NVMe and NVMe-over-Fabrics SSDs. Instead of running parity math on a dedicated ASIC or borrowing cycles from the host CPU, SupremeRAID hands the work to a compact NVIDIA GPU. The name itself is the thesis: GPU plus RAID.
The clever part is where the GPU sits. It stays out of the data path - it tells the SSDs what to do, but the actual data never routes through it. That removes the classic RAID chokepoint and lets flash drives run at close to their native speed while staying protected against failure.
A single SR-1010 card can protect up to 32 NVMe drives and push around 110 GB/s of sequential read throughput. The smaller SR-1001 fits one PCIe slot, runs an NVIDIA T400-class GPU, and still reaches roughly 6 million IOPS - while drawing only 30 watts.
The point is not raw bragging rights. It is that offloading RAID to a GPU frees CPU cores for the actual workload - video rendering, simulations, database queries, AI training - instead of burning them on storage overhead.
PCIe Gen4 card for up to 32 native NVMe SSDs. Up to ~19M read IOPS and 110 GB/s. The data-center workhorse.
Single-slot Gen3 card for up to 8 drives at just 30W - built for edge servers, workstations and high-end rigs.
Tuned for AI/ML with NVIDIA GPUDirect Storage, enabling direct NVMe-to-GPU memory transfers.
Maximum-bandwidth configuration aimed at AI and HPC clusters that need every gigabyte per second.
Editions extending GPU RAID to workstations (with GeForce RTX 5000 support) and specialized high-end enterprise use.
Virtual RAID on CPU with a 24-month roadmap and Tier 1 OEM support, broadening the software-defined portfolio.
Enterprises, cloud and government data centers, media and HPC operators, and AI/ML teams - typically reaching them through server OEMs and distributors.
GPU RAID cards bundled with a software-defined RAID engine and tiered per-drive licenses, sold direct and via OEM and channel partners.
The bridge between raw NVMe flash and the AI/HPC workloads starving for storage bandwidth - a new "GPU RAID" category.
How it differs from competitors. Traditional hardware RAID (Broadcom/LSI MegaRAID, Microchip Adaptec) leans on a dedicated ASIC. Intel VROC and open-source tools like Linux mdadm and ZFS lean on the CPU. Graid leans on the GPU, out of the data path - a genuinely different architecture rather than a faster version of the same one. Analysts have noted the approach gave Graid a multi-year technology lead.
| Round | Amount | Date | Lead / Notable Investors |
|---|---|---|---|
| Seed | $2.5M | 2019 | Early backers |
| Series A | $15M | 2022 | Series A investor group |
| Series B | $30M | Mar 2025 | HH-CTBC Partnership (Foxconn + CTBC), Yuanta Ventures, Delta Electronics Capital, Harbinger Venture Capital |
Total raised ~$47.5M. The Series B was framed as the final private round before a planned public listing in Taiwan.
25+ years in the storage industry and a National Taiwan University computer science graduate. Started Graid with $2.5M in seed funding and a conviction that the GPU was storage's future.
Proposed harnessing GPU acceleration for RAID. Specializes in CUDA development, massively parallel algorithms and GPU-optimized storage pipelines.
SupremeRAID runs RAID I/O on NVIDIA GPUs and supports GPUDirect Storage. Graid was named an NVIDIA "Strategic 50" startup.
SupremeRAID SKUs ship through Dell.com configurators; Supermicro deployments scale to hundreds of systems.
Broad CPU-platform compatibility and validated NVMe SSD ecosystem partners.
Global and public-sector distribution reaching enterprise and government buyers.
Leander Yu launches the company with ~$2.5M seed funding and a thesis to offload RAID onto GPUs.
The company introduces SupremeRAID, positioned as the world's first GPU-based NVMe RAID.
Graid raises a $15M A round and ships its flagship 32-drive card.
Launches the single-slot SR-1001 for edge and workstation use as partnerships expand.
Closes a pre-IPO Series B and releases SupremeRAID AE and Ultra for AI and HPC.
Wins recognition at Interop Tokyo 2026 and the Taiwan AI Awards, and launches VROC by Graid.
SupremeRAID, a GPU-based RAID solution for NVMe and NVMe-over-Fabrics SSDs. It pairs a small NVIDIA GPU card with a software-defined RAID engine to protect and accelerate flash storage.
Traditional controllers process parity on a dedicated ASIC or the host CPU, which bottlenecks fast NVMe drives. SupremeRAID offloads RAID calculations to a GPU that sits out of the data path, so SSDs run at near-native speed.
Roughly $47.5M total: about $2.5M seed (2019), $15M Series A (2022), and a $30M Series B in March 2025 led by HH-CTBC Partnership and Yuanta Ventures.
Founded in 2019 by CEO Leander Yu and Chief Architect David Tseng, with operations in Santa Clara / Sunnyvale, California and Taiwan.
Enterprises, cloud and government data centers, media and HPC operators, and AI/ML teams - usually via server OEMs like Dell and Supermicro and distributors such as Arrow and TD Synnex.