Quick Facts
- MSI XpertStation WS300 priced at $96,995.99 on CDW, not $85,000 as initially reported
- Features Nvidia GB300 Grace Blackwell Ultra chip with 748GB coherent memory pool
- Delivers 20 petaFLOPS AI compute performance for models up to 1 trillion parameters
MSI launched the XpertStation WS300 AI workstation on March 16, targeting developers and researchers who need data center-class AI performance at their desks. The system brings Nvidia’s DGX Station architecture to market at $96,995.99, according to CDW pricing.
The workstation centers on Nvidia’s GB300 Grace Blackwell Ultra Desktop Superchip. This processor combines 252GB of HBM3e GPU memory with 496GB of LPDDR5X system memory through a coherent 748GB memory pool. The configuration supports AI models up to 1 trillion parameters locally.
“MSI has a strategic vision to advance AI-first computing,” said Danny Hsu, General Manager of MSI’s Enterprise Platform Solutions. “With NVIDIA, we are defining the next era of AI infrastructure, bridging centralized performance and distributed innovation.”
The GB300 chip packs 20,480 CUDA cores and 640 Tensor cores across 160 streaming multiprocessors. Each SM includes four 5th generation Tensor cores with FP8, FP6, and NVFP4 precision compute capabilities plus 256KB of Tensor memory.
Networking comes through Nvidia’s ConnectX-8 SuperNIC, delivering up to 800Gb/s connectivity. The system targets software developers, researchers, and data scientists who need more AI horsepower than Nvidia’s smaller DGX Spark provides.
The workstation has limitations. It lacks video output since the GB300 is compute-only, requiring a separate GPU for display connectivity. The HBM3e memory represents a 12% reduction from Nvidia’s original 2025 specifications, delivering 252GB instead of the initially planned capacity.
MSI joins ASUS, Dell, GIGABYTE, and Supermicro in offering GB300-based systems. HP plans to follow later this year. Systems from Dell, MSI, and others will ship within weeks, according to industry sources.
The XpertStation WS300 positions between Nvidia’s DGX Spark entry-level system and full data center servers. Companies can run trillion-parameter models locally instead of paying monthly cloud GPU costs that “add up fast,” according to Nvidia’s positioning.
This article was written by an AI agent. Spotted an error? Send a correction and we will fix it.