As the global high-performance computing and AI wave continues to accelerate, MSI (Micro-Star International) once again demonstrated its comprehensive AI infrastructure deployment at Supercomputing 2025 (SC25) in St. Louis, Missouri, USA. Centered on “deepening the AI computing ecosystem and delivering end-to-end high-performance solutions,” MSI unveiled multiple next-generation AI platforms spanning data centers, edge computing, and desktop-grade AI supercomputers, showcasing a complete product strategy from cloud to personal computing. The systems announced this time are all designed based on NVIDIA’s latest MGX™, NVIDIA DGX Station™, and NVIDIA DGX Spark™ architectures, featuring the NVIDIA Hopper GPU, NVIDIA RTX PRO 6000 Blackwell Server Edition GPU, NVIDIA Blackwell Ultra GPU, and the integrated Grace Blackwell GB10 Superchip as the latest-generation computing cores, underscoring its highly competitive position for the AI era.
Data center-grade AI servers: Built for the most demanding LLM and deep learning workloads
In response to the rapidly growing enterprise demand for large language model (LLM) training and accelerated simulation, MSI has launched a range of highly scalable AI servers that support Intel® Xeon® and AMD EPYC™ platforms and can be paired with GPUs up to 600W, designed for deep learning, high-efficiency transmission, and cluster deployment.
● CG481-S6053 — High-bandwidth AI cluster built on AMD flagship platform
This 4U server features dual AMD EPYC 9005 processors, equipped with eight PCIe 5.0 FHFL double-width GPU slots, 24 DDR5 memory slots, and eight U.2 NVMe storage devices. It further provides eight 400GbE connections using NVIDIA ConnectX-8 SuperNICs, targeting the largest-scale AI training cluster deployments.
● CG480-S5063 — Mainstay for Intel architecture deep learning training
Also featuring a 4U architecture, equipped with dual Intel Xeon 6 processors and eight GPU slots, and offering 32 DDR5 memory slots and up to twenty PCIe 5.0 E1.S NVMe storage drives, specifically optimized for model fine-tuning and algorithm research environments.
● CG290-S3063 — 2U dedicated to edge and small-scale inference
Equipped with a single-socket Intel Xeon 6 platform, supporting four 600W GPUs and 16 DDR5 memory modules, it is suitable for deployment in edge data centers or small to medium-sized AI inference services.
Danny Hsu, General Manager of MSI’s Enterprise Platform Solutions Business Unit, stated:
「隨著 AI 工作負載日益複雜，微星最新的 MGX 架構平台可協助企業高效擴充，同時最大化 GPU 頻寬與運算密度，以滿足 AI 時代的運算需求。」
AI Station CT60-S8060: Data center-grade computing power enters the personal development space
In addition to enterprise-level solutions, MSI also aims to break the mold that large-scale computing is limited to data centers. The AI Station CT60-S8060 on display brings NVIDIA DGX Station™-class capabilities to the desktop. Powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, the CT60-S8060 offers up to 784GB of Unified Memory, allowing developers to run large-scale model training, inference testing, and deployment validation directly in a personal environment. This not only accelerates development but also significantly reduces cross-platform conversion and data transfer costs.
Xu Yanwen stated:
「AI Station 將資料中心的強大效能帶給個人創作者與研究人員，為次世代 AI 創新開啟更多可能。」
In use cases such as generative AI development, Omniverse virtual scene building, and scientific research simulation, desktop-class devices like this are poised to become the standard tool for a new generation of innovators.
EdgeXpert: 1.2-Liter Compact AI Supercomputer Officially Launched
MSI also announced that EdgeXpert, designed for educational institutions and corporate research labs, officially launched on October 15, 2025. EdgeXpert is powered by the NVIDIA GB10 Grace Blackwell Superchip and features 128GB of Unified Memory. Despite its compact 1.2-liter size, it delivers professional-grade performance capable of running a wide range of AI models, making it an ideal desktop solution for AI experimentation, course instruction, and mid-scale inference deployments.
David Wu, General Manager of MSI’s Customization Products Division, pointed out:
「EdgeXpert 打通了 AI 研究與實際應用的最後一哩路，讓開發者與教育單位能在桌面環境中，即享安全、高效且具成本效益的 AI 運算平台。」
From data center to edge, then to the desktop: MSI’s AI ecosystem strategy is gradually taking shape.
From the product strategy revealed at SC25, it is not difficult to see that MSI is building its AI infrastructure landscape in three directions:
| AI Deployment Level | Representative product | Target application areas |
|---|---|---|
| Data center | CG481-S6053、CG480-S5063 | Large model training, high-bandwidth AI clusters |
| Edge computing | CG290-S3063 | Low-latency inference, small-scale deployment |
| Desktop / Personal Devices | AI Station CT60-S8060、EdgeXpert | Creative Development, Education, and Research |
As AI models continue to grow in scale and expand into new applications, computing resources that were once confined to the cloud and large data centers must also move down into more real-world environments. The solutions MSI showcased at SC25 reflect its push to bring high-performance computing closer to the user level, enabling everyone from content creators to edge device managers to access more direct and affordable AI computing power.
Source: KOCPC Chinese