EIA Rack
AI Liquid-Cooled Rack
2025SC
2025SC
AI CLUSTER POWER
Cool Fast. Scale Faster.
At SC25, MiTAC Computing looks ahead to the future of computing—showcasing its mastery in end-to-end system engineering, transitioning from individual servers to fully integrated racks. From AI and HPC racks to clusters, MiTAC demonstrates seamless integration of direct-to-chip liquid cooling and high-bandwidth fabric, enabling the performance, scalability, and efficiency that will drive the next generation of datacenter breakthroughs.
From AI Servers to Clusters
Debuting at SC25, the MiTAC x AMD liquid-cooled AI cluster redefines efficiency and scale—unlocking the full power of the AMD Instinct™ MI355X through advanced liquid cooling technology, and paving the way for the next generation of large-scale AI deployments.
Exceptional Cooling Performance
Advanced liquid cooling guarantees non-throttled AI throughput
Ultimate Scalability
Modular scalability from 64 to 256 GPUs
High-Efficiency Network Architecture
400/800 Gb/s networking for ultra-low latency AI efficiency
MiTAC Rack-Scale Manufacturing Power
13+ MW (2025)
Global Manufacturing Centers
3,000 /month
Rack Capacity
Up to 800 Gb/s
Network Speed Environment
Up to 480 VAC
Power Readiness
- Global Footprint Including North America, EMEA, SEA, and China.
- Powered by Local Built on local expertise, localized production, and readily available inventory for greater agility and responsiveness.
- Full Vertical Integration In-house design from PCB, motherboard, and barebone to system and cluster.
- Dynamic Customization Customizable rack configurations tailored to customers’ needs.
Versatile Rack Integration Services
MiTAC provides a full spectrum of rack integration services, from AI liquid-cooled solutions to general air-cooled systems. Our expertise extends to both industry-standard EIA racks and Open Rack v3 architectures, enabling versatile deployments for diverse data center needs
EIA Rack
AI Air-Cooled Rack
G8825Z5
ORv3 Rack
HPC Liquid-Cooled Rack
C2811Z5
ORv3 Rack
HPC Air-Cooled Rack
C2810Z5
Meet Our Heroes
Featured are select MiTAC Computing products that epitomize the crop of the crème in AI inference, high-performance computing, cloud infrastructure, and enterprise workloads.
Cloud Infrastructure
M2810Z5
2U4N Single-socket platform, optimized for high-density cloud deployments and enhanced core computing performance
Learn More- 2U4N architecture maximizes single-socket density for cloud deployments
- Supports AMD EPYC™ 9005/9004 CPUs with up to 3TB DDR5 RAM
- PCIe Gen 5 NVMe plus HH/HL PCIe slots ensure ultra-fast data access
Enterprise Workloads
R2520G6
2U2S server platform with maximum memory capacity for high-performance in-memory computing
Learn More- High-density DIMM slots with maximum memory
- Multiple PCIe slots with I/O versatility
- Advanced manageability with DC-SCM v2.0
AI Training & Inference
G4826Z5
4U2S liquid cooling server platform supporting 8 AMD Instinct™ MI355X GPUs for large-scale AI infrastructure
Learn More- Liquid cooling for peak efficiency
- 8-GPU platform with up to 6TB DDR5-6400 memory
- OCm™ open software stack
AI Training & Inference
G8825Z5
8U AMD Instinct™ MI350X GPU server platform designed for advanced large-scale AI inference and training
Learn More- 8-GPU platform for extreme scalability
- Up to 6TB DDR5-6400 memory for peak AI/HPC performance
- ROCm™ open software stack
High-performance Computing
G4520G6
4U2S 8-GPU server platform designed for various parallel AI workloads
Learn More- Dual CPUs, massive memory capacity up to 8TB
- 8 GPUs, 11 expansion slots for AI acceleration
- 9.6kW Titanium-grade redundant power
High-performance Computing
C2811Z5
Liquid-cooled multi-node server built for performance-critical workloads
- 3TB DDR5-6400 RAM for extreme performance
- PCIe Gen 5 NVMe and OCP v3 slots enable ultra-fast data access
- Liquid-cooled ORv3 multi-node server ensures superior efficiency and density
Success Cases and Solutions
Data Center Rack Integration Service: Partnering with a global cloud-based network security provider
MiTAC provides customized, ready-to-deploy rack integration services that enhance operational efficiency and reduce time-to-market, setting a new standard for fast and sustainable data center deployment.
- Delivered reliable rack deployment worldwide within 48 hours
- Supported over 380 rack configurations across approximately 150 data centers globally
- Enabled rapid GPU infrastructure integration within a 6-month timeframe
Rethinking Sustainable High-Performance Computing in Europe
Qarnot (French cloud service provider) partners with MiTAC to integrate our Capri OCP servers with Qarnot's direct water-cooling and heat recovery system, enabling sustainable HPC for engineering and finance industries while drastically reducing energy waste.
- Pioneered sustainable HPC by capturing 95% of server heat for local reuse
- Reduced operational costs by 50% compared to traditional data centers
- Achieved Power Usage Effectiveness of 1.01 (36% below global average) and Energy Reuse Effectiveness of 0.05 (96.8% below global average)
MiTAC Eliminates Bottlenecks to Maximize GPU Performance with Software-Defined Storage Partners
MiTAC partners with software-defined storage leader to deliver optimized server architectures with consistent throughput and low-latency performance for AI.
- Eliminated I/O bottlenecks via pre-validated design, improving GPU utilization by 35%
- Achieved 200+ GB/s throughput and microsecond latency, cutting AI training time by three times
- Delivered a turnkey solution with deployment in days, backed by single-point accountability
OCP Data Center Deployment: Partnering with a global IT solution provider in Japan & USA
MiTAC offers Open Rack V3-compliant solutions and servers for high-performance, energy-efficient, and sustainable next-generation data centers. The achievements included:
- Adopting a power system compliant with ORv3 to minimize energy loss compared to traditional designs, optimizing overall energy utilization
- Faster rollouts with flexible rack configuration capabilities and reduced resource demands
