MSI Unveils Next-Generation AI Computing Solutions at Supercomputing 2025

MSI, a recognized leader in high-performance computing, is set to showcase its latest AI computing platforms at Supercomputing 2025 (SC25) in St. Louis, Missouri. The new lineup is built on advanced NVIDIA reference architectures, including NVIDIA MGX, NVIDIA DGX Station, and NVIDIA DGX Spark, and leverages the power of NVIDIA Hopper GPUs, NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs, NVIDIA Blackwell Ultra GPUs, and the Grace Blackwell GB10 Superchip.

Comprehensive AI Infrastructure for Diverse Workloads

MSI’s new AI servers are engineered to deliver data-center-class performance across a wide range of artificial intelligence workloads. From large-scale model training and simulation to edge inferencing and desktop AI development, these platforms are designed to meet the evolving needs of enterprises, researchers, and developers.

The servers are optimized for large language models (LLMs), deep learning, and NVIDIA Omniverse applications. They offer flexible configurations with support for both Intel Xeon and AMD EPYC processors, and accommodate up to 600 W GPUs, ensuring maximum performance and scalability for demanding AI tasks.

Key AI Server Platforms

  • CG481-S6053 (4U, AMD Platform): Features dual AMD EPYC 9005 CPUs, eight FHFL PCIe 5.0 dual-width GPU slots, 24 DDR5 DIMMs, eight U.2 NVMe bays, and eight 400 GbE ports powered by NVIDIA ConnectX-8 SuperNICs. This configuration is ideal for building high-bandwidth AI clusters.
  • CG480-S5063 (4U, Intel Platform): Equipped with dual Intel Xeon 6 CPUs, eight FHFL dual-width GPU slots, 32 DDR5 DIMMs, and twenty PCIe 5.0 E1.S NVMe bays. This platform is optimized for deep learning training and fine-tuning workloads.
  • CG290-S3063 (2U): Offers a single Intel Xeon 6 CPU, 16 DDR5 DIMMs, and four dual-width GPU slots (up to 600 W each), making it suitable for compact edge computing and small-scale inferencing deployments.

According to Danny Hsu, General Manager of MSI's Enterprise Platform Solutions, these platforms enable customers to efficiently scale their AI infrastructure while maximizing GPU bandwidth and computational density to address the increasing complexity of AI workloads.

AI Station CT60-S8060: Bringing Data-Center Performance to the Desktop

For developers and researchers who require data-center-level compute power in a workstation form factor, MSI introduces the AI Station CT60-S8060. Built on the NVIDIA DGX Station architecture and powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, this workstation offers up to 784 GB of unified memory. The AI Station CT60-S8060 delivers DGX-class performance, enabling users to develop, train, and deploy large-scale AI models directly from their desktop environment.

As Danny Hsu notes, the AI Station empowers individual creators and researchers by providing access to high-performance AI capabilities outside the traditional data center.

EdgeXpert Personal AI Supercomputer: Compact Power for Education and Research

The MSI EdgeXpert Personal AI Supercomputer is now available, offering professional-grade AI performance in a compact 1.2-liter form factor. Powered by the NVIDIA GB10 Grace Blackwell Superchip with 128 GB of unified memory, EdgeXpert is designed for education, research, and enterprise AI labs seeking secure, high-performance, and affordable AI computing solutions.

David Wu, General Manager of MSI's Customized Product Solutions BU, highlights that EdgeXpert bridges the gap between AI research and real-world deployment, making advanced AI computing accessible to developers and educators directly on their desktops.