MiTAC Computing Showcases Scalable AI Infrastructure for the Agentic Era at AI Infra Summit 2026
MiTAC
Computing Showcases Scalable AI Infrastructure for the Agentic Era at AI Infra
Summit 2026
Rack-scale liquid cooling, GPU-accelerated
systems, OCP platforms, and storage-dense servers help data centers support the
evolving demands of training, inference, and agentic AI
SAN JOSE, California — [September 16, 2026] — MiTAC Computing Technology Corporation, a global leader in
high-performance, energy-efficient server solutions and a subsidiary of MiTAC
Holdings Corporation (TWSE: 3706), today announced that it will showcase its
latest diversified AI infrastructure and solutions at AI Infra Summit 2026,
taking place September 15–17, 2026, at the Santa Clara Convention Center in
Santa Clara, California. Visitors can explore MiTAC Computing’s solutions
at Booth 1146.
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.
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 EPYCTM “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 GPUs. Purpose-built for
large-scale Generative and agentic 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 reference 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 CPUs 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.
