The load of a data center was divided into 2 tasks. One is a basic task which is related to the IP traffic, and another is an AI task which is not directly related to the traffic volume such as deep learning.
　As a result, the power consumption of the data center in 2018 was estimated to be 14 TWh in Japan and 190 TWh in the world. Assuming that the current trend of computational load increase continues in the future and the latest equipment currently available is used (not incorporating future technological progress), the power consumption will be 90 TWh (2030) and 12,000 TWh (2050) in Japan, and 3,000 TWh (2030) and 500,000 TWh (2050) in the world, respectively.
　Currently, servers account for 50% of data center power consumption, followed by power and cooling systems with 25-30% and storage systems with about 10%. Servers are expected to make up 60-80% of data center power consumption in the future. It is thus most important to reduce the power consumption of this particular component. CPUs and GPUs have great potential to reduce power consumption. The energy efficiency (Gflops/W) needs to be about 3 to 10 times higher than the current one in 2030, and 1,000 times higher in 2050. The power consumption of memory, power supply, and storage should also target to be 1/10 in 2030 and 1/1,000 in 2050.