In 2026, while the demand for AI computing power is growing exponentially, its energy consumption and carbon emissions are also receiving increasing attention. According to the latest report from the International Energy Agency (IEA), global data center electricity consumption accounts for 2-3% of the world's total electricity generation, with AI training and inference being the fastest-growing part. In this context, the concept of "green AI" has emerged as a key proposition for the sustainable development of the AI industry.
At the hardware level, the energy efficiency of AI chips is rapidly improving. By 2026, AI chips based on 3nm and 2nm processes will be mass-produced on a large scale, with a single watt performance improvement of 40-60% compared to the previous generation. More noteworthy is that emerging technologies such as integrated storage and computing architecture and photon computing are moving from the laboratory to industrialization, with the potential to reduce AI inference energy consumption by over 90%. Nvidia, AMD, Huawei and other manufacturers have all released a new generation of low-power AI acceleration cards that support dynamic power management and intelligent sleep strategies.
At the software level, optimizing model efficiency has become an industry consensus. By combining model quantification (INT4/INT8), knowledge distillation, pruning, and other techniques, large models can compress the parameter count to 1/5 of its original size while maintaining over 90% performance, and increase inference speed by 3-5 times. The popularity of MoE (Mixed Expert) architecture has also significantly reduced the proportion of activation parameters for each inference, from 100% to 10-20%.
In addition, more and more AI companies are beginning to commit to carbon neutrality goals by offsetting the carbon emissions from AI operations through purchasing green electricity, participating in carbon trading, and investing in forest carbon sinks. Google, Microsoft, Amazon and other tech giants have announced the goal of achieving operational carbon neutrality by 2030, and Chinese tech companies are also accelerating their follow-up. The combination of AI and sustainable development is not only a technological proposition, but also an industrial responsibility.
[Reference source] This article is a comprehensive compilation of information publicly released by the industry.