Chairman’s Message

<Main Points>

AI is driving increased electricity demand in data centers, necessitating huge investment in electricity supply infrastructure.

Improving the energy efficiency of AI must be pursued, which can potentially reduce electricity demand in data centers.

Data center energy savings potential amounts to 20% by 2035, with IT energy efficiency improvements representing the largest contribution.

AI must be fully utilized to enhance energy efficiency of various activities.

Energy efficiency improvement in the industry sector using AI represents the greatest sub-sector potential, resulting in 2-5% additional energy efficiency improvement by 2035.

Energy efficiency improvement in the industry, road transport, offices/commercial, and residential sectors can amount to 178 Mtoe, or 1.7% energy efficiency improvement globally by 2035 (compared with Advanced Technologies Scenario).

Wider utilization of AI for energy efficiency requires training, incentives, data standardization/protection, and technology transfer.

1. Massive electricity demand for data centers magnified by AI

Global electricity demand for data centers was flat until 2019. It started to rise in part by the shift to working from home, which in turn has driven the growth in cloud service after the breakout of COVID 19 in 2020. The growth accelerated with the introduction of generative AI in 2022. Since then, electricity demand expansion for data centers and AI has become the main topic for not just energy experts but also among IT sector leaders. While there is a wide range of projections of future demand for data centers, there is a consensus that electricity demand will grow substantially, and more capacity for electricity generation and transmission is essential. The Strategic Energy Plan of Japan, decided this February, expects up to 20% increase in electricity demand by 2040, which is a sharp contrast to the continuous decline in electricity demand since 2007.
　With strong demand for electricity expected, policymakers and IT leaders are emphasizing the urgent need for more investment in electricity generation and transmission. To be sure, it is not just AI. Electrification and reshoring are pushing the electricity demand in advanced economies, and economic growth and rising living standards are massively expanding electricity demand in the Emerging Markets and Developing Economies (EMDE). But there is limited discussion about the need to make AI and data centers more energy efficient. There is even less discussion about the need to utilize AI to make our economic and social activities more efficient.
　Isn’t it time for the world to talk seriously about making AI more energy efficient, and using AI to make our activities more energy efficient, not just demanding more electricity?

2. Energy efficiency improvement of AI and data centers must be pursued

The breakdown of the composition of electricity consumption by data centers tells us that we will have to tackle both IT equipment and cooling.

Source: Hongyu Zhu et. al. (2023). Future data center energy-conservation and emission-reduction technologies in the context of smart and low-carbon city construction. Sustainable Cities and Society | VOL 89 | 2023 |

　The current measure of energy efficiency, PUE (Electricity Usage Effectiveness), is focused on cooling. PUE is a measure of data center energy efficiency, which can be calculated by dividing the total data center facility energy by the IT equipment energy consumption. The closer PUE gets to 1, the more efficient a data center is. As Figure 1 shows, cooling represents about 35% of the electricity consumption at traditional data centers with PUE representing 2, the focus on cooling is not wrong. In fact, we need to work much harder to make the cooling more energy efficient. Innovation in cooling is necessary.
　At the same time, we should not ignore IT equipment, which represents 50% of electricity demand. Servers and storage must be made more energy efficient and efforts to make chips more energy efficient must be accelerated. Innovation led by NTT of Japan, called IOWN, which aims to make the transition from electrons to photons, can be a game changer technology.
　It is not just about hardware. It is generally understood that generative AI requires 10 times more electricity to answer inquiries compared with conventional Google searches. Innovation in algorithms can rationalize necessary data processing. Matching the necessary data processing with the required task can rationalize the use of data and data processing.
　As the energy efficiency of IT equipment and data processing is important, I believe that we need to develop an appropriate measure to capture such energy efficiency, not just PUE focusing on cooling for data centers.
　The IEEJ estimates that electricity demand for data centers can potentially be saved by 20% by 2035. It can be realized through the combination of improving PUE, energy efficiency improvement of IT equipment, and rationalization of data processing.

Source: IEEJ. (2025).

　A 20% improvement in energy savings of data centers may not compensate for the increase in electricity demand driven by the need for more AI related activities, but it is significant enough to pursue seriously.

Figure 3. Data center electricity demand (Reference vs. Advanced Technologies Scenario) Note: Reference Scenario (Continuation of current trends in energy and environmental policies.) and Advanced Technologies Scenario (ATS) (Maximum implementation of policies for energy security and climate action, with technologies deployed to maximum extent, considering feasibility and acceptance)
Source: IEEJ. (2025).

3. AI must be fully utilized to enhance energy efficiency of various activities.

Energy efficiency is often called the “first fuel” of choice in global efforts toward achieving carbon neutrality. It is generally most cost-effective to promote energy efficiency compared with other supply options. At COP 28, member countries pledged to double the annual energy intensity improvement through this decade. As this pledge was based on IEA’s Net Zero Scenario, the assumption was raising the annual improvement to 4%, double the achievement of 2% in 2022. But 2022 was a special year. With the energy crisis, many countries and companies engaged in broad energy saving efforts that led to the 2% improvement in 2022. But as the crisis receded, the annual improvement dropped to 1% in 2023 and 2024. If we extrapolate this 3 percentage point gap for this decade, the energy demand in 2030 can be 28% larger than the path to realize net zero by 2050. It is clear that we must do more to drastically improve our energy efficiency to achieve net zero.
　As many efforts have been made to improve energy efficiency, especially in advanced economies, the marginal room for improvement by conventional means may be limited. This is why I hope that AI can play a major role in further improving energy efficiency of various activities.
　AI can help the energy sector in multiple ways: it can detect failure in production processes, predict renewable energy supplies based on weather changes, and can accelerate development of new materials through simulation. For energy efficiency, the most important role of AI is that it can optimize various activities including production processes, better operation of buildings and optimizing transport system operations.

4. The industry sector has the greatest potential for energy efficiency through AI

The industry sector has been the main target of energy efficiency efforts, but they were basically dependent on the experience and knowledge of veteran operators and were focused on particular processes. With AI, optimal operation can be derived through analyzing large datasets, factory-wide optimization, and dynamic control on a real time-based forecast. Energy efficiency improvements can be expected in various sub-sectors including steel, chemicals, cement, paper and pulp, and other industries including machinery. Regionally, substantial energy efficiency gains can be expected in Advanced Economies, China and India.
　According to the analysis by the IEEJ, energy savings in the industry sector can potentially be improved globally by 2-5% by 2035. This improvement is on top of the improvement reflected in the ATS of the Global Energy Outlook issued by the IEEJ.

Note: AI’s energy savings potential is estimated as the addition to technological/operational improvements in the ATS. Also, different diffusion rates are assumed by country/region and by sub-sector.
Source: IEEJ. (2025).

5. Energy efficiency improvement through AI in various sectors

Energy efficiency improvements through AI can be expected in various sectors beyond industry. In road transport, including passenger vehicles, buses, and trucks, AI can potentially lead to energy savings by 36.5 Mtoe globally, about 0.9-2.9% savings by 2035 on top of the ATS. Better projections can avoid traffic congestion and optimization of freight truck load. AI can also optimize the routing of vehicles. Through autonomous driving, fuel efficiency can be improved by maintaining constant speeds and minimizing air resistance.
　Energy efficiency of offices and commercial sectors can be improved as well. Through better design using AI, energy efficiency of buildings and regional districts can be enhanced. By analyzing sensor data of movement of people and air flow, lighting, air conditioning, and elevators, we can make buildings and regional districts more energy efficient in the operation. The total of energy efficiency savings can potentially reach 8.8 Mtoe or 102 TWh globally. This scale is equivalent to about 10% of the electricity demand for data centers at 1,080 TWh in 2035.
　AI also has potential in the residential sector. AI can optimize air conditioner operation by analyzing data such as room and outside temperature, household electricity use, solar PV generation, and electricity tariffs. The IEEJ expects that energy efficiency gains of 1.5 Mtoe (2.3% improvement) in advanced economies and 0.9 Mtoe (1.5% improvement) in EMDEs by 2035 are possible.

　If we add all the potential gains in energy efficiency through AI, the total scale can potentially reach 178 Mtoe by 2035. This is equivalent to 1.7% of global energy demand in 2035 based on the ATS. While energy efficiency improvements will be divided between heat and electricity, electricity efficiency gains will be 68 Mtoe, representing 2.4% of global electricity demand in 2035 based on the ATS. These numbers may not sound dramatic, but in fact, the potential electricity savings use could represent 74% of the electricity demand expected for data centers globally in 2035 based on the Reference scenario. As the growing electricity demand triggered by AI and data centers is the topic of the day, it is very much worthwhile utilizing AI to make various activities more energy efficient. Source: IEEJ. (2025).

6. How can we utilize AI for energy efficiency?

As discussed, there is substantial potential for energy efficiency improvement by utilizing AI. However, this is only potential. We need to address multiple impediments which can limit the realization of such potential.
　A shortage of skills is the most common impediment across sectors. We need to train talent at scale to use AI effectively to improve energy efficiency. Lack of knowledge is another barrier. We need to have technology transfers from large companies to SMEs, and from the advanced economies to EMDEs. Data is not readily available in some sectors. Data is often not shared among companies or different sectors or countries. Standardization of data formats and protection of data are essential. As energy efficiency improvements using AI require investment, there must be effective policies to incentivize investment in energy efficiency improvement using AI. We need comprehensive policies and actions to realize the potential for energy efficiency improvement using AI. I believe that it is worthwhile considering the offsetting impact it can have to minimize the increase in electricity demand from AI.

7. Energy efficiency of AI & Energy efficiency by AI must be pursued

It is not good enough to only urge massive investment in electricity generation, transmission, and distribution to fill the ever-growing appetite for electricity by AI. We need to seriously promote energy efficiency enhancements of AI and data centers. An appropriate measure of energy efficiency of data centers beyond cooling must be developed and be deployed broadly.
　AI should not be just a consumer of electricity. AI should play a greater role in improving energy efficiency among various sectors to help offset some of the increased electricity demand triggered by AI. This should be the responsibility of policymakers, companies in various sectors using energy, and also the hyperscalers that are pushing AI and profiting from it.

The topics introduced in this message will be explained in detail in the webinar below.
Please feel free to click the hyperlink below to register.
For those who miss it, a video recording will be available afterwards.

The 20th IEEJ Webinar for the World
Date/Time: 2AM ET /8AM CET/ 10AM AST/ 3PM SGT/ 4PM JST on Wednesday, December 10, 2025
2. Webinar App.：Zoom
3. Agenda：
　Presentation：“The IEEJ Outlook 2026: The Future of AI and Energy Demand”
Naoko DOI, Ph.D., Senior Research Director, Assistant Director,Climate Change and Energy Efficiency Unit, IEEJ
　QA Session: (Moderator: Toshiyuki SAKAMOTO, Board Member, Director,Climate Change and Energy Efficiency Unit, IEEJ)
4.Registration: https://us02web.zoom.us/webinar/register/WN_6nZlp1dGRSqBIZChzv6BPQ 5.Video
You can watch the video here (it will be uploaded shortly after the webinar). https://eneken.ieej.or.jp/en/seminar/other/ieej_seminar.html