St. Louis, Missouri – November 18, 2025 – MiTAC Computing Technology, a global leader in high-performance and energy-efficient server solutions, is proud to announce its participation at the Supercomputing (SC) 2025 (November 18–20, St. Louis, Missouri), booth 3916. Under the theme “AI Cluster Power – Cool Fast. Scale Faster.”, MiTAC Computing will showcase its modular, highly scalable rack infrastructure designed for demanding AI and HPC workloads. The exhibition will demonstrate the company's full spectrum of capabilities, ranging from single AI servers to complete cluster-level integration, with a focus on cutting-edge liquid cooling and energy-efficient designs to maximize performance and sustainability.

 

Collaborating with key industry players such as AMD, Broadcom, CoolIT, Intel, KIOXIA, Micron, NVIDIA, Samsung, and Solidigm, MiTAC Computing is dedicated to accelerating the development of advanced computing and high-efficiency data centers.

 

 

Rack-scale Innovation｜From Standard Architecture to AI Cluster Excellence

At SC 2025, MiTAC Computing will present its full range of rack-level solutions, embodying the commitment to cluster-scale deployments. The exhibit includes liquid-cooled and air-cooled AI and HPC racks in both OCP ORv3 and EIA formats, addressing the needs of next-generation open architectures as well as traditional enterprise data centers.

 

AI Liquid-Cooled Rack with AMD Instinct™ MI355X GPUs | Optimized for Ultra-Scale AI with 64 to 256 GPUs

The MiTAC 48U EIA AI liquid-cooled rack MR1100 series is introduced for the hyperscale data center market, specifically targeting the exponential demands of large-scale AI training and Generative AI inference. This high-density solution is engineered for unparalleled AI performance and is ready for future Exascale-class deployment, supporting 64 to 256 AI GPUs within a single, industry-standard footprint. Debuting at SC 2025, the MiTAC x AMD AI cluster features the AI liquid-cooled platform, utilizing the latest AMD Instinct™ MI355X GPUs and high-performing AMD EPYC™ 9005 CPUs. This robust design uses advanced cold-plate technology for direct-to-chip liquid cooling, ensuring exceptional, non-throttled AI throughput during intensive LLM training. Its high-efficiency network architecture (400/800 Gb/s) ensures ultra-low latency, and the standardized 48U EIA design allows for seamless integration into hyperscale data centers globally.

 

AI Air-Cooled Rack with AMD Instinct™ MI350X GPUs｜Standardized Architecture with High-Speed Interconnect for Rapid AI Deployment

MiTAC Computing will also present its standard EIA 45U air-cooled AI rack, pre-configured with four MiTAC G8825Z5 systems, leveraging AMD Instinct™ MI350X/MI325X GPUs. This robust setup delivers significant compute power, optimized for large language model (LLM) training and generative AI inference tasks. The 800Gb/s network switch on the rack is powered by the Broadcom Tomahawk 5 chipset, which guarantees reliable, low-latency data transmission among compute nodes. Furthermore, the system is complemented by the MiTAC GC68C-B8056 management server and MiTAC TS70A-B8056 storage server, facilitating the cohesive integration of high-performance computation with rapid data access. By unifying compute, networking, management, and storage, MiTAC enables enterprises to rapidly deploy compatible, highly scalable, and available AI/HPC cluster environments.

 

OCP ORv3 Liquid-Cooled Rack｜Modular Power and Advanced Thermal Management for Sustainable HPC

MiTAC Computing’s OCP ORv3 43OU liquid-cooled rack is designed for sustainable HPC, accommodating up to 14 C2811Z5 multi-node servers. Powered by AMD EPYC™ 9005 series processors and expansive DDR5 memory, the rack integrates MiTAC Lake Erie storage modules and networking switches for seamless interaction. Power is managed by a 33kW Power Shelf, while cooling is handled by the CoolIT 200kW CHx200+ In-Rack CDU, delivering a 200kW-class capacity. This open, modular design ensures stable, energy-efficient operation, maximizing performance and reliability for high-density compute servers.

 

AI Acceleration Platforms: Liquid-Cooled GPU Solutions for Massive Training & Generative AI

• G4527G6: This 4U AI Accelerator stands as a powerhouse built on the NVIDIA MGX™ architecture. Driven by dual Intel® Xeon® 6767P CPUs, the G4527G6 is engineered for massive parallel workloads, hosting up to 8 NVIDIA RTX PRO™ 6000 Blackwell Server Edition GPUs. With fast Solidigm D7-P5520 SSDs, it provides the I/O speed and density essential for cutting-edge deep learning, demanding computer vision, and expansive enterprise AI applications.
• G4826Z5: The G4826Z5 is the foundational, liquid-cooled compute node specifically engineered to fit within the MiTAC AI liquid cluster. This node ensures sustained performance by incorporating liquid cooling for both CPU and GPU, supporting up to eight AMD Instinct™ MI355X GPUs alongside AMD EPYC™ 9005 series processors. Boasting an immense 6TB capacity across 24 DDR5-6400 memory slots and equipped with the Broadcom P2200G network adapter, the G4826Z5 is purpose-built to deliver the sustained high-throughput and minimal latency critical for massive AI model training.
• G8825Z5: As a premier high-performance AI platform, the G8825Z5 handles the most demanding computational requirements with ease. Its capabilities are defined by its support for up to 8 AMD Instinct™ MI325X or MI350X GPUs, and AMD Pensando™ Pollara 400 AI NICs. Furthermore, the integration of Micron 6550 ION NVMe SSDs ensures instantaneous data access and processing speeds, which are vital for drastically accelerating large-scale model training cycles and maximizing generative AI inference rates.

HPC & Cloud Computing Frameworks: Optimized Platforms for Scalable Workloads

• G4520G6: It is a highly adaptable server platform, excelling in both flexible cloud infrastructure and traditional HPC roles. Powered by the latest Intel® Xeon® 6 CPUs, this server delivers a significant leap in performance per watt for high-density, scalable workloads. Engineered with ample expansion options, G4526G6 includes support for NVIDIA RTX PRO^TM 6000 Blackwell Server Edition and NVIDIA Hopper GPUs. Furthermore, it leverages Micron DDR5 DRAM to deliver superior memory performance, resulting in efficient and accelerated processing within scalable, data-heavy cloud and distributed environments.
• C2811Z5: This OCP-compliant, liquid-cooled multi-node server is the ideal solution for high-density compute environments. Featuring AMD EPYC™ 9005 series processors, the C2811Z5 provides robust stability for lengthy, complex computational workloads, supported by 12 DDR5-6400 memory slots (3TB capacity per node). Optimized with NVMe E1.S storage interfaces and Micron 9550 NVMe SSDs, it guarantees the sustained, high-speed data delivery essential for intensive HPC applications, such as detailed scientific simulations and meteorological modeling.

Enterprise Data Solutions: High-Reliability Storage and Processing for Data-Intensive Applications

• M2810Z5: The M2810Z5 is a specialized enterprise server optimized to deliver peak storage performance. By deploying Kioxia XD8 E1.S Gen5 SSDs, this platform unlocks maximum bandwidth and near-instantaneous data retrieval, making it indispensable for enterprises running the most rigorous, I/O-intensive database and transactional applications.
• M2510G6: Built to provide a reliable framework for demanding data processing, the M2510G6 platform ensures high-endurance and unwavering stability in mission-critical enterprise settings, a reliability driven by its choice of Samsung MZTL67T6HBLC-00AW7 SSDs.
• R2513G6: Tailored for organizations with immense archival needs, the R2513G6 platform offers maximum data retention capacity. It is designed to accommodate vast storage volumes, featuring integration with large-capacity Seagate EXOS M 30TB HDDs for dependable, petabyte-scale storage and long-term data archiving.
• R2520G6: The R2520G6 is a high-capacity 2U server ready for intensive data processing. Supporting dual Intel® Xeon® 6700P series processors and up to 24 NVMe U.2 SSDs, it integrates Solidigm D7-PS1010 SSDs with PCIe 5.0 bandwidth to guarantee exceptional long-term stability. This makes the R2520G6 an optimal server for massive data warehousing, executing complex big data analytics, and providing the robust foundation for enterprise AI data preparation.

 

Global Success Stories: Demonstrating MiTAC’s Value in AI & HPC Deployments Worldwide

At SC 2025, MiTAC Computing will showcase real-world success cases from across the globe, highlighting its proven expertise in delivering comprehensive rack-level solutions. These deployments reflect MiTAC’s unwavering commitment to cluster-scale integration and innovation in AI and HPC infrastructure.

• MiTAC and French CSP pioneer sustainable HPC, achieving world-class PUE and operational cost reduction: MiTAC partnered with Qarnot by integrating Capri OCP servers in their direct water-cooling and heat recovery system, featured alongside the Broadcom N1400GD network adapter at SC25. This system captures 95% of server heat for local reuse, achieving a world-class PUE of 1.01, and reducing operational costs by 50% for clients.
• OCP data center deployment in the U.S. in partnership with CTCA: MiTAC partnered with CTCA, a global IT solutions provider, to accelerate OCP deployment in data centers across the U.S.. This success leveraged the ORv3-compliant OCP server, featuring a 48V power system for minimized energy loss. At SC, we showcase the OCP server featuring Samsung M321R8GA0EB2-CCP Memory, highlighting faster rollouts and resource efficiency.
• Data center rack integration service partnering with a global cloud-based network security service provider: For a leading global cloud-based network security service provider, MiTAC provided customized, ready-to-deploy rack integration services, utilizing platforms such as the MiTAC GC68C-B8056 server. This comprehensive service dramatically streamlined operations—delivering over 380 configurations across 150 data centers worldwide, meeting 48-hour turnaround times, and enabling a GPU-first infrastructure shift within half a year. MiTAC provides the consistency, agility, and innovation customers need to stay ahead.
• MiTAC and its software-defined storage partner maximize GPU performance by eliminating AI training bottlenecks: Through collaboration with a leading software-defined storage partner, MiTAC leveraged the GC68A-B8056 server to remove critical I/O constraints in large-scale AI training. The pre-validated architecture boosted GPU utilization by 35% and delivered over 200 GB/s throughput with microsecond latency, effectively tripling the speed of AI training completion through a unified, turnkey solution.
• For more SC information and product catalogs, please visit:

MiTAC Computing SC 2025 event site

Intel Platform Brochure 

AMD Platform Brochure