From the self-built power stations established by American tech companies to the energy efficiency plans for data centers proposed by the European Union, the “energy consumption bill” behind AI has been placed on the discussion table of energy and technology policies in many countries. What challenges does the AI energy consumption issue pose to global energy? Can “green computing power” be the solution? How can the international community better cooperate to address these issues? In response to these questions, our newspaper recently invited experts to provide interpretations.
How much power does AI consume?
“Electric Tiger”, “Energy-Eater Beast”, “Power Hog”… These are the nicknames given by many Chinese and foreign media to AI data centers recently. Just how much power does AI consume?
A set of data provides the answer. The report “Energy and Artificial Intelligence” by the International Energy Agency (IEA) shows that data centers, as the core carriers for AI operations, have witnessed a rapid increase in their power consumption. In 2024, the global electricity consumption of data centers reached 415 terawatt-hours, accounting for 1.5% of the global total electricity consumption, which is comparable to the annual electricity consumption of the United Kingdom. As AI training and inference tasks surge, this figure is expected to reach approximately 945 terawatt-hours by 2030, approaching the annual electricity consumption of Japan.
Among them, the electricity consumption increase of data centers in developed economies is rapidly escalating. The IEA report indicates that from 2024 to 2030, the increase in electricity consumption of data centers in the United States will account for nearly 50% of the total increase in electricity demand in the country; by 2030, the electricity consumption for AI data processing in the United States will exceed the combined electricity consumption of traditional high-energy-consuming industries such as aluminum, steel, cement, and chemicals. During the same period, the increase in electricity consumption of data centers in Japan will account for more than 50% of the total growth in electricity demand in the country; the electricity consumption of data centers in the European Union is expected to reach 150 terawatt-hours in 2030, an increase of approximately three times the current scale.
Chai Qiming, the director of the Strategic Planning Department of the National Center for Climate Change Strategy and International Cooperation, told a reporter, “Compared with traditional computing models, AI large models have trillions of parameters, and the computing volume increases exponentially, consuming an enormous amount of power. Take the United States as an example. In the past 15 years, its power consumption has been basically stable. However, in the past 3 to 5 years, it has grown rapidly, largely due to the fact that it has deployed nearly half of the large data centers worldwide. The energy consumption and power consumption surge brought about by the rapid expansion of AI have already posed a real challenge to global energy supply.”
Wang Juan, a specially-appointed associate researcher at the National Engineering Laboratory for Big Data Analysis and Application Technology of Peking University, told a reporter that the peak power demand for training advanced AI models is expanding at a rate of 2.2 to 2.9 times per year. Currently, the energy consumption of AI has already shown an impact on the energy supply of many countries and regions. Morgan Stanley predicts that the cumulative power shortage of data centers in the United States from 2025 to 2028 will reach 47 gigawatts, which is equivalent to the total electricity consumption of 9 Miami or 15 Philadelphia. Insufficient power supply has become the core bottleneck restricting the expansion of AI computing power.
High energy consumption has also led to a series of problems such as carbon emissions. The IEA predicts that by 2035, the global carbon emissions from data centers will rise from 180 million tons in 2024 to 300 million tons. Although the total emissions are less than 1.5% of the total emissions of the energy industry, data centers have become one of the fastest-growing emission sources.
“At present, the issue of AI energy consumption has rapidly become a global concern. Its impact goes far beyond the technical and economic realm and has escalated into an important topic related to the global energy landscape and global climate governance,” Wang Juan said.
It is both “the electric tiger” and “the green giant”
The energy consumption of AI is huge, but it is not an “unsolvable problem”.
Experts point out that the pressure on energy consumption is driving the global energy system to accelerate the green and low-carbon transformation. Green computing power, which means that data centers achieve a low-carbon, efficient and sustainable development model throughout the entire process of providing computing services, is becoming the key path to resolving this challenge and is expected to drive AI data centers to transform from “electricity hogs” to “green giants” with low carbon and intelligence.
From a global perspective, major economies have, based on their own realities, been exploring diversified energy solutions. Wang Juan introduced that the European Union has, through policy guidance, established an energy-aware data center ecosystem. Currently, 90% of the electricity used in European data centers comes from renewable energy, and 70% of the operators have reached at least a 75% renewable energy or hourly carbon-free energy standard. In the United States, while relying on natural gas and nuclear energy to ensure the base load, tech giants are purchasing large amounts of renewable energy through long-term power purchase agreements and are exploring the restart of traditional nuclear power plants and investing in small modular reactors to supply power to data centers. Saudi Arabia is relying on desert photovoltaic power and green hydrogen storage to plan to build gigawatt-level zero-carbon AI factories and explore all-weather clean power supply solutions.
The power demand of AI is a complex systemic issue involving total volume, stability, latency and geographical distribution, which is forcing a comprehensive upgrade and strategic adjustment of global energy infrastructure. The IEA predicted in the “World Energy Outlook 2025” that in the next 10 years, more than 85% of the new capacity will still be concentrated in the United States, China and the European Union. Moreover, more than half of the ongoing data center projects are located or close to large cities with a population of over one million, posing new challenges for local power grids and regional coordination. In response to the challenge of AI energy consumption, “green computing power”, which integrates the supply of computing power and the consumption of clean energy with efficient energy-saving technologies, has become a solution of common concern for the international community, said Wang Juan.
In China, data centers are accelerating their green transformation, actively adopting renewable energy sources such as solar, wind, and hydropower, and significantly increasing the proportion of green electricity usage. From “East Computing – West Storage” to “Computing and Electricity Synergy”, China is building a national integrated computing power network to promote precise matching of computing load and green electricity in terms of time and space. At the “East Computing – West Storage” hub nodes, China has already achieved the goal of having the green electricity proportion in newly-built data centers exceed 80%.
“Computing infrastructure has a strong ‘lock-in effect’ – the IT equipment and related infrastructure in data centers are generally designed to last for 5 to 10 years. If no energy efficiency standards and green power ratios are set during the initial construction phase, the subsequent energy consumption and emissions will be locked in for a long time. That’s why China has clearly set the PUE (Power Usage Effectiveness) indicators and green power usage ratios during the planning stage, aiming to avoid the old path of ‘building first and then dealing with the problems later’ from the very beginning.” said Chai Qimin.
Chai Qimin pointed out that currently, AI and new energy are showing a mutually reinforcing and collaborative development relationship. On one hand, new energy can provide green and sustainable energy support for the development of AI. On the other hand, AI is also deeply empowering the transformation and upgrading of the energy industry.
“Renewable energy generation has the characteristics of intermittency and instability. Wind power and solar power are greatly affected by the weather and require massive computing power for precise prediction. In the future, as large-scale facilities such as deserts, gobi, desert areas, and offshore large-scale bases, as well as distributed renewable energy sources, are connected, the power and electricity balance across time and space and in multiple dimensions will become even more complex. Deep optimization and scheduling with the help of AI and IoT technologies are urgently needed. Moreover, AI is expected to assist in the research and development of cutting-edge energy technologies such as controllable nuclear fusion, through higher-precision simulations, significantly improving the research and development efficiency. From space-based photovoltaic power to space-based nuclear power plants and other zero-carbon energy innovation directions, they can better serve the layout of AI computing power space, and also cannot do without the support of AI technology.” Chai Qimin said, “The two are intertwined and together weave a vast future technological industry picture, injecting new impetus into economic and social development.”
Challenging yet full of opportunities
Fatiha Biroir, the director of the International Energy Agency, stated that the rise of AI has placed the energy industry at the forefront of the technological revolution. It brings not only challenges but also opportunities. Currently, how to balance AI innovation with energy sustainability and promote the deep integration and synergy of the two industries is an important issue that all countries need to jointly address and solve.
Wang Juan believes that, given the global nature of the energy challenges posed by AI, there is ample room for multilateral cooperation in the field of green computing power. For instance, in the area of cutting-edge technology research and development, all parties can collaborate in the most advanced fields such as space data centers by sharing large-scale scientific equipment and experimental data, jointly conducting feasibility studies and early technology validation, and sharing the high costs and risks of research and development. At the same time, in the field of standard setting and mutual recognition, all parties can enhance communication and collaboration within frameworks such as the International Electrotechnical Commission and the International Organization for Standardization to promote the comparison, mutual recognition, and transformation of green computing power-related standards, and establish globally recognized measurement scales and interface specifications. Moreover, in the field of coordinated scheduling of computing power and electricity, countries can promote the establishment of a regular coordination mechanism between energy and digital economy authorities at the institutional level, and develop unified data interfaces and collaborative optimization algorithms at the technical level to achieve real-time linkage between computing power load and green electricity supply.
Chai Qimin stated that as emerging industries such as AI shift to new energy on a large scale, the global energy landscape based on oil and gas may gradually change, leading to the migration of global energy centers and thereby triggering profound changes in the political security, industrial development, and investment flow patterns. This transformation process provides the international community with a historical opportunity to jointly explore new paradigms.
During this process, China has achieved a complete system from equipment manufacturing to technological innovation through the rapid development of the new energy sector. It has accumulated systematic advantages in the underlying energy field and can contribute valuable experience to the world. Chai Qiming said, “First, at the level of technological and economic synergy, the development model of combining AI with new energy in China is both innovative and feasible in engineering, and has obvious cost advantages. It can coordinate the relationship between development and transformation, promoting growth while controlling costs, and generating a positive spillover effect. Second, at the infrastructure construction level, practices such as ‘East Computing West Storage’ have demonstrated how to achieve coordinated development and green electricity through spatial optimization layout, providing a reference sample for the cross-regional allocation of computing power resources and energy resources worldwide. Third, at the level of pioneering research and development, in fields such as space energy and controlled nuclear fusion, which require joint investment from the global scientific community, China can play a greater role and work with the international community to break through the boundaries of human energy and computing power.”
“From a long-term perspective, the two global innovation processes of digitalization and greenification are interweaving and integrating, bringing new possibilities for the economic and social development of all countries. In the face of the common challenges brought by AI energy consumption, the international community is expected to achieve breakthroughs through joint cooperation, seize opportunities amid challenges, and jointly pave the way for the future of green computing power.” Chai Qimin said.