Summary
At its September 18, 2026 event in Seoul, ASUS presented an AI-infrastructure strategy built around integrated “AI factories” rather than compute alone. The company showcased enterprise servers, HGX training systems and rack-scale AI POD deployments powered by NVIDIA Vera Rubin NVL72.
ASUS used its AI Tech 2026 event in Seoul to present an infrastructure strategy focused on integrated “AI factories” rather than computing power alone. The company said increasingly autonomous and complex AI workloads are creating demand for systems that coordinate compute, power, cooling, connectivity and automation.
The event, held on September 18, 2026, included enterprise AI server platforms, HGX training systems and rack-scale ASUS AI POD deployments powered by NVIDIA Vera Rubin NVL72. ASUS said these systems are intended to support workloads ranging from enterprise inference to large-scale training and supercomputing.
Why AI infrastructure is becoming a broader systems problem
AI models require substantial computing resources, but the cost and efficiency of an AI deployment also depend on the systems around the processors. Power delivery, thermal management, networking and software automation affect how effectively that computing capacity can be used.
In his keynote, CH Hsieh, ASUS corporate vice president of its R&D Center, framed the issue around the cost of producing a token—the basic unit of text or other data generated by many AI systems. The question, according to ASUS, is shifting from simply building larger and faster data centres to understanding how efficiently those facilities can produce useful AI output.
ASUS summarised its approach with the phrase “Every Signal. Every Watt. Every Degree. Every Second.” The four measures refer broadly to the information moving through an AI system, the energy it consumes, the heat that must be managed and the time required to complete workloads. Treating these factors as connected parts of one infrastructure design can help organisations plan capacity and operating efficiency alongside raw performance.
What ASUS means by an AI factory
An AI factory is ASUS’s term for an integrated environment that supports the full AI lifecycle. Instead of treating model training, deployment, operations and governance as separate projects, the approach combines computing, storage, networking, deployment software and AI operations within a unified framework.
Training uses computing resources to adjust a model’s parameters, while inference runs a trained model to generate results for users or applications. The infrastructure requirements can differ significantly between those workloads. Training may require large, closely connected groups of accelerators, while enterprise inference may prioritise predictable response times, efficient resource use and integration with business systems.
ASUS also linked the infrastructure shift to the growth of agentic AI and physical AI. Agentic systems are designed to carry out multi-step tasks with greater autonomy, while physical AI applies AI capabilities to systems that perceive and act in the physical world. Both can create workloads that are more varied and operationally demanding than a single model responding to an isolated prompt.
From enterprise servers to rack-scale systems
ASUS presented a portfolio intended to cover different stages and scales of AI adoption. Enterprise AI servers are aimed at individual organisational workloads, while HGX training platforms provide infrastructure for developing and training larger models. At the rack scale, ASUS AI POD deployments combine multiple infrastructure components into larger integrated systems.
The company said its AI POD examples used NVIDIA Vera Rubin NVL72 and were designed for large-scale AI training and supercomputing workloads. Taken together, the products illustrate ASUS’s stated strategy of offering infrastructure that can extend from enterprise inference to more demanding training environments.
The announcement presents AI factories as a long-term operating model for AI adoption: a way to connect deployment, management, governance and continuous optimisation instead of building isolated systems for each project. Its central shift is from viewing AI infrastructure as a collection of powerful processors to treating it as a coordinated system in which energy, cooling, networking and automation are part of the performance equation.