MSI has launched XpertStation WS300, a deskside system built on NVIDIA DGX Station architecture for running LLM, generative AI, and data science workloads locally. The system is powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip and is available to order starting today.
XpertStation WS300 supports up to 748 GB of “large coherent memory,” combining HBM3e GPU memory and LPDDR5X CPU memory into a unified memory domain for CPU-to-GPU data sharing during model training and fine-tuning. For networking, the system includes dual 400GbE via an NVIDIA ConnectX-8 SuperNIC, providing up to 800 Gb/s of aggregate bandwidth for distributed AI workloads and scaling across multiple nodes. MSI also lists high-speed PCIe Gen5 and Gen6 NVMe storage support aimed at accelerating dataset ingestion and AI data pipelines, alongside support for the NVIDIA AI Software Stack.
For data center teams, the interesting angle isn’t that “AI is going deskside”—it’s that workstation-class placement is being paired with data-center-style fabrics. Dual 400GbE in a deskside box is a clear signal that MSI expects these systems to plug into serious networks and participate in multi-node workflows, not just run single-user experiments under a desk.
Danny Hsu, General Manager of MSI’s Enterprise Platform Solutions, said, “With NVIDIA, we are defining the next era of AI infrastructure, bridging centralized performance and distributed innovation, and enabling organizations to move from experimentation to production with greater speed, scale, and confidence.”
MSI also says the system can be used as a centralized AI compute node for collaborative fine-tuning and on-demand deployment, while keeping proprietary data and IP under local control. In addition, MSI points to NVIDIA NemoClaw—described as an open-source stack that installs an OpenShell runtime with a policy-controlled sandbox for autonomous AI agents—and says developers can run trillion-parameter models locally on XpertStation WS300 with up to 20 petaFLOPS of AI compute and 748 GB of memory.
Source: MSI
