The new ExpertCenter Pro ET900N deskside supercomputer was showcased by ASUS at the Computex 2026 exhibition to address enterprise workloads.
This system delivers up to 20 PFLOPS of artificial intelligence performance for local training and real-time inference. The hardware is designed specifically for autonomous agents and complex data processing tasks.
The system represents a shift in how enterprise-grade computing is deployed outside traditional data centers. It provides local, private compute power for researchers and developers who face data transfer limitations. Consequently, organizations can run large models without relying entirely on cloud infrastructure.
Hardware Architecture and Performance
The ExpertCenter Pro ET900N utilizes the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip.
This architecture combines an NVIDIA Grace CPU and an NVIDIA Blackwell Ultra GPU. These components connect via the NVIDIA NVLink-C2C high-bandwidth, low-latency chip-to-chip interconnect.
For high-speed scaling, the system incorporates ConnectX-8 SuperNIC networking that supports speeds up to 800Gb/s. It also supports NVIDIA DGX OS to provide a standardized development environment. This setup allows developers to run trillion-parameter models directly at their deskside.
Memory and Expandability of the ExpertCenter Pro ET900N
The system features a 748GB coherent memory pool to enable faster data access during model training.
This pool combines 252GB of high-speed HBM3e GPU memory with 496GB of LPDDR5X system memory. Notably, this configuration offers more than double the GPU memory of a standard four-GPU workstation.
For physical expansion, the motherboard includes three PCIe x16 slots for additional graphics cards or accelerators. Storage needs are met through three M.2 SSD slots. Power is supplied via standard ATX and EPS12V connectors, supplemented by three 12V-2×6 connectors delivering up to 1800W to the GPUs.
Software Integration and Enterprise Use Cases
The ExpertCenter Pro ET900N is fully compatible with NVIDIA NemoClaw and NVIDIA AI Enterprise software.
This compatibility allows businesses to deploy autonomous AI agents locally while maintaining strict data privacy standards. Furthermore, the software stack is optimized for data science, machine learning, and analytics.
ASUS Co-CEO S.Y. Hsu stated that the company aims to connect powerful infrastructure with real-world experiences at the edge.
“At Computex 2026, ASUS is bringing its vision of ubiquitous AI to life by connecting powerful infrastructure with meaningful, real-world experiences at the edge.”
S.Y. Hsu, ASUS Co-CEO
Additional Product Announcements at Computex 2026
Alongside the main workstation, ASUS introduced the Ascent GX10 system for agentic applications.
This model is powered by the NVIDIA GB10 Grace Blackwell Superchip with 128GB of unified memory. It serves as a central processor for applications enabled by NVIDIA NemoClaw.
The company also announced the ProArt P16 and P14 laptops for creative professionals. These portable computers feature NVIDIA RTX Spark technology to support personal AI agents. The laptops are designed to bring advanced computing capabilities to mobile workflows.
Source: X (@asus) — researched
