MiTAC Unveils 52U Liquid-Cooled AI Rack, Integrates 96 AMD Instinct MI355X GPUs
MiTAC Computing Technology Corp. unveiled advanced rack-scale architectures at COMPUTEX 2026, including a 52U liquid-cooled AI rack with 96 AMD Instinct MI355X GPUs. These solutions are designed to address scalability and energy challenges for agentic AI workloads, offering turnkey infrastructure and open-source management for data centers. MiTAC is also collaborating with various industry partners to co-create resilient and intelligent AI infrastructure, including modular data centers.
Key Takeaways
- MiTAC's new 52U high-density AI liquid-cooled rack integrates 12 G4826Z5 AI servers with 96 AMD Instinct MI355X GPUs.
- This configuration achieves 50% higher GPU density per rack compared to standard AI setups, designed for agentic AI applications.
- The C2811Z5 liquid-cooled HPC rack was introduced for national and academic HPC centers.
- MiTAC showcased G8825Z5 Diamond-cooled AI servers, providing throttle-free performance up to 95°F inlet air temperatures.
- New partnerships include DDN for data platforms, Rafay for GPU-as-a-Service, Canonical for Ubuntu 26.04 LTS integration, and Tonomia for modular AI data centers.
Why It Matters
The introduction of extreme-density liquid-cooled AI infrastructure directly responds to the escalating compute and energy demands of agentic AI. Achieving 50% higher GPU density per rack signifies a critical step in overcoming data center constraints related to space, cooling, and power. This strategic push, amplified by extensive hardware and software partnerships, points to a concerted industry effort to standardize and scale AI deployments efficiently. Streaming providers and media companies increasingly rely on AI for content creation, optimization, and delivery, making these infrastructure advancements directly relevant to their operational capabilities. Watch for how quickly these high-density, modular AI solutions achieve broad market adoption and impact deployment costs for large-scale AI processing.
Read full article at tradingview.com