Summary
ASUS described a four-stage AI infrastructure lineup at DCD Connect London, spanning developer systems, enterprise computing, liquid-cooled data-centre servers and a rack-scale AI factory.
ASUS has outlined a four-stage AI infrastructure portfolio intended to take organisations from local development systems to large-scale data-centre computing. The company’s press release, dated September 22, 2026, describes a showcase at DCD Connect London, lists the event dates as September 16–17 and identifies Booth 35.
The lineup combines ASUS hardware with NVIDIA-accelerated computing. It ranges from deskside development platforms to systems ASUS describes as AI factories for model training and token generation.
Four tiers of AI computing
For developers, ASUS presents the Ascent GX10 as a local platform for AI engineering, model iteration and software training. The next tier is the ExpertCenter Pro ET900N, which the company positions as a more powerful deskside system. ASUS also describes pairing it with server-class ET900N and NVIDIA-Certified Systems for enterprise adoption.
At the data-centre level, ASUS lists liquid-cooled servers built on NVIDIA HGX Rubin NVL8. The company says these are designed to deliver dense, multi-GPU computing within conventional data-centre space and thermal constraints.
At the largest scale, ASUS describes an AI factory built on NVIDIA Vera Rubin NVL72: a rack-scale, co-designed system intended for continuous training of multi-trillion-parameter models and high-volume token generation. Tokens are the units of text that language models process and produce; generating them is part of inference, when a trained model responds to a prompt.
The case for a connected stack
ASUS’s central argument is that scaling AI requires more than adding processors. Its proposed portfolio also brings together networking, cooling, storage, deployment and support. Coordinating those layers, the company says, can reduce deployment complexity, shorten time to revenue and lower total cost of ownership.
The progression reflects a practical infrastructure challenge: a system suitable for experimenting with a model locally may not be suited to serving it at enterprise scale. ASUS is presenting the four tiers as a route between those stages, rather than as a single system for every workload. The release is a portfolio overview; it gives no benchmark figures for the performance or cost benefits it describes.