The United States is experiencing a booming demand for generative AI models, despite the threat the hot technology poses for the planet should it remain reliant on fossil fuels. A study recently published in Nature Sustainability explores potential net-zero pathways to mitigate the environmental burden of artificial intelligence servers in the U.S. According to the study by Xiao, Nerini, et al., generative AI server deployment in the US is largely unmitigated, creating a collision course with climate goals. Through the use of bottleneck-based modelling, the authors predict that the environmental burden of AI servers could reach a 44 Mt CO2-equivalent and a water footprint exceeding 1 billion cubic meters by 2030. Those in favor of the AI boom often point towards the technology’s ability to aid in the discovery of climate solutions; however, the study finds that the scale of AI’s computational demand is greatly outpacing the rate of U.S. grid de-carbonization. This makes corporate net-zero goals and promises largely unattainable without “highly uncertain carbon offset and water restoration mechanisms.” The roadmap provided by the study prioritizes optimal siting of servers in regions like the Midwest that experience low-carbon and low-water-stress, instead of the current trend of building data centers in water-scarce regions like Nevada and Arizona. When combined with accelerated grid de-carbonization and key technologies like advanced liquid cooling, a sustainable and competitive foundation for future AI growth may not be out of reach. Following the outlined environmental incentives also aligns with for-profit interests, as placing facilities in regions rich in renewable energy and water protects operations from the volatile utility costs commonly associated with resource scarcity. By following the roadmap outlined in the study, companies can lower their operating expenses in the long term, while simultaneously creating an environmentally-friendly, future-proof foundation for their AI investments.