ASUS ExpertCenter Pro ET900N G3 brings NVIDIA Grace Blackwell Ultra AI supercomputing power to the desktop

Artificial intelligence hardware is escalating at a ridiculous pace lately, but every so often a machine appears that makes even seasoned nerds pause for a second. The ASUS ExpertCenter Pro ET900N G3 is one of those systems. ASUS showed it off at NVIDIA GTC 2026, and the premise is pretty simple: deliver data center level AI compute in a system that can live beside your desk.

And yes, I am going to say it. This thing is gorgeous. I could weep.

The ExpertCenter Pro ET900N G3 is powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Platform. That setup pairs a NVIDIA Grace CPU with a Blackwell Ultra GPU connected through NVLink C2C, which allows the two chips to share memory and communicate extremely quickly. That architecture is designed specifically for AI workloads that constantly move data between the CPU and GPU.

The memory configuration alone is enough to make hardware enthusiasts do a double take. The system features 748GB of coherent unified CPU GPU memory. That means large AI models can run locally without immediately slamming into the limits that traditional workstations often face. ASUS says the configuration can provide more than twice the GPU memory capacity of some four GPU workstation systems.

Then there is the compute power. ASUS claims the ET900N G3 can deliver up to 20 PFLOPS of AI performance. For context, that level of performance used to be associated almost exclusively with rack mounted infrastructure inside a data center. Seeing that kind of capability in a deskside system is honestly a bit surreal.

ASUS says the machine is based on NVIDIA DGX Station architecture, which means it is purpose built for modern AI development tasks rather than being a generic workstation. Workloads like LLM fine tuning, deep learning research, advanced simulations, and multimodal AI development are exactly what this hardware is intended to handle.

The company also frames the system as a solution to what it calls the “last mile” problem in AI development. Cloud platforms provide enormous computing resources, but they are not always ideal for every workflow. Developers often want local systems for privacy, experimentation, or simply to avoid the constant meter running in the cloud. A powerful deskside system like the ET900N G3 gives researchers and engineers the ability to run large models locally while still having access to serious compute power.

The system also integrates with the NVIDIA AI software stack, creating a ready to use environment for machine learning, analytics, and data science workloads. ASUS says multiple units can also be interconnected, allowing organizations to scale their compute capacity even further when needed.

Most people reading this will never have one of these machines sitting next to their desk. The ExpertCenter Pro ET900N G3 is clearly aimed at enterprises, research institutions, and serious AI development teams. Still, from a technology perspective, it is hard not to appreciate what ASUS and NVIDIA are doing here.

A deskside system with 748GB of unified memory and up to 20 PFLOPS of AI compute would have sounded absurd not very long ago.

ASUS says the ExpertCenter Pro ET900N G3 should become available worldwide in late Q2 2026. Pricing has not been announced yet, but given the hardware involved, it is safe to assume this will be a very expensive piece of kit.

Written by

Brian Fagioli ✔

Technology journalist and founder of NERDS.xyz

Brian Fagioli is a technology journalist and founder of NERDS.xyz. A former BetaNews writer, he has spent over a decade covering Linux, hardware, software, cybersecurity, and AI with a no nonsense approach for real nerds.

📄 More by Brian Fagioli ✖ Follow on X ▶ YouTube @ Threads 🐘 Mastodon