ASUS readies NVIDIA HGX Rubin NVL8 servers for Q4 2026 launch

Two HGX Rubin NVL8 server designs

According to ASUS, the new portfolio includes two models: the 3U hybrid-cooled XR R1I12B-M1BHFS and the 2U, 100% liquid-cooled XR R1I12B-M1BLFS. The company describes them as flexible, high-performance platforms built for massive AI reasoning and workloads.

The 3U XR R1I12B-M1BHFS combines direct-to-chip (D2C) liquid cooling for the NVIDIA HGX Rubin 8-GPU baseboard with air cooling for dual Intel Xeon 6 processors. ASUS states that the air-cooled CPU sled reduces maintenance time by up to 50%. The system also offers flexible U.2/E1.S storage options to support different deployment requirements.

The 2U XR R1I12B-M1BLFS uses a fully liquid-cooled architecture. ASUS says it is designed for high thermal efficiency, improved PUE and high computing density in demanding AI environments. Its NVIDIA-validated design and cable-less board-to-board architecture are meant to simplify system integration and reduce deployment complexity.

AI factory design and simulation with NVIDIA DSX Sim

ASUS argues that building an AI factory requires optimization across compute, power, cooling and facility infrastructure. To address this, the company combines its systems engineering expertise with the NVIDIA DSX Sim Blueprint. The goal is to let enterprises design, simulate and validate AI factory infrastructure in a digital twin environment while optimizing for large-scale token generation.

ASUS works with ecosystem partners including Schneider Electric, AVEVA and IBM. Together, according to the company, they integrate compute, networking, storage, power, cooling and facility infrastructure into one unified digital twin. Customers can then evaluate infrastructure configurations, identify constraints and optimize performance and efficiency before deployment.

Wider portfolio on show in Berlin

At GTC Berlin 2026, ASUS is exhibiting at Booth #3039 at Station Berlin. Alongside the HGX Rubin systems, it shows NVIDIA Grace Blackwell-powered products. The ASUS Ascent GX10 targets AI model fine-tuning, prototyping and inference, while the ExpertCenter Pro ET900N G3 is aimed at demanding AI development and large-scale model training.

For work beyond the data center, ASUS also presents the ProArt GR1X Mini PC and the ProArt P16, which it positions for developers and creators. The company frames the line-up as covering the full agentic AI journey, from infrastructure and model development to edge and agent computing.

Visitors can also see the edge-to-cloud portfolio in the ASUS VIP Room at the W3 Hub, open October 20-22. It includes edge AI systems and enterprise infrastructure based on NVIDIA HGX and NVIDIA RTX PRO.

What it means for buyers

For enterprises planning AI infrastructure, the announcement offers two cooling approaches on the same HGX Rubin NVL8 platform: a hybrid design that keeps air cooling for the CPUs, and a fully liquid-cooled one for higher density. The source gives no pricing, exact shipping dates within Q4 2026 or performance figures, so those details remain open. The DSX Sim-based simulation offer suggests ASUS wants to help customers test infrastructure choices before committing to a deployment.