ASUS ramps up production of NVIDIA Vera Rubin NVL72 systems: supply reliability becomes a key factor in the era of rack-scale AI

Edited by: Svitlana Velhush

ASUS Ships Vera Rubin Racks in Q4 ASUS announced on November 17 2026 that it is ramping volume shipments of NVIDIA Vera Rubin NVL72 systems during the fourth quarter. The company cites strong market demand for the rack-scale platform. ASUS is presenting an AI Factory reference

Image
Reply

On 17 November 2026, ASUS announced an increase in shipment volumes of NVIDIA Vera Rubin NVL72 systems in the fourth quarter. The company cites sustained demand from customers building large-scale AI infrastructures, while also showcasing its AI Factory reference architecture at SC26.

The focus is on the fully liquid-cooled XA VR721-E3 rack, which combines 72 Rubin GPUs and 36 Vera CPUs. The platform is designed for training frontier models, complex reasoning and high-performance inference. ASUS emphasizes not only the configuration itself, but also automated L10/L11 processes, six-level validation and early assessment of engineering risks through the Predictive Virtual Lab.

In a situation where several system integrators already offer NVL72, access to silicon ceases to be the only competitive advantage. Supply predictability, reduced pre-deployment risks and end-to-end support from facility planning to operations come to the fore. It is precisely on these aspects that ASUS is betting, expanding production capacity and offering digital twins together with Schneider Electric, AVEVA and IBM.

The rack's technical density — up to 227 kW of peak power — makes the manufacturing process an integral part of deployment risk. Automated optical inspection, precision assembly and multi-level testing (from node to full rack) are intended to minimize variations and ensure stable operation of NVLink 6, ConnectX-9 and BlueField-4. This is especially important for clusters where scaling from 256 to 1024 GPUs must maintain efficiency above 97%.

Comparison with previous generations shows that Vera Rubin offers a substantial gain: up to 2× performance per GPU compared with optimized GB300 NVL72 in MLPerf Training. However, for system vendors, the key factor becomes not only silicon but also the ability to integrate the entire ecosystem quickly and reliably — from power and cooling to management and validation.

Strategically, ASUS's move reflects an industry shift: large cloud providers and hyperscalers are increasingly evaluating not only peak characteristics but also the actual speed and predictability of obtaining finished racks. In this context, the ASUS AI Factory is positioned as a comprehensive solution covering planning, manufacturing, validation and subsequent support.

At the same time, the question remains open as to how quickly other integrators will be able to scale similar validation and automation processes. Independent tests of real clusters in production will show how sustainable ASUS's claimed advantages in supply reliability prove to be under mass deployments.

Ultimately, ASUS's increase in production underscores that in 2026–2027, the success of large-scale AI projects will be determined not only by access to advanced GPUs but also by the maturity of the entire supply and integration chain.

3 Views

Sources

  • Modelos y plataformas de IA – Unite.AI

  • unite.ai

  • newswire.ca

  • decisioninsights.ai

Comments

Read more articles on this topic:

Did you find an error or inaccuracy?We will consider your comments as soon as possible.