AI is almost omnipresent in our daily lives, from navigation and voice assistants to social media recommendations and smart customer service. Behind these applications are thousands of servers working silently. These servers are concentrated in places called "data centers," which operate like computational cities, running efficiently and non-stop year-round. However, servers running at high speeds for long periods generate significant heat. If this heat is not properly dissipated, it can not only reduce efficiency but also damage equipment, much like a smartphone overheating in the summer. Therefore, cooling systems have become indispensable for data centers.
The most intuitive way to cool AI servers is to use air conditioning, similar to home air conditioner, to maintain a stable low temperature in the server room. This method is straightforward but consumes high energy in large AI data centers. Many data centers design "cold aisles" to blow cool air and "hot aisles" to exhaust hot air, preventing the mixing of cold and hot air to improve cooling efficiency.
However, as AI models grow larger and power consumption increases, traditional air conditioning is no longer sufficient. This is where liquid cooling technology comes into play, using conductive liquids instead of air to directly remove heat from servers. This method cools faster and is more energy efficient.
The role of cooling systems in AI data centers goes beyond just "cooling." It is also about overall energy efficiency and sustainability. Cooling systems maintain the stability and safety of AI operations, reduce overall energy consumption, lower carbon emissions, save on electricity costs, and improve operational efficiency. Many data centers now even recycle heat for reuse, such as heating building water or providing warmth, ensuring no heat is wasted.
In the future, AI data centers will not only symbolize efficiency but also represent green technology. From intelligent cooling to environmentally friendly designs, every innovation is an effort to coexist with technology and the environment. Just as we need air conditioning for comfortable living, AI needs cooling systems to continue delivering intelligence and serving us every moment.
To enhance the efficiency of cooling systems in AI data centers, Arrow and its partners have been dedicated to developing related solutions. At the upcoming PCIM Asia 2025 exhibition, you will see the results of these efforts.
Cooling systems for AI data centers: liquid and immersion cooling
As AI applications become more widespread, the computational load of AI data centers is rapidly increasing, with server heat reaching scorching levels like summer pavement. Traditional air cooling (using air to cool, like air conditioning) is gradually unable to meet the high-density, high-heat demands of computing. This is where two new technologies, "liquid cooling" and "immersion cooling," come into play.
Liquid Cooling involves using liquids instead of air to dissipate heat from servers. The most common method is to circulate water or other cooling fluids through special pipes close to high-temperature components (such as CPUs and GPUs) in the server, effectively "cooling" the AI's brain and quickly removing heat. This method offers high cooling efficiency, making it particularly suitable for high-performance AI computing. Compared to air cooling, it is more energy-efficient, environmentally friendly, reduces internal temperature differences in servers, and helps extend server lifespan.
Immersion Cooling takes this a step further by submerging entire servers in a special non-conductive cooling liquid that is harmless to equipment. Heat is rapidly transferred from the server surface to the liquid and then expelled through the cooling system. This method provides extremely high cooling efficiency, making it ideal for extreme high-density AI training scenarios. It almost eliminates the need for fans, reducing noise and power consumption while saving space, enabling more compact data center designs.
As AI models become increasingly complex, such as large language models, visual recognition, and real-time translation, requiring massive computational power. Traditional air cooling is not only inefficient but can also lead to high electricity costs and carbon emissions. Liquid and immersion cooling are not only more energy efficient but also ensure stable and safe long-term operation of AI, representing a significant step toward green data centers. Cooling technology has evolved from a simple temperature-lowering tool to an indispensable part of AI infrastructure. Both liquid and immersion cooling are key technologies driving the parallel advancement of AI and sustainability, ensuring AI operates intelligently and coolly. It’s a symbol of the new era where intelligence and environmental responsibility go hand in hand.
Cooling solutions make AI data centers more stable and environmentally friendly
Arrow Electronics and its partners have been committed to making AI data centers more stable and improving energy efficiency by offering a wide range of cooling solutions. Below, we will introduce these solutions by application type, including power conversion, motor control, and smart interconnects.
1.Power conversion
In liquid and immersion cooling systems, power conversion acts like the "power hub" behind the cooling system, responsible for converting electricity into the appropriate voltage and current needed to drive pumps, fans (if required), and cooling modules. Its main tasks include converting grid input into stable low-voltage power for cooling equipment, controlling the operational efficiency of pumps or heat exchange units, and integrating with sensor systems for intelligent regulation to avoid energy waste. Through efficient power conversion solutions, liquid and immersion systems can operate stably and energy-efficiently, providing reliable cooling for AI data centers.
3 Phase 3 Level T-Type Bidirectional Power Converter with on-grid function — This solution is a high power three-phase Active Front End (AFE) bidirectional rectifier. Which is based on the three level T-Type topology. It can be widely used for industrial power supply, electric vehicle fast charging and high-power motor driver applications. It features fully digital control for high performance microcontrollers of ST and providing full control of PF, THD, bidirectional power flow and on-grid DC-AC function. It using 3 Level T-Type topology provided higher efficiency and lower power losses.
Download related information
• 3 Phase 3 Level T-Type Bidirectional Power Converter with on-grid function
2.Motor control
Smart fan demo — This demo uses far-infrared thermal sensors to detect heat sources and indicates them via an LED board. When a heat source is detected in the middle of the detection zone, the motor starts. The higher the temperature, the faster the motor spins. It uses NXP's MCU MXCA153 to receive temperature data and controls the motor via PWM. When the temperature exceeds a threshold, the fan speed increases. This smart fan demo combines edge AI, motor control, and multi-point thermal detection, making it suitable for intelligent applications in various fields.
Download related information
• Smart fan demo
10 kW PMSM motor driver — This motor drive control board, introduced by STMicroelectronics (ST), is based on the STM32G4 platform and includes one motor with a three-phase Vienna PFC and one MCU for high-efficiency cooling. This high-power cooling system can be used for data center infrastructure management, such as direct liquid-cooled AI servers and telecom base stations. It employs WBG materials, and the motor variable-speed drive can adjust motor speed based on current demand, saving 20-60%. The drive optimizes motor flux based on load (SRM), saving up to 20%, and supports distributed dPFC to improve power quality. System analysis enables cooling optimization, saving approximately 30% in electricity.
3. Smart interconnects
In liquid and immersion cooling systems, smart interconnects act like a neural network, linking sensors, controllers, motors, and energy systems to enable real-time monitoring and intelligent collaboration throughout the cooling process. The main functions of smart interconnects include transmitting data such as temperature, pressure, and flow rates for precise control, enabling remote monitoring and diagnostics to improve maintenance efficiency, and integrating with the energy and computing management systems of AI data centers for overall optimization. Through smart interconnects, cooling systems are no longer standalone mechanical devices but become part of the intelligent operation of AI data centers, making the entire system more energy-efficient and reliable.
OSFP interconnect system — Molex's Octal Small Form Factor Pluggable (OSFP) interconnect system and cable assemblies provide 56 or 112 Gbps PAM-4 per lane, single-port, 8-lane I/O connectivity. The OSFP connector system supports 16 channels of 112 Gbps PAM-4, with 4 Rx and 4 Tx pairs per row, delivering an optimized interface for aggregated data rates of 400 and 800 Gbps. It improves thermal management and complies with IEEE and OSFP industry standards. The compact DAC, AOC, ACC, and optical modules offer flexibility for data center solutions, telecom/network high-density switch applications, and meet the growing demands of AI, machine learning (ML), high-speed networking, and other high-speed applications.
Download related information
• OSFP 112 Gbps PAM-4 interconnect system and cable assemblies
Conclusion
As AI continues to drive innovation, the data centers behind it are quietly evolving. From liquid to immersion cooling, these continuously optimized cooling technologies are the key forces ensuring AI operates stably, efficiently, and with low carbon emissions. Cooling is not just a technical detail but a commitment to sustainability. As we make AI smarter, let’s also ensure that every kilowatt of electricity and every drop of cooling fluid brings us closer to future green standards.