PFAS FLOWS, TREATMENT AND REDUCTION IN AI INFRASTRUCTURE
Access to this document is restricted. Some items have been embargoed at the request of the author, but will be made publicly available after the "No Access Until" date.
During the embargo period, you may request access to the item by clicking the link to the restricted file(s) and completing the request form. If we have contact information for a Cornell author, we will contact the author and request permission to provide access. If we do not have contact information for a Cornell author, or the author denies or does not respond to our inquiry, we will not be able to provide access. For more information, review our policies for restricted content.
The rapid expansion of generative artificial intelligence (AI) is reshaping semiconductor manufacturing and data center infrastructure, driving increasing reliance on per- and polyfluoroalkyl substances (PFAS) and raising growing concerns over their environmental and health impacts. In this work, we develop a unified framework to quantify PFAS-related burdens and mitigation trade-offs across key components of AI-driven infrastructure. We first examine PFAS emissions from electronics and semiconductor manufacturing and show that mitigation strategies introduce inherent trade-offs between health risk reduction and associated environmental and economic burdens, establishing a cost–health nexus that underpins system-level decision-making.Building on this foundation, we analyze PFAS use in photolithography processes supporting generative AI, where consumption is projected to reach 21.0–73.1 kilotons between 2025 and 2030, with anti-reflective coatings as the dominant application. Using a functional unit defined per kilogram of PFAS managed by waste medium and scaling to projected demand, we quantify cumulative impacts of 622.0 kilotons of CO2 emissions and $1109.8 million in costs under aggressive scenarios. These burdens are strongly influenced by PFAS partitioning across waste streams and removal efficiency. In parallel, we investigate PFAS use in data center cooling systems under increasing power density. While PFAS-based two-phase cooling significantly reduces electricity demand compared to single-phase systems (181.7 ± 28.8 TWh vs. 608.5 ± 61.8 TWh), it introduces higher treatment burdens, revealing a fundamental trade-off between energy efficiency and environmental impacts. We further characterize these trade-offs across operating conditions and develop a PUE-based decision framework to guide technology selection. Across these domains, we evaluate mitigation strategies including improved separation and destruction, source control, cascade utilization, and coolant recycling and redesign, achieving up to 35.9% reductions in global warming potential and up to 81% reductions in coolant demand. By integrating PFAS flows across manufacturing, process-level use, and end-use systems, this work provides a quantitative basis for understanding how PFAS-related risks propagate across AI infrastructure and for balancing performance, cost, and environmental impacts.