神達雲端運算在OCP大會展示全棧AI基礎設施，與ZutaCore合作無水兩相液冷PoC，僅為產品展示無訂單

MiTAC Computing Technology featured a comprehensive suite of full-stack AI infrastructure at the Open Compute Project (OCP) Global Summit in San Jose, California, centered around the theme "Scaling Open AI Infra with Full-Stack Innovation." The company demonstrated integrated solutions spanning compute, storage, liquid-cooled servers, firmware, operating systems, and management software.

MiTAC noted that AI workloads have evolved beyond model training, inference, and retrieval-augmented generation (RAG) toward agentic AI. As enterprise infrastructure requirements vary significantly by compute scale and deployment location, MiTAC's AI infrastructure stack consolidates previously siloed hardware and software layers into a deployable, manageable, and scalable ecosystem.
To address large-scale centralized AI environments, MiTAC leverages its technology stack and rack-level system integration to build scalable AI factories capable of supporting high-density compute and continuous horizontal expansion. For distributed edge sites or projects requiring rapid bring-up, the company introduced modular data center options. The featured TonoForge Modular Data Center, developed in partnership with Tonomia, can be deployed in just 12 weeks, according to MiTAC, underscoring the speed of modular architectures in bringing high-performance GPU AI clusters online.
Advanced liquid cooling represents another core pillar of MiTAC's lineup. In a joint proof-of-concept (PoC) demonstration with ZutaCore, MiTAC integrated its G4826Z5 server with ZutaCore's HyperCool waterless two-phase direct-to-chip liquid cooling technology. The system uses dielectric fluid to undergo a phase change directly over chip hot spots, efficiently pulling heat away from high-power CPUs and GPUs.
Engineered for next-generation silicon, the technology is designed to handle thermal loads exceeding 4,000W per chip with an average heat flux over 250W/cm2. MiTAC said annualized modeling under specific operating conditions indicates partial power usage effectiveness (pPUE) could be as low as 1.03. During the Expo Hall Session on October 14, MiTAC and its ecosystem partners will discuss transitioning high-density AI systems to next-generation rack-scale platforms and overcoming thermal limits via advanced liquid cooling.
Advancing its open-architecture initiative, MiTAC continues to expand its open-source firmware portfolio. Its in-house MiOPF and MiOBMC solutions enhance system boot speeds, hardware telemetry, and remote monitoring capabilities while minimizing vendor lock-in for customers.
The software ecosystem incorporates Canonical-certified Ubuntu 26.04, Kubernetes, DDN's intelligent AI data platform, and the AMD ROCm open software stack. MiTAC is also collaborating with Rafay to integrate GPU cluster management, enabling pooled GPU-as-a-Service (GPUaaS) resource models.
During the summit's Executive Sessions, MiTAC highlighted how open standards like open platform firmware (OPF), OpenBMC, and Redfish foster interoperability across compute, networking, power, and liquid cooling infrastructure. Paired with its MiCoreView software, MiTAC extends monitoring and orchestration from individual nodes up to the rack and cluster levels.
The event also marked the global debut of MiTAC's C2611Z6 server platform and showcased its latest G4520G6E3BC-N configuration, engineered for OCP high-density compute and enterprise AI inference workloads, respectively. By integrating processor architectures across AMD, Intel, and Nvidia, the new platforms broaden MiTAC's portfolio from general high-performance computing (HPC) to specialized GPU acceleration.
MiTAC emphasized that competition in AI infrastructure has shifted from raw single-server benchmarks toward full-stack integration capabilities encompassing hardware, liquid cooling, firmware, and orchestration software. As AI factories expand in scale, open standards and modular deployment will remain essential for reducing construction timelines, avoiding supplier lock-in, and ensuring long-term scalability.