Environmental impact and net-zero pathways for sustainable artificial intelligence servers in the USA

A new study projects that U.S. AI server deployment will generate an annual water footprint of 731-1,125 million m³ and 24-44 Mt CO2-equivalent emissions by 2030, making the industry's net-zero aspirations by 2030 largely unattainable without substantial, uncertain carbon offsets and water restoration. While best practices in efficiency, optimal spatial distribution (e.g., Midwestern states with abundant renewables), and grid decarbonization could reduce these impacts by up to 73% for carbon and 86% for water, significant infrastructure challenges in optimal regions and the rebound effect from efficiency gains pose considerable risks. This highlights critical long-term operational costs, potential regulatory scrutiny, and the need for strategic investments in green infrastructure and transparent sustainability measures for AI companies.

Analysis

The rapid expansion of AI servers in the U.S. is projected to generate significant environmental impacts, with annual water footprints ranging from 731 to 1,125 million m³ and carbon emissions from 24 to 44 Mt CO2-equivalent between 2024 and 2030. This scale suggests the AI server industry is unlikely to achieve net-zero aspirations by 2030 without substantial reliance on highly uncertain carbon offset and water restoration mechanisms. The highest impact scenario largely surpasses previous forecasts for the entire U.S. data-center market, underscoring environmental risks.

While best practices in industry efficiency, optimal spatial distribution, and grid decarbonization could reduce carbon emissions by up to 73% and water footprints by 86%, achieving these reductions faces significant hurdles. Optimal locations, such as Midwestern states like Texas, Montana, Nebraska, and South Dakota, offer abundant renewables but require substantial investment in new renewable capacity and transmission infrastructure. Texas alone may need to support an additional 74–178 TWh of AI server demand, potentially exceeding its current total renewable generation.

Efficiency gains in AI hardware and software risk a "rebound effect," where lower costs per computing task lead to increased overall application volume and higher total demand. This dynamic could amplify total environmental footprints, complicating the path to sustainability. Unmodelled uncertainties like supply-chain bottlenecks beyond CoWoS technology and geopolitical influences could also significantly alter projections.

The study highlights the need for strategic coordination between private actors and regulatory interventions for sustainable AI sector development. Challenges in deploying AI servers at optimal locations and achieving ideal efficiencies suggest AI companies will increasingly rely on offset mechanisms, raising concerns about long-term contract security and verification. This implies potential for increased regulatory scrutiny and a shift towards more transparent, verifiable sustainability practices.

Key Decisions for Investors

• Evaluate AI-exposed companies' long-term sustainability strategies, focusing on verifiable net-zero pathways beyond reliance on uncertain carbon offsets.
• Monitor regulatory developments and potential policy interventions aimed at mitigating AI's environmental footprint, particularly concerning data center siting and energy procurement.
• Assess infrastructure investment opportunities in renewable energy and transmission in optimal Midwestern states, which are critical for sustainable AI growth.
• Consider the "rebound effect" risk when evaluating efficiency-driven growth projections for AI companies, as increased demand could offset per-unit environmental gains.