In this report, we quantitatively studied the energy saving technologies for data centers, focusing on servers, which account for most of the power consumption, to evaluate technologies that can be used to suppress the increase of power consumption in the IT society. In particular, we focused on processors, which consume the largest amount of energy among the server devices, and then on memory and storage, for the case of small improvement (Modest Case) and large improvement (Optimistic Case).
　For the year 2030, in the Modest case, improvements in current technologies over the next five to ten years, particularly multi-core, miniaturization, and integration technologies in CPUs, while in the Optimistic case, developments in accelerator technologies were taken into consideration. As a result, the power consumption of data centers in 2030 was estimated to be 24 TWh in Japan and 670 TWh worldwide in the Modest case, and 6TWh in Japan and 190 TWh worldwide in the Optimistic case. Of this result, server power consumption was estimated to be 17 TWh in Japan and 510 TWh worldwide in the Modest case, and 5 TWh in Japan and 140 TWh worldwide in the Optimistic case.
　Since 2050 is in the distant future, the reliability of that prediction is not high. In the Modest case, the improvement rate will be the same as in 2030, and in the Optimistic case, non-von Neumann CMOS computing and quantum annealing as new computational principles were expected to contribute. As a result, the power consumption of data centers in 2050 is estimated to be 500 TWh in Japan and 16,000 TWh worldwide in the Modest case, and 110 TWh in Japan and 3,000 TWh worldwide in the Optimistic case. Of this, server power consumption was estimated to be 330 TWh in Japan and 11,000 TWh worldwide in the Modest case, and 50 TWh in Japan and 1,600 TWh worldwide in the Optimistic case.
　These results show that the power consumption of data centers will not reach an unacceptable level until 2030 due to improvements in current technology. On the other hand, current technologies are expected to face their limits in the next 10 years, therefore, when we look ahead to 2050, when AI will become more widespread in society and autonomous cars will become a practical reality, the development of innovative new technologies will be required. In particular, research and development of CPUs and computational functions that complement CPUs will be extremely important. For this purpose, long-term and continuous investment in basic research, applied research, and human resource cultivation is required.