White Paper


For more information, please get the full access of the white paper. 

White Paper on Global Artificial Intelligence Environmental Impact


Escalating Environmental Impact of AI and Data Centers

As the demand for AI models and data centers continues to surge, the associated environmental impacts are becoming increasingly significant. Data centers now consume around 1-2% of the world's electricity, a figure that is expected to grow alongside the expansion of cloud computing and AI technologies. The computational demands of machine learning (ML) models, particularly large-scale auto-regressive models, have also escalated significantly. For example, training large AI models such as GPT-3 can consume an immense amount of energy, equivalent to the yearly electricity usage of approximately 120 U.S. homes. This substantial energy consumption also results in a significant carbon footprint, with some large AI models emitting more than 626,000 pounds of CO2 during training, which is nearly five times the lifetime emissions of an average American car. Additionally, data centers contribute to environmental strain through their extensive water usage, necessary for cooling, often consuming hundreds of thousands of gallons daily. The rapid pace of hardware obsolescence further exacerbates environmental concerns by contributing to electronic waste, while the heat emissions from these centers can lead to local temperature increases and additional energy consumption for cooling. Additionally, The environmental impact of ML encompasses not only computing-related impacts, like computations and hardware production, but also immediate application impacts from deploying ML in various sectors like power system, transportation, optimizing processes or substituting traditional methods, as well as system-level impacts, which reflect broader changes in societal and infrastructural systems. 


Regulatory Gaps in Addressing Environmental Challenges

Given the urgency and significance of the matter, assessing and regulating these environmental impacts has become crucial.  Although there are existing efforts in the industry, academia, research, and regulation, they often have limitations such as the absence of specific guidelines for stakeholders to follow or the lack of consideration for certain aspects of the full scope or life cycle of environmental impacts. For instance, many existing frameworks focus on isolated parts of the energy usage or emissions from data centers, without accounting for the broader supply chain impacts or the carbon intensity of energy sources at various stages. Furthermore, regulatory efforts tend to lag behind technological advancements, leaving gaps in enforcement or failing to address emerging environmental concerns, such as the indirect effects of increased AI deployment on energy consumption. The lack of harmonization between global and regional regulations further complicates the ability to form a cohesive, effective response to the environmental challenges posed by the industry.


Introducing the AI Green Index for Holistic Environmental Assessment

This paper provides in-depth insights into the assessment and regulation of the environmental impacts of AI and data centers. We propose the AI Green Index, a novel evaluation framework to comprehensively assess both the carbon footprint and water footprint of AI models and data centers. The need for the Green AI Index stems from the lack of standardized, comprehensive tools to evaluate the environmental costs associated with the rapid expansion of AI technologies and data centers. By implementing this index, stakeholders can better understand and mitigate the environmental impact across the entire life cycle of AI applications. The objectives of the Green AI Index are to provide a unified and quantifiable metric and a step-by-step guideline for assessing the environmental impact of AI models and data centers. We hope that through the Green AI Index, we can promote transparency, encourage more sustainable practices, incentivize carbon-efficient innovations, and guide policymakers, organizations, and developers in reducing their environmental footprint.


Comparative Analysis of Regional AI Environmental Policies

Additionally, we review the latest AI and data center environment-related policies in major regions, including China, the United States, and the European Union. We conduct an in-depth analysis of the key points of each regulation and compare them, while also providing insights into the strengths and weaknesses of each region in AI and data center environmental governance.


Collaborative Path Forward for Sustainable AI Development

We hope that this white paper will assist academic researchers, industry professionals, and policymakers by fostering collaboration and building bridges between different sectors to promote sustainable development in the AI industry.