How to deal with AI’s “power consumption”?

The ultimate goal of AI is to achieve efficiency leaps and intelligent transformation in various industries through the combination of algorithms and data. However, every breakthrough in AI is inseparable from the processing of massive data and the training of complex models, and the core support of these processes is "computing power".

 

"Computing power" is the core driving force for the implementation of AI. The rapid development of AI has led to an exponential growth in the demand for computing power, and the problem of energy consumption has also become prominent. The power consumption of data centers, the core infrastructure of computing power, has surged.

 

According to the International Energy Agency's forecast, from 2025 to 2027, China's data center and 5G network power consumption will grow rapidly, and the proportion of data center power consumption is expected to increase from the current 3% to about 6%.

 

AI's huge consumption of electricity is described as "eating electricity". Chu Pan, an expert member of the Energy Storage Application Branch of the China Chemical and Physical Power Industry Association, explained to the reporter of Securities Daily that compared with traditional data centers, the power of a single cabinet in an AI data center has increased by 5 to 8 times, which has led to a 5 to 8 times increase in the demand for electricity in data centers of the same building size. Chu Pan believes that with the popularization of open source LLM large models (large language models) such as DeepSeek, there will be more and more AI computing centers.

 

 

Guosheng Securities Co., Ltd. recently released a research report saying that from a global perspective, overseas technology giants have continued to increase AI computing power, and it is expected that the demand for AI infrastructure will increase significantly, and data centers will usher in a new cycle. According to SemiAnalysis data, the total demand for key IT power in data centers will double from 49GW in 2023 to 96GW in 2026, of which 90% of the growth comes from AI-related demand.

 

The rapid advancement of AI computing power infrastructure has also put forward higher requirements for the stable operation of data centers. Specifically, on the one hand, the power demand of data centers is volatile, especially in computing-intensive tasks, power consumption surges; on the other hand, data centers usually require 24-hour uninterrupted power supply. As a result, data centers have an increasing demand for power reliability.

 

In this regard, Chu Pan said that in order to cope with the diversified power needs of AI data centers, not only does the power system need to provide low-cost and reliable power, but the power system also needs to improve the quality of power. The unit price of AI chips is high, and unqualified electricity will increase the probability of damage to AI chips, reduce the service life of AI chips, and thus affect the normal operation of data centers.

 

Under the development trend of global energy transformation, data centers are also actively exploring the path of green, low-carbon and high-quality development, which puts higher requirements on the cleanliness and sustainability of power supply.

 

Contact now