Press Releases

As AI advances toward autonomous, multi-step agentic applications, infrastructure must support persistent workloads that continuously process data, coordinate models and tools, maintain context, and execute complex workflows. MiTAC Computing addresses these demands through flexible processors, accelerators, cooling, and system architectures, enabling organizations to deploy scalable AI infrastructure aligned with their workloads, data center environments, and business objectives.

“Agentic AI is accelerating the need for infrastructure that can support larger, more dynamic, and increasingly interconnected workloads,” said Raymond Huang, General Manager of MiTAC Computing Technology Corporation. “There is no single architecture or cooling approach that fits every AI deployment. At AI Infra Summit 2026, MiTAC Computing will demonstrate how our diversified platforms and end-to-end integration capabilities help customers build scalable, energy-efficient, and production-ready infrastructure for the agentic AI era.”

Diversified Rack-Scale Infrastructure for Agentic AI

Agentic AI places intensive and interconnected demands across the entire infrastructure stack. In addition to accelerated computing for model execution, production environments require high-speed access to enterprise data, low-latency networking, scalable storage, efficient thermal management, and the ability to operate persistent AI services reliably.

At AI Infra Summit 2026, MiTAC Computing will highlight four rack-scale reference solutions designed to help customers address these requirements across AI, HPC, and cloud data center environments:

52U High-Density AI Liquid-Cooled Rack

This 52U HD rack integrates 12 MiTAC G4826Z5 liquid-cooled AI servers, supporting up to 96 AMD Instinct™ MI355X GPUs and dual AMD EPYC™ processors per server. Delivering 50% greater GPU density than a standard 48U configuration, the solution maximizes vertical compute capacity while minimizing data center footprint. Advanced cold-plate cooling and an integrated coolant distribution unit (CDU) reduce reliance on server fans and facility air conditioning, helping improve energy efficiency and optimize power usage effectiveness (PUE).

G8825Z5 AI GPU Air-Cooled Rack

Designed for high-performance AI training and inference, this full-rack configuration integrates four MiTAC G8825Z5 systems with up to 32 AMD Instinct™ MI350X GPUs. The solution delivers dense GPU-accelerated compute within an air-cooled architecture, providing a practical deployment option for data centers seeking to scale AI performance without transitioning to liquid-cooling infrastructure.

OCP ORv3 Standard Liquid-Cooled Rack

Built around the MiTAC C2811Z5 system, this liquid-cooled rack complies with OCP Open Rack v3 (ORv3) standards and is designed for standardized data centers, hyperscale cloud environments, and HPC applications. Modular power, networking, and thermal architectures provide greater deployment flexibility, simplify system servicing, and help reduce ongoing maintenance costs.

Advanced Server Platforms for AI, HPC, Cloud, and Enterprise Workloads

M2810Z6: High-Density Multi-Node Computing

The M2810Z6 is a high-density 2U, 2-node server built on a single-socket AMD SP7 architecture per node. Each independently serviceable node supports one 6th Gen AMD EPYC™ Server CPU with up to 600W default PPT, 16 DDR5 DIMM slots, and up to 4TB of DDR5-8000 memory.

Flexible front hot-swappable NVMe storage supports E3.S, E1.S, and U.2 form factors, while OCP 3.0 TSFF expansion provides additional configuration flexibility. By combining two nodes in a compact chassis, the M2810Z6 delivers greater compute density for CPU-intensive AI services, HPC, data processing, orchestration, and cloud-scale workloads.

G8825Z5: World’s First Diamond-Cooled AI Server

The G8825Z5 is a high-performance AI server powered by AMD Instinct™ MI350X GPUs and the AMD Pensando™ Pollara 400 AI NIC. Featuring Akash Systems’ proprietary Diamond Cooling® technology, the platform is engineered to sustain throttle-free performance at inlet temperatures above 95°F while delivering up to 50% more AI tokens per watt under standard operating conditions.

By improving heat dissipation and reducing cooling requirements, the G8825Z5 helps optimize overall system efficiency while lowering capital and operating costs. Its combination of GPU acceleration, high-speed connectivity, and advanced thermal management provides an energy-efficient platform for large-scale AI training and inference in next-generation data centers.

TN85-B8261: Flexible AMD and NVIDIA GPU-Accelerated AI Server

The TN85-B8261 is a versatile 2U dual-socket server designed for AI training, inference, HPC, and data analytics. Powered by dual AMD EPYC™ 9005 or 9004 Series processors, it supports up to 6TB of DDR5 memory and four double-width PCIe 5.0 GPUs, with flexible accelerator options including AMD Instinct™ MI350P PCIe® Cards and NVIDIA RTX PRO™ 4500 Blackwell Server Edition

or NVIDIA RTX PRO™ 6000 Blackwell Server Edition .

Support for the AMD Pensando™ Pollara 400 AI NIC, eight hot-swappable NVMe U.2 drive bays, and Titanium-efficiency redundant power supplies delivers high-speed connectivity, local storage performance, and operational reliability. This balanced architecture provides the compute density and scalability required for model training, inference, RAG, and multi-step agentic AI workflows.

G4827Z6: Next-gen G Series High-throughput 4U AI Powerhouse 

Based on the NVIDIA MGX reference architecture, this 4U, 2-socket server features the latest dual AMD EPYC^TM “Venice” processors and supports up to 8 double-width GPUs. Designed for deployment flexibility, this platform can be equipped with NVIDIA RTX PRO 4500 Blackwell Server Edition or NVIDIA RTX PRO 6000 Blackwell Server Edition as well as NVIDIA H200 GPU. Purpose-built for large-scale Generative AI training and inference, the system delivers the massive throughput, low latency, and scalability required for modern AI factories and enterprise AI deployments.

G4520G6: Multi-modal Generative AI for Fine-tuned Training & Inference

Powered by Intel® Xeon® 6 processors, the G4520G6 AI PCIe GPU server supports eight dual-slot GPUs and up to 8,192GB of DDR5-6400 memory. PCIe 5.0 connectivity provides high-bandwidth communication across processors, accelerators, storage, and networking resources.

Advanced thermal management and a 9,600W Titanium-efficiency redundant power supply enables scalable performance and power efficiency under demanding workloads. Its flexible PCIe GPU architecture supports generative and agentic AI, analytics, HPC, cloud computing, and enterprise applications across diverse accelerator and software configurations.

R1917GC: High-Efficiency NVIDIA Grace and Vera-Powered Server for Core, Edge, and Management

Built on the NVIDIA MGX™ modular architecture, the R1917GC delivers high-density, energy-efficient compute designed to anchor modern AI infrastructure across core data center, edge compute, and unified management layers. Powered by NVIDIA Grace™ or next-generation NVIDIA Vera CPUs, the 1U platform harnesses onboard LPDDR5X memory to provide exceptional memory bandwidth and compute density while operating within strict power envelopes. Designed for versatile deployment, the R1917GC functions seamlessly as an always-on cluster orchestrator managing Kubernetes control planes and Slurm dispatching, a high-speed data mover and storage head node feeding GPU clusters via RDMA, or an efficient edge engine running local inference and microservices.

C2811Z5: Multi-Node OCP Liquid-Cooled Server

Built on AMD EPYC™ 9005 Series processors, the C2811Z5 is a multi-node, OCP-based server designed for energy-efficient, sustained high-performance computing. Each node supports up to 12 DDR5-6400 memory slots, NVMe E1.S interfaces, high-performance NVMe SSDs, and direct liquid cooling.

The combination of multi-node density and liquid-cooled thermal management makes the C2811Z5 well suited for advanced HPC and AI clusters. It is purpose-built for compute-intensive applications such as scientific simulation, engineering design, climate research, and weather modeling.

About MiTAC Computing Technology Corporation

MiTAC Computing Technology Corporation, a subsidiary of MiTAC Holdings Corporation (TWSE: 3706), delivers comprehensive, energy-efficient server and infrastructure solutions backed by industry expertise dating to the 1990s. Specializing in AI, HPC, cloud, enterprise, storage, and edge computing, MiTAC Computing provides solutions ranging from server barebones and complete systems to integrated racks and clusters.

With global research and development, manufacturing, system integration, validation, and support capabilities, MiTAC Computing delivers scalable and customized platforms designed to meet diverse data center and business requirements. Its portfolio encompasses GPU-accelerated AI servers, multi-node systems, OCP platforms, enterprise servers, high-density storage, rack-scale integration, and advanced liquid-cooling solutions—helping customers prepare their infrastructure for generative and agentic AI.

For more information, visit www.mitaccomputing.com.