ASUS AI Tech 2026 in Seoul: Denser AI Systems and Liquid-Cooled Racks

ASUS AI Tech 2026 in Seoul: Denser AI Systems and Liquid-Cooled Racks

ASUS AI Tech 2026 in Seoul, South Korea, brought together ASUS systems, suppliers, partners, and customers to showcase the company’s latest AI servers and rack-scale platforms. The event included NVIDIA HGX Rubin NVL8 systems, ASUS AI Pod trays, and newer liquid-cooled designs based on NVIDIA GB200/GB300 NVL72 and Vera Rubin NVL72 platforms.

NVIDIA HGX Rubin NVL8 systems

The event display included NVIDIA HGX Rubin NVL8 solutions with eight Rubin GPUs. ASUS presented several versions, including a liquid-cooled design that differs substantially from the Ampere, Hopper, and Blackwell generations.

The system changes the GPU orientation on the baseboard to support more efficient liquid cooling. NVIDIA also introduces a new generation of NVLink Switch technology, along with high-speed NVIDIA networking for scaling out to larger clusters.

With liquid cooling, the eight GPUs can fit in a 1U configuration. Air-cooled HGX baseboards generally require 6U or more, so the new arrangement allows substantially higher density.

ASUS uses another chassis segment for the CPUs, SSDs, DPUs, and related components. This approach places a large amount of compute in a smaller rack footprint. The nodes can use hybrid cooling or all-liquid cooling, allowing them to support data centers with different cooling capabilities.

Although individual GPUs can use more power than Hopper-generation GPUs, the higher-performance Rubin GPUs provide more compute resources and memory within the NVL8 topology. Keeping more communication inside the node reduces data movement across the scale-out fabric. As a result, each GPU can process more work, while the NVL8 domain can handle more data and provide higher internal communication performance.

ASUS AI Pod and rack-scale systems

ASUS also displayed trays from its NVIDIA GB200/GB300 NVL72 and Vera Rubin NVL72 platforms, which it refers to as the ASUS AI Pod. The exhibition included liquid-cooled ORv3 trays from these systems.

All-liquid cooling improves efficiency, an important factor in determining how much computing performance a data center can deliver. The same focus on cooling and efficiency was a recurring topic among the event’s speakers.

The Vera Rubin NVL72 design illustrates the shift toward removing fans from server chassis and using liquid cooling throughout the rack. This arrangement can increase compute density and maximize performance within a defined power footprint.

ASUS has also described its testing processes in its thermal lab, including work involving AI server designs. The event continued with storage systems and a panel discussion focused on the broader AI infrastructure ecosystem.