AI’s Hot Pursuit: Cooling Down the Environmental Impact of Data Centers

With the exponential rise of artificial intelligence (AI) and machine learning, entire industries are being transformed—helping boost productivity and reshaping how business is done and services are delivered.

However, this progress casts a long shadow, with data centers driving up energy demand and greenhouse gas emissions. Recently, several large tech companies reported increases in their emissions—attributing it in part to AI and data center electricity usage.

The concern over AI’s environmental and energy impacts has fueled headlines such as The New York Times’ 'A.I.’s Insatiable Appetite for Energy' and Bloomberg’s ‘Microsoft Wanted to be Carbon Negative. Then It Went Big on AI.’ While concerns over meeting the power demand and addressing emissions are critical challenges to address, no conversation on the future of AI is complete without considering strategies to reduce energy use in data centers—and that includes rethinking the cooling systems.

Data centers not only need large amounts of electricity for computing power but also for the necessary cooling to manage the enormous amount of heat generated by the hardware. At the same time, data centers consume large amounts of water and require large swaths of land. In the face of these challenges, there is an opportunity to move toward a new innovative system architecture known as two-phase immersion cooling (2-PIC), a potentially game-changing solution to make data centers more energy efficient. The process involves immersing electronic components in a dielectric heat transfer fluid within a hermetically sealed tank. The liquid boils from the server heat and the vapor that forms removes the heat, condenses, and returns to liquid form—creating a circular system.

Traditionally, data centers rely on air- and water-cooling technologies. While effective, this method can be inefficient and uses large amounts of energy. In fact, about 40% of data center energy use is generated by these traditional technologies.

2-PIC relies on low global warming potential fluorinated gases, or F-gases, which enable industry to cut emissions, increase system life, and reduce waste, as they can be recovered, recycled, and reclaimed at their end-of-life. Compared to air-cooled systems, this technology can significantly reduce cooling energy consumption in data centers by up to 90%, which equates to a 40% reduction in total data center energy consumption. Moreover, this technology can significantly decrease water use and reduce the physical size of data centers by up to 60%--helping to address other resource concerns associated with data centers.

As more companies and industries embrace AI, the energy and environmental concerns surrounding data centers will only be magnified. Some estimate that data center power demand will increase 160% by 2030, while companies like Google and Microsoft have acknowledged that AI poses a challenge to meeting their climate pledges.

At the same time, we must acknowledge the profound influence of AI across the economy. Its effects are far-reaching, contributing to enhanced patient care in healthcare, supporting risk management and fraud detection in finance, and boosting crop yields within agriculture, among other advancements, yet we've merely begun to uncover the full extent of its potential. The momentum behind AI will continue to grow, and this presents us with a unique opportunity to further its progress while employing technologies that reduce energy consumption and address the environmental impact of data centers. Although 2-PIC may not offer a complete answer, it represents a significant stride toward a sustainable solution for the future of data centers and AI.

Izabela Jasinska is the Immersion Cooling Venture Leader, Thermal & Specialized Solutions at Chemours. She has 20 years of chemical industry expertise, with a proven track record of delivering strong financial performance, leading business transformation, and developing purpose-driven strategies centered on innovation