Nvidia has unveiled a new category of high-speed storage server aimed squarely at the demands of modern AI data centers, and the numbers are staggering. The system, known as a SCADA server, offers up to 2.9 petabytes of storage and is built around the next-generation PCIe 6.0 interface — a standard that roughly doubles the data transfer bandwidth of its predecessor.
According to Tom's Hardware, Wiwynn is among the first manufacturers to demonstrate the Nvidia SCADA server in public. The design relies on GPU-accelerated storage, meaning Nvidia's graphics processors aren't just crunching AI calculations — they're also being put to work speeding up how quickly data moves in and out of storage. This approach is intended to eliminate the bottlenecks that emerge when AI systems need to rapidly access enormous datasets during training and inference.
To put the scale in perspective, 2.9 petabytes is roughly equivalent to about 2.9 million gigabytes — enough to store tens of millions of high-resolution images or years' worth of video footage. For AI workloads that constantly shuttle massive amounts of data between memory and storage, having that capacity accessible at extreme speed is a critical engineering challenge.
Wiwynn's demonstration suggests the hardware ecosystem around Nvidia's AI infrastructure ambitions is beginning to take concrete shape, moving from concept to physical product.
As AI models grow larger and training runs consume ever-greater volumes of data, storage performance — not just raw compute power — is becoming a defining bottleneck, making this class of server a potential cornerstone of next-generation AI infrastructure.