MiTAC Computing showcases liquid-cooled AI servers at Shanghai conference

MiTAC Computing Technology Corporation, a subsidiary of Taiwan-listed MiTAC Holdings Corporation (TWSE:3706), unveiled a comprehensive AI server portfolio at the World Artificial Intelligence Conference (WAIC) 2026 held in Shanghai on July 17, 2026. The presentation took place under the core theme "Liquid Cooling. Infinite Compute. Sustainable Future" and focused on addressing the enterprise shift from generative AI to agentic AI—a class of AI systems capable of autonomous planning and task execution.

The company introduced four primary rack-scale solutions, each engineered for distinct use cases. The flagship offering is a 52U high-density liquid-cooled rack that integrates 12 G4826Z5 liquid-cooled AI servers, collectively packing up to 96 AMD Instinct MI355X GPUs along with dual AMD EPYC processors. This configuration achieves a 50% boost in GPU density compared to standard 48U racks and employs advanced cold plates and a CDU (coolant distribution unit) to reduce power consumption from fans and air conditioning, thereby optimizing the facility's Power Usage Effectiveness (PUE). A second offering, the G8825Z5 AI GPU air-cooled rack, is tailored for AI training and inference workloads and houses 4 G8825Z5 units configured with up to 32 AMD Instinct MI350X GPUs. For standardized internet data centers and hyperscale public cloud deployments, MiTAC offers the C2811Z5 system, which complies with OCP ORv3 standards and features modularized power, networking, and thermal designs to maximize deployment flexibility and reduce maintenance costs. The fourth solution, the TS70A-B8056 high-density storage rack, integrates the DDN Infinia Intelligent AI Data Platform to eliminate end-to-end data bottlenecks and accelerate model inference, RAG (retrieval-augmented generation), and agentic AI workflows.

Beyond the rack-scale portfolio, MiTAC Computing highlighted a broader server platform lineup engineered for GPU acceleration, multi-node OCP (Open Compute Project) architectures, and high-density data infrastructure. The company positioned these solutions as essential building blocks for next-generation data centers capable of supporting the increased complexity and density demands of agentic AI workloads. All solutions were on display at booth H2-C1327 during the conference.

MiTAC Computing's presentation at WAIC 2026 reflects a strategic shift in how enterprises are preparing their data center infrastructure. The company's emphasis on transitioning from generative AI to agentic AI highlights a broader industry recognition that AI workloads are becoming more complex and autonomous. Agentic AI systems, which can plan and execute tasks independently, demand higher compute density, faster inference, and more sophisticated data pipelines than traditional generative AI models.

The technical choices MiTAC showcases—liquid cooling, high-density GPU packing, modularized designs, and integrated storage solutions—directly address these demands. The 50% GPU density improvement in the 52U rack, achieved through liquid cooling, represents a tangible way to reduce both the physical footprint and operational power costs of large AI deployments. The inclusion of DDN's Infinia platform alongside storage infrastructure signals recognition that data bottlenecks are as critical as compute capacity in agentic workflows, where rapid model inference and retrieval-augmented generation (RAG) depend on fast data access.

No comments yet. Be the first to share your thoughts!