MSI Brings Liquid-Cooled NVIDIA MGX and NVIDIA DGX Station to COMPUTEX 2026

Taipei, Taiwan – Jun. 3, 2026 – MSI, a global leader in high-performance server solutions, returns to COMPUTEX 2026 (Booth #J0605a) with liquid-cooled AI platforms built on NVIDIA MGX and NVIDIA DGX Station architectures for large-scale AI infrastructure and deskside AI development. Designed for demanding AI workloads, the CG681-S6093 AI server and XpertStation WS300 based on NVIDIA DGX Station architecture, deliver high-performance AI computing from data centers to deskside AI workflows.

“As AI infrastructure continues scaling in performance and density, liquid cooling is becoming essential for modern AI data centers,” said Danny Hsu, General Manager of MSI’s Enterprise Platform Solutions. “MSI combines liquid-cooled system design with NVIDIA AI platforms to help customers deploy scalable AI infrastructure with greater performance and efficiency.”

Built on NVIDIA MGX architecture, the CG681-S6093 6U liquid-cooled AI server supports dual AMD EPYC™ processors and up to 8 NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs supporting up to 1200W TDP, delivering the compute density required for AI training and HPC workloads. Equipped with NVIDIA ConnectX-8 SuperNICs providing up to 8×400Gbps Ethernet connectivity, the platform enables high-bandwidth, low-latency communication for distributed AI environments, with a liquid-cooled architecture designed to sustain peak multi-GPU performance.

At the rack level, the liquid-cooled rack-scale system supports up to 4 CG681-S6093 GPU systems within a 48RU configuration, enabling scalable compute for multi-node AI workloads. NVIDIA Spectrum-4 SN5600 scale-out Ethernet and SN2201 out-of-band switches provide high-performance connectivity across AI clusters, while integrated power and thermal design help sustain performance and streamline deployment for large-scale AI training and inference.

The XpertStation WS300 based on NVIDIA DGX Station architecture, brings data-center-class AI computing to the desktop for AI development, fine-tuning, and inference workloads. Powered by NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip with up to 748GB coherent memory and 7.1TB/s HBM3e bandwidth, the system supports Windows-based AI development while enabling high-speed CPU-GPU data sharing for large AI models and data-intensive workflows. Dual 400GbE networking powered by NVIDIA ConnectX-8 SuperNIC supports high-bandwidth connectivity for multi-node AI workflows, while the liquid-cooled design helps sustain AI performance within a compact form factor. During COMPUTEX week, the XpertStation WS300 is also showcased in NVIDIA GTC Taipei, at Taipei International Convention Center, which brings together more applications and use cases.