Supermicro unveils petascale storage server with NVIDIA Grace CPU for AI workloads

Supermicro’s new petascale storage server represents a significant advancement in AI infrastructure by combining NVIDIA’s Grace CPU Superchip with high-density storage capabilities specifically designed for AI workloads. This collaboration between Supermicro, NVIDIA, and WEKA addresses the growing demand for efficient, high-performance storage solutions needed to support the massive datasets required for AI and ML operations, potentially transforming how organizations build their AI computing environments.

The big picture: Supermicro has introduced a first-of-its-class petascale all-flash storage server optimized for high-performance software-defined storage workloads in AI environments.

• The system utilizes the NVIDIA Grace CPU Superchip with 144 Arm Neoverse V2 cores, providing a power-efficient alternative to traditional x86 servers for software-defined storage workloads.
• This release represents the first major Tier 1 provider to offer this specific combination of technologies for AI-focused storage infrastructure.

Key specifications: The new storage server delivers unprecedented storage density with support for 16 hot-swap EDSFF PCIe Gen5 E3.S NVMe drives in a single unit.

• Each server supports up to 983TB of raw storage capacity when configured with 61.44TB SSDs.
• At scale, a single rack containing 40 systems can provide an impressive 39.3PB of raw storage capacity.

Strategic partnerships: Supermicro collaborated closely with NVIDIA and WEKA to create a solution specifically tailored for AI factory environments.

• Rob Davis from NVIDIA highlighted how customers can benefit from “NVIDIA’s innovations in both CPUs and DPUs to accelerate networking protocols, such as GPUDirect Storage, while using less power than comparable x86 servers.”
• WEKA’s Nilesh Patel noted that initial testing shows “significant performance scaling and power efficiency” when running WEKA Data Platform software on the pre-production cluster.

Why this matters: As AI models grow increasingly complex, the infrastructure supporting them requires specialized solutions that balance performance, storage capacity, and energy efficiency.

• The petascale architecture provides the massive storage capacity needed for large AI training datasets while maintaining the performance characteristics required for time-sensitive AI operations.
• The solution’s power efficiency addresses growing concerns about the environmental impact of AI infrastructure.

Between the lines: This announcement reflects the industry’s shift toward purpose-built hardware solutions for AI rather than repurposing general-purpose computing infrastructure.

• By integrating NVIDIA’s ARM-based Grace CPU rather than traditional x86 processors, Supermicro is betting on specialized, energy-efficient architectures for the next generation of AI infrastructure.