The rise of generative AI models, such as ChatGPT, has ushered in a new era of possibilities. These models, fueled by deep learning and trained on massive datasets, can perform complex tasks in mere seconds, from crafting term papers to generating unique artwork. However, with great power comes great energy consumption, and the energy demands of AI are raising concerns about their environmental impact.
The soaring growth of AI computing demand
Alphabet’s John Hennessy estimates that creating AI-generated content like artwork can require ten times the computing power of a standard internet image search. As AI continues to grow in popularity and accessibility, the collective energy demand generated by 5.3 billion internet users harnessing AI capabilities is a topic of intense debate. The projections and opinions surrounding the environmental impact of AI are wide-ranging, with some fearing it could lead to climate destruction while others envision a smoother transition to clean energy.
Data centers: The heart of the AI revolution
Data centers play a pivotal role in AI’s energy consumption. These massive facilities, often referred to as “hyper-scale” data centers, serve as the backbone of the internet, processing and storing vast amounts of digital information. Understanding the implications of increased computing demand on electricity consumption requires a closer look at data centers, which nearly all of the world’s internet traffic passes through.
Current data center energy consumption
In 2021, data centers are estimated to have consumed between 220 to 330 Terawatt hours (TWh) of electricity, according to the International Energy Agency (IEA). To put this into perspective, the entire state of California’s annual electricity consumption is around 278 TWhs. These figures raise questions about how rapidly expanding AI applications might further impact data center electricity demand.
Estimates and projections vary widely
Estimates regarding the future electricity consumption driven by AI vary significantly. Some predict that AI could consume the entire global energy supply, while others anticipate that US data center demand may more than double by 2030. Such projections are alarming, given the broader challenges of decarbonization. However, historical data suggests that past forecasts may have overestimated data center electricity use.
A crucial factor in understanding data center efficiency is “power usage effectiveness” (PUE). This metric measures how effectively data centers convert electricity into computing output. Surprisingly, global data center energy use increased by just 6% between 2006 and 2018, despite a sixfold increase in computing output and a twenty-fivefold increase in storage capacity. This remarkable efficiency improvement was attributed to advancements in PUE.
PUE efficiency gains and recent stagnation
Efficiency gains in PUE before 2019 were remarkable, but recent years have seen these improvements stall. If efficiency gains remain stagnant while computing demand surges, a significant energy consumption problem may arise. This potential challenge underscores the importance of balancing the growth of data centers with energy efficiency and sustainability.
Despite concerns, there is an optimistic perspective that data centers can contribute to climate solutions. Barry Fischer, Google’s “Data Storyteller,” highlights the potential for data centers to procure their renewable energy, meet their electricity needs, and provide flexibility to the grid. Additionally, AI applications supported by data centers can optimize and accelerate the clean energy transition.
Tensions between decarbonization and data center growth
Already, tensions are emerging between decarbonization goals and the increasing electricity demand of data centers. Examples include new data centers in Virginia driving unexpected electricity demand, potential delays in coal plant retirements to accommodate data centers in Wisconsin, and applications for exempting backup diesel generators from environmental regulations in Maryland. These situations highlight the challenges posed by data center growth to clean energy efforts.
The need for data on data center energy use
Efforts to address data center energy consumption are hindered by a lack of reliable information on data center energy use. Collecting accurate data on data centers is essential to anticipate, plan for, monitor, and manage their environmental impacts. Currently, data centers’ energy data remains fragmented and incomplete.
While debates continue about the environmental impact of AI and data centers, the necessary policy tools to address these challenges are already available. Collecting accurate and comprehensive data on data center energy consumption is a crucial first step. Europe has taken the lead by establishing detailed reporting requirements for data centers’ energy performance and sustainability, set to take effect next year. Similar legislation is being considered in the United States.
The rapid growth of AI and data centers presents both promise and peril. As AI’s accessibility expands, its energy consumption becomes a pressing concern. To ensure responsible growth, policymakers, industry stakeholders, and researchers must work together to collect accurate data, address energy efficiency, and balance the potential of AI with environmental responsibility. In this evolving landscape, data centers can play a pivotal role in shaping a sustainable and efficient future.
