Asus Turns a Desktop Into a Home AI Datacenter Unveiled

Imagine a server room shrunk to the size of a PC case. Quiet. Clean. Sitting on your desk. That is the sales pitch, but this is more than marketing flair.

Asus has revealed the ExpertCenter Pro ET900N G3, a desktop that hides a data center's compute density inside a conventional tower. The chassis measures roughly 23.2 by 58.4 by 56.5 centimeters, small enough to live under a monitor and large enough to rewrite expectations about where heavy AI work can run.

Inside the desktop datacenter

At its core sits Nvidia's GB300 Grace Blackwell Ultra platform, pairing a 72 core ARM Grace CPU with a Blackwell Ultra GPU. The combo delivers up to 20 petaflops of AI performance, a level of raw throughput you normally find in rack cabinets, not on home desks. Memory is staggering: 748 gigabytes of unified capacity split into 496 GB of LPDDR5X for the CPU and 252 GB of HBM3e for the GPU.

This desktop can run trillion-parameter models locally, without sending data to the cloud.

A 1,600 watt power supply and an aggressive cooling system are paired to keep sustained workloads stable. That means long training sessions and heavy inference jobs will not force premature throttling. For researchers and small teams who want predictable, local performance, the ET900N G3 aims to remove the need for dedicated server rooms.

Connectivity and expansion are practical and plentiful. Asus included three PCIe 5.0 slots and support for up to four M.2 2280 NVMe drives. For multi-GPU or node-to-node traffic, NVLink-C2C and Mellanox ConnectX-8 networking are available. External I O is generous as well: eight USB 3.2 ports, one USB 2.0 port, and two 10 gigabit Ethernet ports.

• CPU: 72-core ARM Grace
• GPU: Nvidia Blackwell Ultra (GB300)
• Memory: 496 GB LPDDR5X (CPU) + 252 GB HBM3e (GPU) = 748 GB
• Compute: up to 20 petaflops AI performance
• Power: 1,600 W PSU; robust cooling
• Expansion: 3 x PCIe 5.0; 4 x M.2 2280 NVMe
• Networking: NVLink-C2C, ConnectX-8, 2 x 10 GbE
• Ports: 8 x USB 3.2, 1 x USB 2.0

Software choices land the unit firmly in professional hands. The ET900N G3 ships with Ubuntu and Nvidia's AI toolchain preinstalled; Windows support is listed as coming soon. That default configuration makes the machine ready for common AI frameworks and enterprise deployment scripts out of the box.

Who is this for? Data scientists who need local, low-latency inference. Startups that must keep sensitive models on-premises. Developers who want reproducible training runs without queuing for remote clusters. The machine is overkill for casual users, but for teams that need raw throughput and data control, it changes the cost-benefit math of buying rack space.

Asus priced the ExpertCenter Pro ET900N G3 starting at approximately €93,999. It is available in the United States now, with broader region availability expected to follow.

Whether this marks the start of more desktop-sized datacenters remains to be seen. For now, Asus has put the tools for large-scale AI research on a tabletop, and that alone is a conversation starter.