MSI’s deskside NVIDIA DGX Station implementation · GB300 Grace Blackwell Ultra Desktop Superchip · agentic AI with NemoClaw, OpenShell and NeMo Switchyard
MSI’s deskside AI supercomputer, built on the NVIDIA DGX Station architecture.
Compact always-on agent appliance derived from the NVIDIA DGX Spark platform.
Both platforms ship validated against NVIDIA’s current agentic software layers.
MSI (Micro-Star International)
XpertStation WS300
EdgeXpert
Deskside AI supercomputer
Compact agent appliance
NVIDIA DGX Station architecture
NVIDIA DGX Spark platform
GB300 Grace Blackwell Ultra Desktop
GB10 Grace Blackwell (EdgeXpert)
Up to 748 GB (WS300)
128 GB (EdgeXpert)
Up to 1 petaFLOP (EdgeXpert)
2 × 400GbE ConnectX-8 SuperNIC
ConnectX-7 (EdgeXpert)
2 × WS300 direct connect
4 × EdgeXpert via ConnectX-7
PCIe Gen5 / Gen6 NVMe
Nemotron 3.5 Lightning
Models up to 200B parameters on-device
NVIDIA NeMo Switchyard
NVIDIA NemoClaw with OpenShell
Shipping since 27 August 2026
ASI, D&H, Newegg
Quote
MSI XpertStation WS300 is a deskside AI supercomputer built on the NVIDIA DGX Station architecture and powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip. With up to 748 GB of coherent memory and dual 400GbE networking through the ConnectX-8 SuperNIC, it is designed to run demanding agentic and multi-model AI workloads locally rather than against a metered cloud endpoint. Two units can be clustered directly for scale-out, and PCIe Gen5/Gen6 NVMe storage sustains multi-model serving.
MSI is shipping the XpertStation WS300 through ASI, D&H and Newegg as of 27 August 2026. The system supports the NVIDIA NemoClaw security layer with the OpenShell runtime for policy-controlled autonomous AI agents, and runs Nemotron 3.5 Lightning — a 30B-parameter mixture-of-experts model with 3B active parameters and context windows up to one million tokens — alongside NVIDIA NeMo Switchyard, an open-source routing library that sends each workflow step to the best-suited model without manual configuration.
Alongside it, MSI EdgeXpert extends the same software story to a compact, always-on appliance based on the NVIDIA DGX Spark platform. Powered by the GB10 Grace Blackwell Superchip with up to 1 petaFLOP of AI compute and 128 GB of coherent unified memory, it runs models up to 200 billion parameters continuously on-device, and clusters up to four units over ConnectX-7. Both platforms target the same shift: enterprises keeping inference on premises to cut recurring cloud cost and retain control of proprietary models and data.
Need MSI XpertStation WS300?
Contact QS Compute for availability, configuration, and volume pricing.
Request Quote