Why liquid cooling is key to the future of AI data centers
As artificial intelligence (AI) continues its unstoppable rise, the surge in computing power brings with it the “happy problem” of enormous energy consumption and intense heat generation. Chip lifespan tends to decrease as temperature rises. There is a saying in the industry that for every 10°C increase in core temperature, a chip’s expected lifespan is halved. If heat dissipation issues are not effectively resolved, chips may overheat, leading to degraded performance or even damage, seriously affecting the stable operation of data centers. To prevent chips from suffering the damaging consequences of persistent “high fever,” emerging liquid cooling technologies are reshaping the rules of chip thermal management.
Liquid Cooling Technology: Keeping AI Computing Power Cool
Liquid cooling encompasses a range of technologies, including cold plate and immersion or spray-based cooling systems. The preferred solution to address thermal challenges in data centers, liquid cooling technology uses liquid mediums to directly absorb and transfer heat, enabling highly efficient heat dissipation. Liquids have significantly higher specific heat capacity and thermal conductivity than air. Under the same conditions, liquid cooling can be 1,000 to 3,500 times more efficient than air cooling. This exceptional thermal performance makes liquid cooling ideal for high-density, high-power server environments, effectively lowering equipment temperatures and enhancing system stability.
In early real-world data center applications, liquid cooling technology has delivered remarkable results. Studies have shown that, compared to traditional air-cooled data centers, liquid-cooled data centers can reduce energy consumption ratios, saving around 30% in energy use. Liquid-cooled servers can operate at higher temperatures, further reducing reliance on additional cooling resources. In short, liquid cooling systems enable higher computing density and enhance overall energy efficiency.
Immersion Cooling: Pushing the Limits of Heat Dissipation Efficiency
Immersion cooling is a classic example of direct-contact liquid cooling, where heat-generating components are fully submerged in a coolant. The circulating liquid transfers the heat generated during operation. Since the heat-generating components are in direct contact with the coolant, this method delivers more efficient heat dissipation and, thanks to the absence of fans, lower noise levels. Depending on whether the coolant undergoes a phase change from liquid to gas, immersion cooling can be classified into single-phase and phase-change immersion cooling systems.
Although this setup requires each server to be equipped with an individual enclosure, thus occupying a certain amount of data center space, it integrates the server and coolant within the enclosure without altering the existing deployment architecture. Compared to traditional distributed cooling solutions, this design effectively increases equipment deployment density per unit of space.
Microfluidic Technology: A Chip-Level Precision Cooling Powerhouse
As a cutting-edge approach to applying liquid cooling at the chip level, microfluidic technology offers exciting new possibilities for chip thermal management. By integrating tiny fluid channels within the chip or its packaging, this technology delivers coolant directly near heat sources, thus enabling highly efficient heat transfer. What sets it apart is its ability to precisely control the heat transfer path, significantly reducing thermal resistance and effectively lowering the chip’s operating temperature.
At the forefront of microfluidic technology, 3D microfluidic liquid cooling systems embed microchannels directly inside the chip, allowing liquid to flow through and absorb and transfer heat. A new type of water-cooling system adopts a 3D microfluidic channel structure, incorporating capillary features and manifold distribution layers. This design achieves a coefficient of performance (COP) as high as 10⁵, far outperforming traditional cooling technologies. This research offers a new approach to tackling the thermal management challenges of high-power electronic devices.
Avnet: A Pioneer in Liquid Cooling
As AI computing power continues to rise, related technologies are increasingly turning to liquid cooling as the next strategic step for controlling processor temperatures. Liquid cooling has already been validated at the packaging level. Experimental GPUs have demonstrated that liquid-cooled integrated heat sinks, featuring microfluidic channels that are circulated by microfluidic pumps, can dissipate heat densities of up to 790 W/cm².
At every stage of their customers’ liquid cooling projects, from research and development to production and market entry, Avnet provides unparalleled professional support. The global technology solutions provider goes to great lengths to understand the specific requirements of each of its customers’ projects, including their underlying reasons for adopting liquid cooling technology, before recommending a tailored solution. Whether it’s to achieve more efficient energy use, cope with harsh environmental conditions, reduce noise levels, save space, or more effectively cool faster and hotter-running chips, Avnet has the solution.
Cooling the “Fever” of Computing Power
In the near future, the explosive growth of artificial intelligence will see more and more data centers established around the world. This trend poses significant challenges for data centers with regard to power supply and thermal management. To ensure stable and efficient performance, data centers must adopt effective measures to control equipment temperatures. Not surprisingly, industry research and development is focused on advances in thermal management technologies. From chip-level microfluidic cooling to server liquid cooling solutions employing direct-to-chip and immersion cooling methods, related technologies are continually being developed, tested, and applied. Liquid cooling is expected to become a key technology in supporting the rapid development of artificial intelligence, while paving the way for the proliferation of data centers in the new era of AI.
