Supermicro Unveils Advanced AI Storage Server with NVIDIA Grace Chip
San Jose, Wednesday, 19 March 2025.
Supermicro has launched a petascale all-flash storage server powered by the NVIDIA Grace CPU Superchip, enhancing AI data processing capabilities across European business and research centers.
Revolutionary Storage Architecture
The newly announced system, unveiled at GTC 2025 on March 17, represents a significant leap in storage capabilities, featuring 16 hot-swap EDSFF PCIe Gen5 E3.S NVMe drives that enable up to 983TB of raw capacity using 61.44TB SSDs [1]. When fully deployed, a single rack containing 40 systems can provide an impressive 39.3PB of raw storage capacity [1]. The server’s architecture incorporates 144 Arm cores and integrates 960GB of LPDDR5X memory, alongside support for two NVIDIA BlueField-3 or ConnectX-8 SuperNICs [2].
Industry Collaboration and Performance
Supermicro has strategically partnered with NVIDIA and WEKA to optimize the system’s performance [1]. According to NVIDIA’s vice president of storage networking technology, Rob Davis, this collaboration enables customers to leverage NVIDIA’s innovations in both CPUs and DPUs, accelerating networking protocols while consuming less power than comparable x86 servers [1]. Initial testing of the WEKA Data Platform software on this pre-production cluster has demonstrated remarkable performance scaling and power efficiency [1].
Market Impact and Industry Response
The launch comes amid a broader industry movement toward advanced AI infrastructure, with eleven major storage suppliers aligning their technologies with NVIDIA’s ecosystem [5]. The timing is particularly significant as it coincides with Supermicro’s expansion of its enterprise AI portfolio, which now includes over 100 GPU-optimized systems [4]. Market response has been notably positive, with industry leaders emphasizing the system’s potential for AI workloads and enterprise applications [5].
Technical Specifications and Future Outlook
The system features a 600W TDP NVIDIA Grace CPU Superchip and comes equipped with redundant Titanium 1600W Power Supplies [2]. The storage server will be showcased in detail at NVIDIA GTC in San Jose through an On-Demand session (S74296) [1]. While specific pricing details remain [alert! ‘pricing information not provided in sources’], the system’s architecture suggests a focus on high-performance computing centers and enterprise AI applications requiring substantial storage capabilities.