i卡會員

With the booming development of artificial intelligence (AI) and high-performance computing (HPC) applications, the energy consumption from data centers around the world has been rising sharply; as a result, cooling technologies are at the center of unprecedented challenges and opportunities. In particular, and as recently widely discussed, generative AI such as ChatGPT requires massive numbers of GPUs for training and inference. The advanced chips deployed often consume power at the kilowatt level, generating enormous heat within server racks. Cooling is no longer merely a supporting element. Rather, it has become a critical factor determining whether AI computing power can operate reliably and whether data centers can improve their energy efficiency. 

The so-called kilowatt-level power consumption of high-end chips refers to their Thermal Design Power (TDP), i.e., the maximum amount of heat a chip generates under heavy workload and measured in watts (W). The higher the TDP, the greater the heat generated by chips during high-speed operation, and the more demanding the requirements for cooling systems. This indicator not only affects the operational stability of servers, but is also one of the key considerations for the planning of AI data centers. 

Since 2017, all the flagship CPUs and GPUs of the three major chipmakers, namely Intel, AMD, and NVIDIA, have shown a significant jump in TDP. Intel’s Xeon server processors have increased from around 200 W and are expected to reach 500 W by 2025. AMD’s EPYC processors have similarly risen from 180 W to about 500 W. Meanwhile, NVIDIA’s data-center GPUs are projected to climb from 300 W (V100) to as high as 1,000 W (B100/B200), representing the largest increase among the three. These figures reflect the rapid rise in computing density required by AI and HPC, which in turn drives up chip thermal design power and cooling demands.

As chip power consumption continues to escalate, traditional air-cooling methods that rely on fans and heat sinks are no longer sufficient to handle such high thermal densities. Therefore, the industry is accelerating its transition toward more efficient technologies such as liquid cooling and two-phase cooling. In particular, once chip TDP exceeds 1.5 kW, conventional single-phase water cooling—which removes heat by circulating water—becomes increasingly inadequate. It is necessary to adopt a two-phase cooling solution that absorbs heat through liquid vaporization, or even immersion cooling, where entire servers are submerged into special non-conductive liquids. It is through cycles of boiling and condensation that these liquids are able to carry away heat effectively and significantly improve the heat transfer and cooling performance. 

For example, Supermicro introduced its Data Center Building Block Solutions (DCBBS) and next-generation DLC-2 (direct-to-chip) liquid-cooling technology at COMPUTEX 2025. This system can improve cooling efficiency using inlet water temperatures of up to 45°C and achieving a cooling efficiency of as high as 98%. In practical terms, it can reduce power consumption by around 40%, cut space requirements by 60%, and lower water usage by 40%. As a result, the total cost of ownership (TCO) can be reduced by roughly 20%, furthermore, deployment can be completed in as little as three months. Meanwhile, Amazon Web Services (AWS) unveiled its In-Row Heat Exchanger (IRHX) liquid-cooling technology in 2025. The greatest advantage of this technology lies in its support of high-power GPUs without the need for extensive retrofitting of existing data centers.

In the global race for cooling technologies, Taiwan has already become the major supplier of key components such as fans, heat pipes, and thermal modules, with a market share exceeding 70% for related products. Thanks to its strong foundation of the electronics industry and government support, Taiwan has made a number of breakthroughs in recent years. The technology initiatives orchestrated by the Department of Industrial Technology of the Ministry of Economic Affairs has long focused on research and development in high-performance computing and advanced cooling technologies. Working together with the Industrial Technology Research Institute (ITRI) and domestic companies including AEWIN Technologies, I-Chiun Precision Industry, Kenmec Mechanical Engineering, Fusheng Group, Cooler Master Corp., and GIGA-BYTE Technology, these efforts aim to build a comprehensive AI server cooling system supply-chain ecosystem and transform Taiwan from a leading supplier of cooling components into an exporter of high-end system solutions. 

One representative achievement in this field is the kilowatt-class vapor-chamber lid. This technology introduces a specially designed vapor-chamber lid (VC Lid) onto the surface of the chip which utilizes the phase-change properties of water to dramatically improve heat transfer efficiency. Its thermal performance per unit area can be up to ten times that of conventional copper material, resulting in chip cooling performance improvements of 30% to 100%. This technology has entered the supply chain of major U.S. high-performance computing chip companies and become a core cooling solution for next-generation high-power AI chips. 

Another breakthrough is the two-phase immersion cooling system, in which entire server units are submerged in a non-conductive liquid. As heat is removed through cycles of the liquid boiling and condensing, the thermal dissipation capacity surpasses the previous ceiling of around 600 W for single-phase immersion and jumps to above the 1,500 W level. This breakthrough effectively raises the bar for existing cooling techniques and represents the world’s first kilowatt-class chip cooling technology.

By delivering integrated cooling solutions spanning chips to systems, Taiwan is able to demonstrate not only its R&D capability for cutting-edge cooling technologies, but also its collaborative ties with leading global players. With the support of the Department of Industrial Technology under the Ministry of Economic Affairs, kilowatt-class thermal modules and two-phase immersion cooling systems have been successfully adopted by major global IC design firms. These solutions help to address the thermal management needs of the next-generation high-power AI chips, improving both chip stability and energy efficiency and ensuring Taiwan’s presence in the global HPC industry chain. Looking ahead, these technologies are expected to find their way into the fields of edge AI, automotive HPC, defense communications, high-speed data exchange, and satellite ground stations—all of which involve high thermal density equipment. Such developments will further strengthen Taiwan’s strategic position in the global supply chain for computing infrastructure.