MiTAC Computing Unveils Full-Stack Innovations for AI Infrastructure at OCP Global Summit 2026

At the recent OCP Global Summit 2026 held in San Jose, California, MiTAC Computing Technology Corporation showcased its latest advancements under the banner "Scaling Open AI Infra with Full-Stack Innovation." As a pivotal subsidiary of MiTAC Holdings Corporation (TWSE: 3706), MiTAC is actively leading the charge in providing high-performance server solutions tailored for the demanding needs of open AI infrastructure.

Transforming AI Cluster Management

As the world increasingly gravitates towards large-scale AI clusters, the requirements extend beyond mere computational power. MiTAC's full-stack approach transforms fragmented infrastructures into a cohesive, scalable foundation that supports deployment, management, and expansion seamlessly. Recognizing the varied infrastructural requirements that come with different AI workloads, MiTAC has developed a robust AI infrastructure stack. This stack includes powerful computing resources, liquid-cooled servers, open platform firmware, operating systems, and a comprehensive ecosystem for AI software management.

The scalable AI factory, a centerpiece of MiTAC's offerings, combines full-stack capabilities with rack-level integration, fostering high-performance computing and continuous scalability. For sites requiring modular solutions or rapid deployments, MiTAC also presents innovative modular data center solutions, including the TonoForge™, developed in collaboration with Tonomia. This modular data center solution features MiTAC's liquid-cooled AI servers, designed for accelerated deployment of powerful GPU-based AI clusters.

Advanced Thermal Management Solutions

As the demand for higher computational density mounts, effective cooling technology becomes essential. MiTAC is pushing the envelope with advanced liquid cooling solutions. Notably, they introduced a proof-of-concept integrated with ZutaCore, showcasing how their MiTAC G4826Z5 server utilizes ZutaCore® HyperCool® technology with a groundbreaking two-phase liquid cooling method. This approach leverages a dielectric fluid that transitions from liquid to gas directly at the chip level, effectively dissipating heat generated by CPUs and GPUs right at the source. Such technology illustrates its potential to manage over 4,000 watts per chip while achieving an annualized pPUE of just 1.03.

On October 14th, MiTAC will delve into the implications of high-density AI systems transitioning to rack-scale platforms, emphasizing how this advanced cooling can mitigate thermal limitations for ever-increasing compute power needs.

Pushing the Open Standards Agenda

Amidst hardware performance advancements, MiTAC is committed to further developing its open-source firmware solution. The G4826Z5 server exemplifies full-stack integration, from bare metal to sophisticated AI applications. At the platform level, their internally developed MiOPF™ facilitates rapid system initialization, while MiOBMC™ provides real-time hardware monitoring and telemetry. This strategy enhances customer autonomy over firmware configurations and minimizes dependency on single-vendor solutions.

Furthermore, the software ecosystem on the G4826Z5 operates on the Ubuntu 26.04 OS, certified by Canonical, integrating with Kubernetes and the DDN Data Intelligence Platform. They also fully leverage the open software stack AMD ROCm™ to unlock the computing power embedded in AMD Instinct™ MI355X GPUs. In collaboration with Rafay, MiTAC is enhancing GPU cluster management, thus enabling flexible 'GPU-as-a-Service' (GPUaaS) resource pools that can dynamically allocate computing power based on actual AI workload demands, streamlining deployment and operational management.

In an upcoming executive session, MiTAC will also shed light on open infrastructure management, discussing significant technologies, including Open Platform Firmware (OPF), OpenBMC, and Redfish. They will illustrate how open standards and collaboration within the OCP community can improve infrastructure interoperability, alongside demonstrating MiCoreView™, their infrastructure management software built on open-source principles. Practical insights into deployment, telemetry, and rack and cluster management will be shared, revealing how these innovations can evolve data center orchestration more efficiently.

New Platforms for AI Inference

MiTAC also made headlines by unveiling the C2611Z6 server and the latest G4520G6E3BC-N configuration, thereby expanding their high-density computing and AI inference portfolio for OCP environments. The C2611Z6 is designed specifically for OCP-ORv3 racks, equipped with the latest AMD EPYC™ 9006 series processors, housing four compute nodes in a compact 2OU form factor. Each node supports 16 DDR5-8000 RDIMM slots and has OCP 3.0 expansion capabilities, reflecting their commitment to high-density liquid cooling infrastructure.

Meanwhile, the G4520G6 (G4520G6E3BC-N) servers are crafted for enterprise-level AI inference tasks, powered by Intel® Xeon® processors and accommodating up to eight NVIDIA RTX PRO 4500/GPU models. This design merges high-density GPU computing with high-speed storage solutions in a single platform, optimized for data-intensive inference workloads.

MiTAC Computing invites all attendees at the OCP Global Summit to visit Booth D22 to explore how full-stack innovations are pivotal in scaling open AI infrastructures. Discover more about their solutions through their website.