Inside ASUS’s 2026 Thermal Lab for AI Server Testing

ASUS’s Thermal Testing Facilities

ASUS operates a server thermal testing lab in Taiwan for customer validation, thermal qualification, and long-term reliability testing. The facility includes several specialized areas, including environmental chambers that simulate operating conditions and accelerate the effects of thermal stress.

Walk-In Environmental Chamber

AI server development moves quickly, so manufacturers need facilities that can simulate years of thermal exposure within shorter test periods. Environmental chambers help validate heat sink designs, fans, complete servers, and rack-scale systems under different customer conditions.

The walk-in environmental chamber supports operating-range validation for individual servers and complete racks. ASUS runs tests from 25 degrees Celsius to 45 degrees Celsius, covering common data center ambient conditions.

During the visit, an ASUS NVIDIA HGX B200 8-GPU server from the “Blackwell” generation occupied the chamber. Its size allows ASUS to test complete rack configurations instead of only individual nodes. This is important for liquid-cooled systems, where coolant distribution and manifold behavior change at rack scale.

Testing a complete rack also exposes thermal interactions between nodes that single-server validation cannot reproduce. At the same time, individual server testing remains necessary. One current 8-GPU server node can draw the power of two older 208V/30A racks, placing substantial demands on a single system.

Many vendors use environmental chambers for individual air-cooled servers. ASUS designed this chamber to accept full racks because liquid cooling creates dependencies between nodes. Coolant temperature, flow rate, and pressure drop behave differently across a 72-GPU rack than across a single compute tray. A partially liquid-cooled server can also behave differently from fully air-cooled or fully liquid-cooled equipment.

AI infrastructure increasingly requires system-level validation rather than component-level checks. Chambers built for systems up to a single 4U server are not large enough for modern rack-scale deployments. The chamber was complete during the June 2026 visit, while an April 2026 visit took place shortly before its completion.

Extreme Environmental Chamber

The extreme environmental chamber is designed for conditions beyond normal data center operation. It accepts complete racks and can simulate temperatures from minus 40 degrees Celsius to 85 degrees Celsius. Humidity testing ranges from 10 percent to 98 percent, covering environments from cold outdoor installations to tropical locations with limited climate control.

Testing at these extremes serves two main purposes. First, it validates whether rack components can withstand shipping and storage in unconditioned environments. Temperature became a more significant consideration when complete NVIDIA GB200 NVL72 racks began moving by air and sea freight. Many NVIDIA GB200 NVL72 and newer racks are shipped in climate-controlled trucks.

Second, extreme testing helps identify failures that appear only after sustained thermal stress. High temperatures accelerate component aging, and operating servers at 85 degrees Celsius for extended periods can reveal weaknesses that might take years to emerge under normal conditions.

During the visit, ASUS operated the rack’s liquid-cooling system with coolant at 20 degrees Celsius while also testing higher coolant temperatures. Warmer coolant can reduce chiller energy consumption, but it may reduce component reliability and performance headroom. Establishing the appropriate balance requires measuring failure rates during thousands of hours of accelerated-aging tests.

The chamber can become very noisy when its environmental systems operate at full capacity. Its rear section contains the power and fluid connections, along with the environmental equipment. Remote monitoring is therefore important for long test runs, allowing technicians to keep the chamber sealed and operate it continuously without remaining inside.

ASUS AI System R&D Lab

The AI System R&D Lab focuses on rack-scale liquid-cooling integration rather than validation of individual servers.

Rows of racks fill the lab, with power distribution installed overhead and liquid-cooling infrastructure beneath the raised floor. This arrangement resembles customer production data centers and allows ASUS to validate installation procedures and maintenance workflows before systems are shipped.

The piping network beneath the raised floor distributes coolant throughout the facility. Each rack connects to a manifold system, demonstrating how liquid cooling scales from individual racks to multi-rack pods. Pipe diameter, valve placement, and leak detection must be coordinated across the installation. ASUS also verifies that sensors operate correctly and report consistently through its management software.

Liquid-cooling validation now involves more than detecting leaks. Testing must also determine how a leak affects the rack and the operation of the wider cluster.

The facility supports up to 1.1 megawatts of power across its rack positions. Liquid-cooled AI racks have significantly higher power density than traditional enterprise servers, so validating thermal performance requires infrastructure that reflects production deployments. Air-cooled testing facilities cannot reproduce all the coolant-distribution challenges that occur at these power levels.

The integration of an NVIDIA GB300 NVL72 rack differs substantially from testing individual NVIDIA GB300 nodes. Rack-scale testing addresses cooling across 72 GPUs, coolant flow through multiple cold plates, and the operation of leak-detection and monitoring systems. The lab is used to validate these behaviors before customers deploy production hardware.

The work also includes detailed electrical and system diagnostics. Although complete NVIDIA GB300 NVL72 racks are visible in deployed environments, validation may involve only a few nodes connected to a Tektronix scope while engineers examine system behavior.

Because the lab uses a raised floor and modern server equipment is heavy, a ramp connects the staging area to the raised floor. The facility supports both liquid-cooled and air-cooled equipment, so it also uses air handlers similar to those in a conventional data center. Warm air from the hot aisle is drawn toward the ceiling, cooled, and returned to the cold aisle.

Noise levels can be high, making hearing protection important. A power and chiller room supporting the R&D lab is located nearby.