ASUS Unveils Ascent GX10 AI Supercomputer and Full-Stack Infrastructure at Ai4 2026

At Ai4 2026 in Las Vegas, ASUS unveiled a comprehensive AI infrastructure lineup spanning enterprise rack servers to local deskside supercomputing. The showcase was headlined by the Ascent GX10 AI Supercomputer—powered by NVIDIA DGX Spark architectures—and the ExpertCenter Pro ET900N G3, allowing enterprise development teams to run local 70B+ parameter model training and RAG inference workflows without cloud dependency.

Deskside AI Supercomputing for Enterprise Production Workflows

As enterprise engineering teams transition from AI experimentation to local model fine-tuning and retrieval-augmented generation (RAG), the demand for quiet, high-density deskside compute has surged. The ASUS Ascent GX10 integrates multi-GPU NVIDIA MGX and DGX Spark architectures into a compact office form factor, featuring quiet liquid cooling and liquid-to-air heat exchangers suited for standard 120V/240V office power.

Key Hardware & System Specifications

• NVIDIA DGX Spark Architecture: Integrates unified high-bandwidth memory for running 70B to 120B parameter LLMs locally.
• ExpertCenter Pro ET900N G3: Deskside AI workstation based on NVIDIA DGX Station for multi-tenant developer workflows.
• Full-Stack Deployment: Combines local training workstations with MGX data center server racks for seamless model migration.
• Enhanced Security & Thermal Efficiency: Built-in hardware security modules and quiet liquid-assisted cooling for office deployment.

Bridging Desktop Development and Data Center Scale

ASUS Vice President of Enterprise Solutions highlighted that offering matching hardware architecture from desktop to rack allows developers to prototype AI agents locally before deploying them directly to cloud data centers. The company also showcased NVIDIA DC-MHS open-standard rack servers for scaled enterprise IT infrastructure.

System Specification Summary