COMPUTEX 2026: MSI Goes All-In on Liquid-Cooled AI Infrastructure and NVIDIA Blackwell Solutions
Taipei, Taiwan — As compute density reaches new heights, MSI is utilizing its floor space at COMPUTEX 2026 (Booth #J0605a) to signal a massive shift toward liquid-cooled data center solutions. From massive rack-scale deployments to desktop-bound supercomputing, the message is clear: the AI era is running hot, and MSI has the plumbing to handle it.
The showcase features a comprehensive range of platforms built around the NVIDIA MGX and DGX Station architectures, alongside modular DC-MHS designs for cloud and enterprise scaling.
Cooling the AI Beast: Rack-Scale Liquid Infrastructure
Thermal management is the primary bottleneck for 2026’s hardware. MSI is addressing this with its OCP ORv3 Liquid-Cooled Rack Architecture. This 21-inch, 44OU setup is designed for massive 100kW deployments, utilizing a Liquid-to-Liquid Coolant Distribution Unit (L2L CDU) to keep high-density AI nodes from throttling.
For standard data centers, MSI also displayed 19-inch EIA air-cooled racks, supporting both AMD EPYC™ 9005 and Intel® Xeon® 6 platforms, providing a flexible path for enterprises not yet ready to make the jump to full immersion or direct-to-chip liquid cooling.
NVIDIA MGX & The Blackwell Era
MSI’s collaboration with NVIDIA is front and center, focusing on the NVIDIA Blackwell and Grace Blackwell architectures.
The Powerhouse (CG681-S6093): A liquid-cooled 6U beast supporting up to eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. With 32 DDR5 DIMMs and 400Gbps networking via ConnectX-8 SuperNICs, this is built for the most demanding LLM training and HPC workloads.
Edge & Inference (CG290-S3063): For space-constrained environments, this 2U Intel Xeon 6 platform supports four double-wide GPUs, targeting the growing need for local AI inference.
XpertStation WS300: The Supercomputer Under Your Desk
Perhaps the most exciting reveal for developers is the XpertStation WS300. Built on the NVIDIA DGX Station architecture, this is a “desktop supercomputer” designed for local AI model fine-tuning and agent development.
Key Specs:
Silicon: NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip.
Memory: 748GB coherent memory.
Bandwidth: A staggering 7.1TB/s HBM3e.
Networking: Dual 400GbE via ConnectX-8.
Modular Scaling with DC-MHS
MSI is leaning heavily into the Data Center-Modular Hardware System (DC-MHS). This modular approach allows for easier transitions between compute environments. The lineup includes:
AMD-Based: 2OU 4-node systems optimized for inference.
Intel-Based: 2U systems featuring Intel Xeon 6/6+ with up to 288 E-cores, targeting high-density containerized deployments and cloud infrastructure.
BTR Take: The Infrastructure Behind the AI
While most consumers focus on the end-user AI applications, COMPUTEX 2026 shows that the real innovation is happening in the “pipes” and “power.” MSI’s transition to in-house server development is paying off with highly specialized, liquid-cooled hardware that can actually sustain the wattages demanded by NVIDIA’s Blackwell generation.
For the BTR audience, this hardware represents the backbone of the services and local development tools we’ll be using for years to come.
Dual-Socket Intel Xeon 6 M-FLW HPMs: D5062
Single-Socket Intel Xeon 6/6+ M-DNO Type-2 HPMs: D3071 / D3061
Single-Socket Intel Xeon 6 M-DNO Type-4 HPMs: D3066
Single-Socket AMD EPYC 9005 M-DNO Type-2 HPMs: D4051
Single-Socket AMD EPYC 9005 M-DNO Type-4 HPMs: D4056
Standard Intel Xeon 6 Enterprise Motherboard: D3060
Standard AMD EPYC 9005 Enterprise Motherboard: D4050
Stay tuned to BabelTechReviews for more live coverage from the COMPUTEX 2026 floor.