Liquid versus Air: Life Cycle Carbon of Cooling Down AI Data Centers

Abstract

The rise of AI and cryptocurrency technologies has led to exponential growth in data centers globally. As these facilities consume substantial energy and contribute to significant greenhouse gas emissions, implementing sustainable cooling technologies in data center facilities is increasingly crucial. Preliminary findings suggested that applying liquid cooling systems within these data centers holds significant potential to markedly decrease energy consumption, greenhouse gas emissions, and water usage (Manganelli et al., 2021). Initial estimations indicate that data center facilities incorporating liquid cooling mechanisms could reduce both embodied and operational carbon emissions by up to 50% throughout their life cycle compared to those employing conventional air cooling methods. This adaptation could result in near-zero water usage and decrease the physical size of data center buildings, further reducing carbon emissions and capital expenditure, potentially achieving cost neutrality over the facility’s life cycle. I primarily sought to answer three questions: How do liquid and conventional air cooling methods compare in terms of carbon emissions when evaluated through a comprehensive LCA? How does adopting liquid cooling influence data center facility design and resource utilization? Considering both upfront and operational costs, what are the cost implications of transitioning from traditional air cooling to liquid cooling? I hypothesized that data centers with liquid cooling systems will demonstrate substantial reductions in embodied and operational carbon emissions and water use, and despite higher upfront costs, liquid cooling systems will prove cost-neutral over their life cycle. To test these hypotheses, I analyzed the environmental and cost implications of data center cooling technologies through a comparative life cycle assessment (LCA) of liquid and traditional air cooling systems. The LCA adhered to the ISO standards (14040, 14044, 14067, and 15686), encompassing all stages of the data center's life cycle, reporting Global Warming Potential in kgCO2e. Material inventory data from equipment manufacturers and data center operators were aggregated and anonymized to create representative models of data centers using both cooling methods. The liquid cooling LCA models achieved up to 50% reductions in embodied and operational carbon emissions compared to conventional air cooling systems. Liquid cooling also enabled near zero water usage and a 50% smaller data center building size, which resulted in embodied carbon reductions and less resource use throughout the life cycle. My research findings present a compelling case for approaching data center cooling solutions with a holistic life cycle perspective, identifying the most environmentally sustainable and cost-effective cooling method, and providing recommendations for reducing data centers' environmental impact. The outcomes should be significant for data center operators, designers, and construction managers, enabling more sustainable and cost-effective design strategies for future data centers.