Asus ExpertCenter Pro ET900N G3 Puts GB300 AI Superchip on the Desktop

The tower weighs 27 kilograms net and measures 584 x 232 x 565 mm, a size and weight that keep it in a fixed desk or under-desk position. It combines 72 Arm cores with a Blackwell Ultra GPU and 748GB of coherent memory, unifying GPU and system memory. TechRadar reported that Asus has not disclosed exact pricing.

Asus gave the workstation a brushed silver exterior, a dark mesh front panel, and two metal handles on the top edge for safe relocation. An enterprise-grade sealed liquid cooling loop with copper cold plates cools the Superchip, adjacent memory modules, and high-speed network card. A dedicated internal fan directs air onto the network card’s optical transceivers to prevent thermal throttling under sustained networking loads.

Power comes from a 1,600-watt power supply certified at 80 PLUS Titanium efficiency. The system ships with a mirrored array of two 2TB NVMe drives to protect the operating system and software stack from single-drive failure. The internal layout leaves room for three PCIe slots and two additional expansion bays for extra storage or supplemental graphics cards.

In hardware testing, the architecture reached 4,793 output tokens per second across 128 concurrent streams on the MiniMax M2.7 benchmark, according to TechRadar. GLM-5.2 pushed 215GB of expert weight data directly through the coherent memory fabric, bypassing latency bottlenecks that TechRadar said are typical on standard system architectures.

The workstation includes two 400-gigabit networking ports for up to 800Gb/s combined bandwidth. Linking two GB300 stations creates a single 1,496GB coherent memory pool, doubling the standalone unit’s capacity. StorageReview reviewer Brian Beeler said, “The GB300 DGX Station remains the most capable thing we’ve ever put on a desk.” Technology YouTuber Alex Ziskind said, “Everything you’re about to see, dozens of AIs all thinking at once, is running on this one machine, not a data center.”

Running dozens of coding agents simultaneously, the machine sustained thousands of tokens per second without individual agents slowing noticeably, TechRadar reported. On GPT-OSS-120B, output climbed to 9,280 tokens per second, with total throughput reaching more than 45,000 tokens per second. Concurrency tests scaled smoothly from a single agent to 128 simultaneous agents without a system crash or slowdown, and Ziskind demonstrated power draw climbing toward 1,000 watts while coding, research, and writing agents ran at once.

TechRadar estimated the device will cost between $100,000 and $120,000, while noting that Asus has not publicly disclosed exact pricing. The report compared the tower visually to Apple’s classic Mac Pro chassis while placing its performance in a different class from standard high-end desktop computers.