At ASUS AI Tech 2026 in Seoul, the company showed off a new generation of AI servers built around NVIDIA's Rubin architecture. The centerpiece was the HGX Rubin NVL8 solution, which packs eight Rubin GPUs into a single 1U chassis. That is a major density jump over earlier air-cooled designs, which needed multiple rack units to house the baseboard and its cooling hardware.
A notable change this generation is the flipped GPU orientation on the baseboard, which allows more efficient liquid cooling. ASUS is offering both hybrid and all-liquid cooling options so the nodes can fit into data centers with different cooling capabilities. The company also displayed trays from its GB200/GB300 NVL72 and Vera Rubin NVL72 systems, branded as the ASUS AI Pod, with the newer Vera Rubin design moving fans out of the chassis entirely.
The broader theme among speakers was efficiency: liquid cooling increases the performance you can extract from a given power footprint, and keeping more communication inside the NVL8 node reduces reliance on scale-out networking. ASUS paid for the author's travel, so the original coverage is disclosed as sponsored.