ASUS AI POD with NVIDIA Blackwell Ultra GPUs + other servers and offerings revealed at GTC 2025

ASUS has officially unveiled the ASUS AI POD, built on the powerful NVIDIA GB300 NVL72 solution. The NVIDIA GB300 NVL72 server platform pushes performance beyond the standard Blackwell architecture by enhancing compute power, memory capacity, and networking. Equipped with 72 NVIDIA Blackwell Ultra GPUs and 36 Grace CPUs, it delivers 40TB of high-speed memory per rack. The system integrates NVIDIA Quantum-X800 InfiniBand and Spectrum-X Ethernet, along with SXM7 and SOCAMM modules for better serviceability, all within a fully liquid-cooled design. This robust setup is purpose-built to handle trillion-parameter large language model (LLM) training and inferencing effortlessly. Complementing the NVIDIA GB200 NVL72 infrastructure, ASUS also showcased the S501A-E12-RS12U software-defined storage server. This solution is designed to reduce latency during data training and inference while boosting compute efficiency, enabling clients to build high-performance AI infrastructure using ASUS's in-house solutions.

GPU servers designed for generative AI workloads

ASUS is also spotlighting its range of NVIDIA-certified servers optimized for generative AI. Among them, the 10U ASUS ESC NB8-E11 featuring NVIDIA Blackwell HGX B200 8-GPU and the ASUS XA NB3I-E12 with HGX B300 NVL16 stand out. The latter delivers enhanced AI FLOPS, 0.3TB of HBM3e memory, and advanced networking through NVIDIA Quantum-X800 InfiniBand and Spectrum-X Ethernet—making it a perfect fit for AI reasoning, agentic AI, and video inference tasks. Additionally, the 7U ASUS ESC N8-E11V packs eight NVIDIA H200 GPUs within a dual-socket configuration, supporting both air and liquid cooling for superior thermal efficiency, scalability, and AI performance.

Scalable servers optimized for AI inferencing

For AI inferencing, ASUS is showcasing the ESC8000 series, designed with NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. The high-density 4U ESC8000-E12P supports eight dual-slot NVIDIA H200 GPUs and fully embraces the NVIDIA MGX architecture for effortless scalability and fast deployment. Similarly, the ESC8000A-E13P MGX server is engineered for seamless integration, performance optimization, and scalability—ideal for modern data centers and fast-evolving IT environments.

ASUS Ascent GX10 – compact AI powerhouse

Beyond large server solutions, ASUS introduced the Ascent GX10, a portable AI powerhouse built with the NVIDIA GB10 Grace Blackwell Superchip, featuring a 20-core Arm CPU and Blackwell GPU. It comes equipped with 128GB of RAM, delivering 1,000 AI TOPS performance and supporting LLMs with up to 200 billion parameters. Also on display is the IoT PE2100N, powered by the NVIDIA Jetson AGX Orin module with up to 275 TOPS performance. It's designed for diverse applications such as generative AI, VLM, and LLM workloads across smart cities, robotics, and in-vehicle systems. Meanwhile, the rugged ASUS IoT PE8000G edge AI GPU computer supports dual 450W NVIDIA RTX GPUs, flexible power input, and dependable operation in extreme environments—perfect for computer vision, autonomous driving, and real-time AI inference tasks.

Building efficient, sustainable AI infrastructure

ASUS pairs cutting-edge performance with a focus on sustainability, integrating energy-efficient power supplies, advanced cooling solutions, and cloud services into its L12-ready infrastructure. Tools like the ASUS Infrastructure Deployment Center (AIDC) and ASUS Control Center (ACC) streamline remote management and large-scale deployments. Full support for NVIDIA AI Enterprise and Omniverse ensures smooth operation, while the comprehensive design reduces power consumption, lowers operational costs, and minimizes environmental impact and total cost of ownership (TCO).