The MSI XpertStation WS300 is now available through ASI, D&H, and Newegg, bringing the architecture of the NVIDIA DGX Station to the desktop. At its heart is the NVIDIA GB300 Grace Blackwell Ultra Desktop superchip with up to 748 GB of coherent memory, which is designed to allow companies to develop, fine-tune, and run large AI models locally, rather than renting cloud capacity for this purpose.
TL;DR: The most important facts about the MSI XpertStation WS300
- Based on the NVIDIA DGX Station architecture with the GB300 Grace Blackwell Ultra superchip
- 748 GB of coherent memory, 20 petaFLOPS of AI computing power (FP4, sparse)
- Two 400-GbE ports via NVIDIA ConnectX-8 SuperNIC for clustering two systems
- Supports NVIDIA NemoClaw and the OpenShell runtime for agent-based AI
- Price in the U.S. is approximately $85,000; available through ASI, D&H, and Newegg
Technical Specifications of the MSI XpertStation WS300
Unlike traditional tower workstations, the WS300 does not use a multi-chip CPU-GPU duo. Instead, the chassis houses a single GB300 superchip that connects the CPU and GPU via NVIDIA NVLink-C2C with up to 900 GB/s of bandwidth, thereby bypassing the bottleneck that typically occurs between the processor and the graphics card. 748 GB of memory in a single desktop case is a figure that, until now, was more commonly associated with data centers than with one’s own desk.
| Component | Specifications |
|---|---|
| Superchip | NVIDIA GB300 Grace Blackwell Ultra (72-core Grace CPU + Blackwell Ultra GPU) |
| Memory | 748 GB coherent (252 GB HBM3e GPU + 496 GB LPDDR5X CPU) |
| AI Computing Power | up to 20 petaFLOPS (FP4, sparse) |
| Network | 2× 400-GbE QSFP112 via ConnectX-8 SuperNIC, 1× 1000Base-T Management |
| Storage | 2× PCIe 5.0 NVMe M.2 (2 TB RAID 1 preconfigured), 2× free PCIe 6.0 M.2 slots |
| Expansion | 3× PCIe 5.0 slots, TPM 2.0, hardware Root of Trust |
| Operating System | Ubuntu 24.04 LTS with pre-installed NVIDIA AI Developer Tools |
| Power Consumption | 1,600 W (single PSU, 80 PLUS Platinum) |
According to MSI, users who need more performance can connect two WS300 systems into a cluster via the integrated 400-GbE ports. This positions the workstation between traditionalhigh-end PC workstations and full-fledged data center racks.
NVIDIA DGX Station: What’s Behind the Architecture
The WS300 is not an MSI in-house design, but rather one of several implementations of the NVIDIA DGX Station reference architecture, which NVIDIA offers in collaboration with partners such as ASUS, Dell, Gigabyte, HP, and Supermicro. All systems share the same GB300 superchip but differ in terms of chassis, cooling, and additional features.
“AI is moving from experimentation to always-on, agentic workloads that demand powerful computing close to where developers and enterprise data reside,” says Allen Bourgoyne, Director of Enterprise Products at NVIDIA. In other words: Whereas individual chatbot requests were previously the focus, continuously running agents and large language models are now intended to perform computations directly on-site, rather than incurring cloud costs and allowing sensitive data to leave the premises.
For German companies that previously considered on-premises AI too costly, this is the real highlight of the DGX Station class: a preconfigured system that doesn’t require its own server room and runs on a standard power outlet.
Price and Availability
MSI itself does not mention a price in the press release and refers interested parties to a custom quote request. According to consistent reports from TechRadar and Newegg listings, the price in the U.S. is around $84,999. Other DGX Station models from ASUS or Supermicro also fall within this six-figure range.
In Germany, specialized system integrators, among others, distribute the DGX Station series, though without a published price list. Offers are typically provided upon request, with a delivery time of about four to ten weeks. Anyone investing in this price range realistically doesn’t compare the purchase to a gaming PC with a high-end graphics card, but rather to reserved cloud GPU time or their own small data center.
NemoClaw and OpenShell: The Step Toward Agent-Based AI
According to MSI, beyond pure model development, the XpertStation WS300 also supports NVIDIA NemoClaw in combination with the OpenShell runtime. NemoClaw is NVIDIA’s response to the growing number of autonomous AI agents that perform tasks independently in the background: The runtime is designed to run such agents in a secure, policy-driven environment rather than granting them unrestricted access to corporate systems.
For companies that—for example, due to AI labeling requirements and other regulatory mandates—must already document in greater detail what their AI systems do, such a security and control model is likely to be at least as interesting as raw computing power.
Who stands to benefit from the MSI XpertStation WS300?
The WS300 is not intended for home users, but rather for AI teams, research labs, and departments that have previously relied on shared cluster resources or cloud GPU instances. Specifically, the following will benefit:
- Teams that want to develop or fine-tune models with up to one trillion parameters locally
- Companies with strict data sovereignty requirements whose data must not leave the company premises
- Developers of inference engines and robotics teams who require reproducible, exclusive access to Blackwell Ultra hardware
Conclusion
With the MSI XpertStation WS300, the NVIDIA DGX Station architecture becomes accessible to enterprises that prefer to run generative AI and agent-based workloads in-house rather than in the cloud in the future. From a technical standpoint, the large coherent memory is particularly impressive, making it possible to handle models exceeding the 500-billion-parameter threshold without swapping to RAM. However, the price of around $85,000 clearly places the system outside the consumer and even the traditional prosumer segments. Here, you’re paying for independence from the cloud, not for a faster gaming experience.
