Asus Launches ExpertCenter Pro ET900N G3 Built On Nvidia DGX Station Architecture

Asus has released the ExpertCenter Pro ET900N G3, a deskside AI supercomputer built on the Nvidia DGX Station GB300 architecture and powered by the Nvidia GB300 Grace Blackwell Ultra Desktop Superchip.

The system is designed to bring data-centre-class AI computing to enterprises, developers and researchers without requiring dedicated data-centre environments, and is available worldwide through local Asus representatives.

Asus is positioning the ET900N G3 as a local alternative to cloud-based AI infrastructure, targeting organisations that need to run large-scale AI training, inference and autonomous agent workloads on premises while maintaining control over sensitive data.

Hardware and architecture

At the core of the ET900N G3 is the Nvidia GB300 Grace Blackwell Ultra Desktop Superchip, connected via Nvidia NVLink-C2C high-bandwidth interconnect technology for coherent memory access between the CPU and GPU.

The system ships with 748GB of coherent unified CPU-GPU memory, which Asus claims is sufficient to run AI models with up to one trillion parameters locally. It delivers up to 20 PFLOPS of AI performance in a form factor designed to sit beside a desk rather than in a server rack.

The 748GB memory pool is intended to address the workflow bottlenecks Asus says are commonly associated with conventional workstations when handling large AI models. By keeping the entire model in coherent memory, the system avoids the overhead of shuttling data between discrete memory pools during training and inference.

The DGX Station GB300 architecture underpinning the system provides enterprise-grade scalability and connectivity, and Asus has designed the platform to support interconnected AI workflows so organisations can scale compute capacity as their requirements grow.

Software stack and supported workloads

The ET900N G3 supports the Nvidia AI software stack out of the box, providing an environment for machine learning, analytics and AI experimentation without significant additional configuration.

Asus says the system is suited to a range of workloads including large language model fine-tuning, generative AI, physical AI, deep learning research and autonomous AI agent development. Future support is also planned for Windows-based AI development and agentic environments.

The platform supports Nvidia NemoClaw workflows, which are designed to help enterprises and developers build and deploy always-on AI assistants and autonomous agents within local environments. Combined with the Nvidia AI software stack, the NemoClaw blueprint is intended to simplify the development of AI applications while maintaining enterprise-grade control and operational flexibility.

Benchmark results

Asus engineering teams conducted stress testing on the ET900N G3 using vLLM with the Qwen open-source AI model.

In those tests, the system achieved approximately 864 tokens per second in output throughput, with combined input and output processing reaching around 1,600 tokens per second. Asus says these results demonstrate the system’s capacity to handle large-scale open-source AI workloads at the deskside.

On-premises deployment and data governance

Asus is targeting the ET900N G3 at organisations that want to keep AI workloads on premises for data governance and privacy reasons, rather than relying entirely on cloud computing.

By running AI locally, enterprises can reduce their dependency on cloud infrastructure while maintaining control over sensitive data. The on-premises approach also offers lower latency for real-time AI operations and more predictable operational costs compared to usage-based cloud pricing.

The system is aimed at AI research labs, enterprise AI deployment teams, content creation studios and organisations running simulation workloads.

Availability

The Asus ExpertCenter Pro ET900N G3 is available worldwide. Pricing and regional configurations are available through local Asus representatives.

Last Updated on June 15, 2026 by Nick Ross