Current Status of Energy Demand for AI Infrastructure
In recent years, the rapid development of AI technology has driven the construction and upgrading of data centers, leading to a significant increase in their energy demand. According to the International Energy Agency (IEA), in 2022, global artificial intelligence, data centers, and cryptocurrencies consumed approximately 460 billion kilowatt-hours of electricity, accounting for about 2% of global total electricity demand. Of this, AI's electricity demand primarily comes from model training and inference, while data center electricity demand mainly comes from computing (40%), cooling (40%), and other related IT equipment (20%).
With the increasing prevalence of generative AI, the energy demand for data centers is further increasing. Deloitte predicts that by 2026, the power required for key components of global data centers (including GPU and CPU servers, storage systems, cooling equipment, and network switches) will nearly double to 96 gigawatts (GW), with AI computing alone potentially consuming over 40% of that power. Furthermore, by 2026, the annual power consumption of global AI data centers is expected to reach 90 terawatt-hours (TWh), approximately ten times higher than in 2022.
Drivers of AI-driven energy demand growth
Increased model complexity
The number of parameters in AI models is constantly increasing, from the initial millions to hundreds of billions or even trillions today. This has led to a significant increase in the computing resources required for model training and inference. For example, ChatGPT processes more than 200 million requests per day, consuming up to 500,000 kilowatt-hours of electricity per day, resulting in an annual electricity cost of 200 million yuan.
Centralization of data centers
The construction and operation of hyperscale data centers concentrate a large amount of computing resources, and their high energy consumption around the clock puts enormous pressure on existing power infrastructure. In addition, the cooling system of the data center is also one of the major energy consumers, accounting for 40% of the total energy consumption.
The widespread adoption of AI applications
The application of AI technology continues to expand across various industries, from intelligent transportation and industrial automation to fintech, all of which are placing higher demands on AI infrastructure. This has not only increased the number and scale of data centers but also further driven up energy consumption.
Challenges
Sustainable energy supply
The energy demand for AI may grow faster than the supply of clean energy. According to the World Economic Forum, by 2027, AI-related electricity consumption will reach 134 terawatt-hours (TWh), of which approximately 83.08 TWh will still be supplied by fossil fuels, potentially leading to nearly 40 million tons of carbon dioxide emissions annually. This supply-demand imbalance poses a serious challenge to global climate goals.
Bottlenecks in improving energy efficiency
Despite continuous improvements in AI technology, overall energy consumption continues to rise. For example, the Jevons Paradox states that increased technological efficiency can actually lead to increased overall energy use. Furthermore, the energy consumption of AI chips (such as GPUs and ASICs) remains a challenge; how to reduce energy consumption while maintaining computing performance is a key objective.
Pressure on power infrastructure
The centralization and high energy consumption of data centers pose significant challenges to existing power infrastructure. Many regions face issues such as grid connection delays and complex licensing processes, hindering the rapid adoption of clean energy.
Response strategies and technological innovation
Hardware optimization
Hardware manufacturers are improving energy efficiency by optimizing chip design and developing dedicated chips (such as NPUs and TPUs). For example, DeepSeek significantly improves energy efficiency by reducing the energy consumption per query by 90% through reduced computing power consumption.
Energy management technology
AI technology itself is also used to optimize energy management. For example, smart grids use AI technology to achieve functions such as accurate electricity demand forecasting, real-time monitoring and fault diagnosis, and intelligent energy dispatch, thereby improving energy efficiency.
Application of clean energy
Many tech companies are increasing their investments in renewable energy to meet the energy demands of data centers. For example, BlackRock plans to partner with Microsoft to launch a more than $30 billion AI investment fund to fund data center and clean energy projects.
Policy and market synergy
Governments and industries need to work together to address the growing energy demand driven by AI. For example, China has included "AI+" in its government work report for the first time, promoting the accelerated penetration of AI applications across industries, while simultaneously planning investment in power infrastructure.
Future Outlook
While the increasing energy demands of AI present numerous challenges, sustainable development is expected in the future through technological innovation and policy support. For example, breakthroughs have been achieved in the research and development of nuclear fusion energy, with the prospect of nuclear fusion power plants becoming operational by 2050. Furthermore, the widespread adoption of AI on mobile devices could reduce cloud computing power requirements by more than 30%, thereby alleviating energy pressures on data centers.
Summarize
Energy demand for artificial intelligence (AI) infrastructure is one of the major challenges facing the world today. With the widespread adoption of AI technology and the centralization of data centers, the rapid growth in energy demand poses a severe test to sustainable development. However, through hardware optimization, innovation in energy management technologies, the application of clean energy, and policy support, sustainable development of AI infrastructure is expected to be achieved in the future. Global technology companies and governments need to work together to promote the deep integration of AI technology and energy management to address future energy demand challenges.
