Inside the ASUS AI Lab in Taiwan: where ASUS "solves the problem" of operating AI before going to the data center
Delicate recently visited ASUS AI Lab in Luzhu area, Taoyuan (Taiwan) and saw that the company is no longer just a hardware manufacturer, but is operating a large-scale AI infrastructure "test workshop", simulating almost all data center conditions before the system is deployed to customers globally.
ASUS AI Lab is designed as a closed ecosystem, covering the entire lifecycle of AI infrastructure: from research, design, testing, validation to implementation and optimization of actual operations. Instead of focusing solely on computing power, every process in the lab revolves around three factors: stability, scalability and energy efficiency of large-scale AI systems.
At the heart of this complex is RD Lab, where ASUS builds AI server racks with both air and liquid cooling solutions, simulating modern data center architectures. Here, the firmware is tested, debugged and performance tweaked right on configurations almost equivalent to the production system, helping to significantly shorten the time to validate and put the product into mass production.
To handle the heat and power problem, ASUS invested in a separate circulating cooling system for the AI lab, gradually replacing the traditional chiller architecture with a solution that operates at higher temperatures. According to the technical team, this approach saves nearly 30% of power per year, while still meeting the cooling requirements of high-density GPU clusters.
All operating data is collected in the Chamber Monitoring area, which can be considered the "nerve center" of ASUS AI Lab. Engineers monitor real-time air temperature, liquid temperature, humidity, pressure, air flow, power consumption, and alarms for leaks or system failures, thereby simulating different PUE scenarios before bringing the same configuration to the customer's data center.
In parallel, QTR Lab is responsible for testing durability in harsh environmental conditions, with the ability to simulate temperatures from -40°C to 85°C and humidity from 10% to 98%. In this region, ASUS is testing systems like the NVIDIA GB200 NVL72 with 72 Blackwell Tensor Core GPUs, targeting AI models with trillions of parameters, to ensure they remain stable in a variety of climates and infrastructure conditions.
Thermal Lab focuses on the actual operation of data centers with simulation of hot aisles and cold aisles, controlling airflow, temperature and pressure. Through this, ASUS can evaluate in detail the efficiency of air and liquid cooling at the rack level, thereby optimizing operating costs as GPU density increases.
Not only stopping at hardware, ASUS also develops a management platform layer to simplify the deployment and operation of AI infrastructure. ASUS Infrastructure Deployment Center (AIDC) helps automate operating system installation, network configuration, and system deployment, while ASUS Control Center Data Center Edition provides a centralized interface to monitor HPC, AI, and enterprise infrastructure environments.
