MSI Launches XpertStation WS300 Powered by NVIDIA Grace Blackwell

The system features up to 748GB of coherent memory and integrates dual 400GbE networking through NVIDIA ConnectX-8 SuperNICs, allowing two units to cluster for increased scalability. By keeping large foundation models resident in coherent memory, the WS300 minimizes CPU-GPU data movement, a configuration designed to improve inference efficiency for multimodal AI and complex fine-tuning tasks.

Danny Hsu, General Manager of MSI's Enterprise Platform Solutions, noted that the hardware is built to eliminate the months of integration typically required for enterprise AI deployment. Beyond pure compute, the machine supports NVIDIA NemoClaw and the OpenShell runtime, providing a framework for secure, policy-controlled autonomous agents. This shift toward on-premises infrastructure allows companies to retain control over proprietary data and reduce the long-term costs associated with cloud-based inference.