This is the 179th in the ‘China Chronicles’ series.
In June 2025, Sam Altman, the Chief Executive Officer of OpenAI and a leading figure in the field of Artificial Intelligence (AI), stated in a blog post that in the 2030s, intelligence and energy would become wildly abundant, and that with these resources, we could have anything we want. What this statement conceals is more important than what it reveals. Data centres (DCs) that are critical to AI applications consume rather than make available abundant quantities of energy. For example, standard information searches on the internet consume about 0.3 watt-hours (Wh) of electricity, while queries that use generative AI use more than 3 Wh, which is over 30 times that of a standard search. China has one of the largest DC capacities in the world, which has increased China’s electricity consumption and its carbon emissions. Given the competition between the United States and China in the context of AI, China has probably built too many DCs too fast, as it has in most other sectors, from steel to solar energy. However, technology has upended the competition with energy efficiency of AI applications rather than DC capacity as the factor that would decide the winner: the lower the watts behind the bytes, the greater the chance of winning.
Technology has upended the competition with energy efficiency of AI applications rather than DC capacity as the factor that would decide the winner: the lower the watts behind the bytes, the greater the chance of winning.
Two of China’s goals are significant in the context of growth in DC capacity. One is China’s deliberate competitive digital strategy that has designated data as a new factor of production. China’s market size, its capacity for developing infrastructures like DCs, and increasing availability of reliable and inexpensive energy sources give it a structural edge. However, increasing the inexpensive energy supply has compromised China’s other significant target of reducing carbon emissions.
Between 2015 and 2024, electricity demand in China grew at 15 percent per year, more than twice the rate seen between 2005 and 2015, attributed to growing electricity consumption by DCs. Estimation of electricity consumption by DCs in China varies by source. Western sources such as the International Energy Agency (IEA) put electricity consumption by DCs in China at just over 100 TWh (terawatt hours) in 2024, which translates into 25 percent of global DC electricity consumption. According to Chinese sources, the electricity consumption of its DCs was over 150 TWh in 2020, which accounted for over 2 percent of the country’s total electricity consumption. If so, electricity consumption in 2025 could be equal to or greater than that of the US, the market leader. The share of electricity consumption by DCs in China is relatively small today, but it is expected to grow faster than demand from other new sources, such as electric vehicles, as AI inference generation becomes more complex.
China’s quest for dominance of the data economy has taken precedence over environmental goals.
China began planning for energy-efficient DCs more than a decade ago. Starting with the 12th five-year plan in 2011, China promoted the construction of energy-efficient DCs. China’s energy and technology institutions have issued a series of guidelines and regulations to ensure that DC construction keeps up with global standards in energy efficiency and carbon dioxide (CO2) emissions. Notwithstanding meticulous plans for efficient and low-carbon DCs, most of the electricity consumed by China’s DCs is produced from fossil fuels, and projections suggest an increase in DC electricity demand will most likely be met with fossil fuels by 2030. Among many reasons is the fact that China’s quest for dominance of the data economy has taken precedence over environmental goals. Other structural reasons include the location of DCs in the east of China, where electricity supply is dominated by coal with a share of over 70 percent. China initiated the East–West computing resources transmission project (EWCRT Project) in 2022 to address imbalances in renewable energy generation and consumption. Most of the renewable electricity is generated in China’s west, while DCs are located in the east. Completion of the project is expected to substantially increase renewable energy use in DCs and consequently reduce carbon emissions.
China has also sought to achieve an average power usage effectiveness (PUE), a metric for energy efficiency (ratio between the power consumption of the whole facility against the consumption of the data equipment), of less than 1.5 in its DCs by 2025. According to the IEA, PUE of China’s DCs was 1.56 in 2023 compared to a global average of 1.43 in 2023. Cooling accounts for more than 40 percent of the total energy consumption of a DC. DCs in the north, where the average atmospheric temperature is less than 20°C (Celsius), have achieved a PUE of 1.3 as they reduce the need for external cooling.
DeepSeek’s technology has cast doubt on China’s assumption that an endless expansion of DCs will guarantee success.
When ChatGPT was launched in 2022 by the US company OpenAI, China responded quickly, accelerating the construction of AI-focused DCs. In 2023 and 2024, more than 500 DC projects were announced nationwide, and at least 150 were completed by the end of 2024. However, nearly 80 percent of this new computing capacity is reportedly lying idle. Ironically, among the reasons for this is the launch of R1, an AI application by the Chinese company DeepSeek, whose open-source reasoning model matched the performance of ChatGPT at a fraction of its energy and economic cost. Whereas ChatGPT-4 was trained using 25,000 GPUs (graphics processing units) and the US company Meta’s Llama 3.1 used 16,000, DeepSeek used just 2,000 chips that required far less energy. While specific numbers vary, estimates suggest DeepSeek's architecture and fewer parameters allow it to consume up to 90 percent less energy for similar performance, challenging the resource-intensive model of its rivals. DeepSeek’s technology has cast doubt on China’s assumption that an endless expansion of DCs will guarantee success. When energy efficiency is at the heart of the battle over dominance in AI, consumers and the environment will be the real winners.
Lydia Powell is a Distinguished Fellow at the Observer Research Foundation.
The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.
Ms Powell has been with the ORF Centre for Resources Management for over eight years working on policy issues in Energy and Climate Change. Her ...
Read More +
PREV
