Sustainable Artificial Intelligence: Power Through Computility–Electricity–Carbon Coordination

Artificial intelligence (AI) has soared to the top of the research area, business agenda, and people’s daily life, accelerating the construction of data centers and driving demand for more high-energy-intensity facilities. The energy-intensive operation of servers in data centers, along with the consequent carbon emissions, poses critical challenges to global energy sustainability. Specifically, data centers suffer from three issues from different aspects, namely, high power usage effectiveness (PUE), high electricity consumption, and high carbon emissions. Owing to the fact that data centers have temporal–spatial operation flexibility, the interaction between data centers and renewable electric power systems can be a promising way to address the three above issues and alleviate the dilemma of balance between AI development and the achievement of energy sustainability. To this end, a novel framework of computility–electricity–carbon coordination (CECC) for data centers is proposed in this article. Meanwhile, key technologies for CECC are discussed, such as computing workload modeling, computility–electricity codispatching strategy, carbon emission tracking of computing power, and so on. Some practical CECC actions around the world are presented, and potential challenges are also discussed.