Powering Possible 2024: AI and Energy for a Sustainable Future

AI and Energy for a Sustainable Future POWERING POSSIBLE

AI’s energy demand 71 Improving the efficiency of AI and data centers 79 Developing new sources of carbon-free electricity 81 Meeting the electricity needs of AI with carbon-free energy 69 04 Seven priority areas 87 – 102 05 Recommendations for realizing AI's potential for the energy transformation 85 Table of contents Executive summary 5 Energy, innovation, and growth 13 Energy and AI at the crossroads 19 The potential for partnership between the energy and technology sectors 23 The challenge and opportunity for energy and AI 11 01 02 Decarbonizing the current energy system 29 Building the energy system of the future 35 The potential for AI to accelerate the energy transformation 25 03 Systems level 49 Operational level 55 Global level 59 AI's potential to evolve more advanced abilities 67 AI and the future energy system 43 2 Powering Possible 1

We are at a pivotal moment for human progress driven by three megatrends, the rise of the Global South, the accelerated energy transformation, and the rapid growth of AI. AI is an era-defining innovation that is altering the pace of change itself - resetting the boundaries of productivity and the possibilities of progress. But in doing so, it is also creating a power surge that nobody accounted for just 18 months ago. By collaborating to solve AI’s near-term challenges, we can also unlock AI’s long-term benefits across the energy value chain – helping to secure a sustainable and prosperous future for generations to come. This new era calls on us to do two things at once: Meet the AI moment while transitioning to a more sustainable economy. In a world that will need more electricity, not less, it's imperative that we generate more carbon free energy to power AI and use that very same technology to increase capacity, optimize transmission, and expand energy access to communities around the world. This isn’t a journey any of us can take alone. It requires working across technology, energy, science, and policy sectors to find solutions and accelerate our collective progress. Dr. Sultan Al Jaber Managing Director and Group CEO, ADNOC and Chairman, Masdar Brad Smith Vice Chair and President, Microsoft 4 Powering Possible 3

Executive summary Energy and technology, longtime drivers of humanity’s progress and prosperity, have powered the world’s industrialization and enabled dramatic improvements in living standards. In just the past 40 years, the global energy system has contributed to lifting over 1 billion people out of poverty,1 a 10-year increase in life expectancy,2 and a rise in literacy rates from 70% to 90%.3 Yet these benefits have not been experienced equally; nearly 750 million people still lack access to electricity, particularly in the Global South, where expanding energy access is fundamental to fostering economic prosperity.4 Since the 1960s, global energy demand has quadrupled (despite increased energy efficiency) and technological advancements have surged.5 The global economy now stands at a crossroads: the imperative to accelerate the net-zero energy transformation, growing energy demand, particularly in the Global South, and the expected development and growth of AI. Convergence of these three megatrends represents an opportunity to shape a sustainable and equitable future. The UAE Consensus, agreed to by 198 parties6 at COP28, laid out a path to achieve a just, orderly, and equitable transition to a net- zero energy system. Amongst other things, that requires tripling renewable energy capacity and doubling the rate of energy This report identifies seven areas of collaboration and action for the energy and technology sectors to accelerate a just, orderly, and equitable energy transformation to net-zero and to unlock the full potential of Artificial Intelligence (AI). efficiency by 2030, while accelerating new carbon-free technologies. AI has emerged as one enabling technology that has the potential to support the achievement of these objectives whilst meeting rising energy demand.7 AI’s recent advancements and increased adoption suggest it can help accelerate the global energy system’s transformation to net zero whilst meeting the projected 3 –4% per year increase in global electricity demand through 2030.8 However, to realize AI’s full potential, much more is needed, and much more is possible. The electricity needed to power the data centers critical to AI and other digital services is expected to grow between 8 –23% per year through 2026, bringing AI use to 0.24% of global electricity demand.9 Despite $250 billion in private investments in AI between 2017 and 2023, only 5% of that funding went to the energy sector. Much more investment is urgently needed in this area to leverage AI’s potential for driving innovation and sustainability in energy systems.10 Of course, even the most effective and widespread deployment of AI must be combined with other enablers of the transition, including policy and regulation, public and private investment, and support for the Global South. SURVEY INSIGHT believe AI will have a transformative impact on the energy system and it increases to Of executives surveyed 92% by 2030 97% by 2050 6 Powering Possible 5

“ As the UAE’s renewable energy champion and a global clean energy leader, Masdar has a proven track record in pioneering clean energy projects for nearly two decades. But AI has changed the game, and it represents one of the greatest opportunities to transform the global clean energy industry in history. AI – and the data centers it requires - will serve as an increasingly important driver of global energy demand. Meeting this demand sustainably will require a multifaceted approach, and it is up to all of us to work together to unleash its full potential and build a more sustainable future.” Mohamed Jameel Al Ramahi CEO, Masdar According to the International Renewable Energy Agency (IRENA), the annual rate of deployment of renewable generation must grow by 16% every year through 2030 to meet the tripling target.11 To achieve this target, it is critical to address key challenges such as permitting, grid infrastructure, financing, and supply chain bottlenecks. For instance, permitting constraints can delay utility-scale projects by seven to 12 years. AI could help to streamline data workflows and speed up permitting and siting by up to 40%.12 By 2040, electricity grids will need to add or refurbish 80 million kilometers of network in order to connect rising demand and more diversified supply.13 AI is already playing a role. For example, companies in India and Ghana are using it to support the expansion of transmission networks and the integration of renewable energy.14 To achieve the goals of the UAE Consensus and the Paris Agreement, methane emissions must fall by over 30% by 2030 under the Global Methane Pledge.15 AI tools are already being deployed to detect methane with greater accuracy at a fraction of the cost. Although AI can help, methane reduction can only be delivered with stricter regulation and enforcement; 50% of methane emissions come from 11 countries where significant progress is needed.16 For this report, ADNOC surveyed approximately 400 global leaders from different sectors on AI and Energy. Among the executives surveyed, 92% by 2030 and 97% by 2050 believe that AI will have a transformative impact on the energy system, with a focus on improving energy efficiency in the short term and advancing emerging energy solutions (e.g., small modular nuclear reactors, carbon capture) in the longer term. Despite the enthusiasm shown by senior leadership, conventional energy has been slower to adopt AI than other industries (e.g., 23% of conventional energy organizations allocate more than five percent of their annual budget to implementation and development of AI-based solutions, compared to 56% of the renewable sector who allocate more than five percent of their annual budget to implementation and development of AI-based solutions).17 The energy system of the future will need to rely more on electrification and distributed generation and place a premium on flexibility, efficiency, and resilience. AI’s capabilities in analytics and optimization can help to manage this variability and complexity, on both the supply and the demand side. For example, AI can be used to model how distributed sources like rooftop- solar affect grid capacity. It can also help match variable supply with evolving demand, which will be essential in the renewables- dominated electricity system of the future. Specifically, AI can dynamically simulate grid load and transmission capacity to improve transmission and distribution efficiency. AI can also support a more resilient grid. For example, smart grids that link sensors and GIS data with AI can better predict disruptions from wildfires, floods, and storms. Finally, AI is already accelerating research and development into new materials that could increase the capacity and stability of energy storage and carbon removal, amongst other areas.18,19 AI capabilities have matured rapidly in recent years, and expectations are high. Investment in AI is projected to rise to $150 billion in 2025.20 Technology and energy leaders must work together to ensure access to reliable carbon- free energy, a critical enabler for AI’s continued development and growth. Together they are well positioned to make this happen. AI depends on data centers and other critical infrastructure. Data centers account for approximately 1–1.3% of global electricity demand, and this is expected to almost double to 2% by 2026.21 Of global electricity demand, AI-driven data centers account for a small but growing share: 0.02% in 2022 and 0.24% in 2026.22 Demand growth for continuous and reliable power can put pressure on electricity grids where data centers are concentrated. In the European Union (EU), electricity demand for data centers is expected to increase at ~9% per year, due to digitalization including AI, and could exceed 5% of total EU electricity demand by 2026. In the U.S.—the largest data center market globally—data centers represented about 4% of the nation’s electricity demand in 2022 and that is expected to rise to nearly 6% by 2026.23 This demand growth can be challenging in regions with aging transmission infrastructure and growing competing demands for new, carbon-free generation. Conventional energy companies have been slower to adopt AI-based solutions than renewable energy counterparts SURVEY INSIGHT The electricity needed to power the data centers is expected to grow between bringing AI use to 0.24% of global electricity demand 9 8 - 23% 8 Powering Possible 7

We see huge potential for collaboration between the energy and technology sectors to accelerate a just, orderly, and equitable energy transformation to net zero and manage the energy system of the future whilst unlocking the full potential of AI with carbon-free electricity. Specifically, we see seven priority areas: 05 Develop AI with and for emerging economies, to meet their unique needs. 06 Establish data standards and protocols for AI to better support the energy sector. 07 Advance policy and governance for responsible, sustainable AI and a secure and inclusive transformation to a net-zero energy system. Build capacity in the workforce to leverage AI for energy transformation. 02 Invest in AI for the energy transformation, with a focus in four key areas: tripling the availability of renewable energy, building resilient grids, reducing methane emissions, and utilizing carbon capture and storage. 01 Increase collaboration between technology and energy companies to deploy more carbon-free energy while making it more available and more affordable for all. 03 Expand and enhance grid capacity, increase availability of carbon-free electricity, especially in locally stressed grids or regions— while continuing to innovate to increase energy efficiency. 04 10 Powering Possible 9

01 ENERGY AND The challenge and opportunity for AI

Energy and technology have underpinned two centuries of global economic growth and prosperity. From the steam engine of the early 19th century through the internal combustion engine of the mid-20th century to the internet of the late 20th century, successive waves of technological innovation have enabled humanity to produce more output with less input. Global GDP per capita— after a long stagnation that may have lasted many thousands of years— took off in the 19th century. Today, it has reached $13,000 per person, a 13-fold increase in a little over two centuries. The widespread adoption of the technological innovations driving that growth relied upon access to affordable and reliable energy Technological innovation and energy have underpinned economic growth Figure 1 in ever-greater quantities. In 1800, primary energy demand was 6,250 TWh per year; today, it is 180,000 TWh.24 It was technological innovation that unlocked that energy—from the introduction of coal-powered steam engines in the coal mines of the late 18th century to the fracking revolution of the 1990s. It has been a cycle of positive reinforcement (see Figure 1). Peng Xiao GCEO, G42 “ AI will be the catalyst for transforming energy systems globally. By harnessing advanced AI solutions, we can expand energy access, accelerate economic development, and enable adopting nations to spearhead the worldwide shift toward sustainable energy practices. This integration promises not only to optimize current systems but also to innovate new pathways for environmental stewardship and energy efficiency.” Efficient Steam engine 1769 Internal combustion engine 1867 Electrification 1890s – 1930s 0 1,000 4,000 7,000 10,000 13,000 1200 1400 1600 1800 1000 2000 20231 0 TWhs 40,000 TWhs 80,000 TWhs 120,000 TWhs 160,000 TWhs 200,000 TWhs GDP per capita (LHS y-axis) GDP per capita (USD) First Industrial Revolution Second Industrial Revolution Third Industrial Revolution Internet 1970s Global primary energy consumption (RHS y-axis) Energy, innovation, and growth The world has witnessed how technological innovation and access to affordable energy have delivered global growth and socio-economic progress for two centuries. Billions have been lifted out of poverty as a result, though millions are still without access to electricity and are at a disadvantage because of it. Note: GDP per capita data for year 0 – 1980 based on data from Maddison Project Database and University of Gronigen; 1980-present based on data from IMF Source: Our World in Data, World Bank, “World Bank World Development Indicators”, 2023; Bolt and van Zanden - Maddison Project Database 2023, University of Gronigen; Maddison Database 2010; Oxford Economics (GDP data from IMF Data Portal-1980 onwards); ADNOC Analysis 14 Powering Possible 13

More recently, primary energy demand has begun to decouple from economic growth in more mature markets. Global GDP has almost tripled since 1990, while energy intensity has fallen 34% (see Figure 2), driven by a shift from industry to services, as well as steady improvements in energy efficiency. Related, the emissions intensity of the global economy has declined by c. 60% since 1970, even whilst absolute emissions have continued to grow (see Figure 3). 1990 2000 2010 2019 China Brazil France UK US World 40 60 80 100 120 50 70 90 110 0 Global GDP (in 2017 $PPP) Energy intensity per unit of GDP (indexed to 1990 = 100, in $PPP) 51T 68T 97T 130T +5% +34% +43% +72% Economic output has become less energy-intensive since 1990 Figure 2 Global emissions continue to grow, even as emissions per capita fall Figure 3 Note: PPP = purchasing power parity; T = trillion Source: BCG, A Blueprint for the Energy Transition, 2023. Global emissions 1970 onwards [All values indexed to 1970 (1970 = 100)] Global annual CO2 emissions Global emissions per GDP $ 42 249 0 250 50 150 200 100 1980 1990 1970 2010 2020 2000 All values indexed to 1970 (1970 = 100). Source: Global Carbon Budget (2023) – with major processing by Our World in Data reduce the energy and emissions intensity Technological innovation has helped of economic output 16 Powering Possible 15

Human prosperity remains highly correlated to primary energy demand Figure 4 But the availability of energy remains strongly correlated to prosperity (see in Figure 4). In just the past 40 years, the global energy system has contributed to lifting over 1 billion people out of poverty, a 10-year increase in life expectancy, and a rise in literacy rates, from 70% to 90%. Approximately 60 countries that rank low or medium on the UN Human Prosperity Index, rely on less than 20 MWh of primary energy consumption per capita per year (see in Figure 4). Moreover, 750 million people still lack any access to electricity at all (see in Figure 4).25 Closing this access gap would be a major contribution toward alleviating poverty, particularly in sub-Saharan Africa. 60 Indonesia Costa Rica Mozambique South Africa China Japan Germany India 3 0 20 40 1.9 2.1 2.6 1.3 Low Medium High Very high 2 1 Primary energy use, by country (MWh per capita) Number of people at each level (billions) Level of human prosperity, based on HDI Titles Font size 9 Note: Countries with HDI >0.8 and with per capita energy consumption >60MWh are not shown. HDI = Human development index. HDI measures acountry's performance in terms of life expectancy at birth, average years of schooling, and gross national income. Low, HDI

“ The rise of AI requires significant infrastructure to support it. Private capital can finance critical investments in data centers as well as energy infrastructure and decarbonization technology needed to meet growing power demand. These investments can help power economic growth, create jobs, and drive AI innovation.” Larry Fink Chairman and CEO, BlackRock AI could contribute between to the global economic output by 2030 26 $1 trillion $7 trillion to AI’s Game-Changing Abilities in Action 27 20 Powering Possible 19 Measure, predict, and optimize complex systems Weather forecasting relies on highly complex simulations involving an array of parameters, including temperature, wind speed, dew point, and more. The latest AI-powered weather models can outperform the European Medium Range Weather Forecasting model, the leading legacy model, on more than 90% of predictive factors.28, 29 Accelerate the development of new solutions Advances in the material sciences are critical to the development of next- generation batteries and energy storage. Microsoft and the Pacific Northwest National Laboratory used AI to scan 32 million material candidates and identify a previously unknown solid-state electrolyte not present in nature that offers real promise for a better battery. This research, which took only nine months, would have required years if AI had not been deployed.30 Empower the workforce Generative AI (GenAI) introduced a new paradigm in conversational interaction between humans and computers. ChatGPT, the AI chatbot, reached 100 million users just two months after its launch—the fastest adoption in the history of the digital era to date.31 GenAI also has the potential to empower the workforce. According to a recent BCG survey, approximately 60% of employees who use GenAI estimate that the tools save them at least five hours per week.32 Energy and AI at the crossroads AI is one of the latest technological innovations with the potential to make a significant contribution to economic growth through its emergent abilities to measure, predict, and optimize systems; accelerate innovation; and empower the workforce. Estimates of AI’s potential contribution to global economic output by 2030 vary from $1 trillion to $7 trillion.26 Today, the global economy is at a crossroads, and energy and AI are at the center of developments. Energy and technology companies have the potential to drive this economic opportunity together. For example, technology companies can serve the latent demand for digitalization and deployment of AI in the energy sector; energy companies, on the other hand, can help technology companies with their net-zero goals by serving the necessary demand for carbon-free electricity.

Meanwhile, climate change threatens to further undermine economic growth. As a result of anthropogenic greenhouse gas emissions, global average temperatures have risen 1.2°C since the pre-industrial period.33 This rise has led to climate-related natural disasters causing significant and rising socio-economic impacts. If we want an even chance of limiting global temperature rise to 1.5°C by 2100, cumulative emissions since 1850 must be limited to c. 2,900 GtCO2; as of now, we have emitted more than 80% of that. If not controlled, we will exhaust the rest of it by 2030 (see Figure 5).34 The energy system, consisting of producers and consumers, accounts for approximately 75% of greenhouse gas emissions and has a central role to play. For context, this rate of emissions reduction implies an energy transformation roughly three times the speed of any previous one, and from a far larger base.35 Figure 5 Global remaining carbon budget dictates that this decade is critical 2390 2000 Historical Since 2020 2020 Historical emissions 1850-2019 Remaining carbon budgets (until 2100) 2020-2030 CO2 emissions assuming constant at 2019 level 1.5°C1 2°C2 (>50%) (83%) Carbon budgets 0 1000 2000 0 1000 500 1500 500 900 Cumulative CO2 emissions (GtCO2) 1.5° C 2° C This line indicates maximum emissions to stay within 2°C of warming (with 83% chance) Note: Figures in parentheses denote level of confidence; 1. 1.5°C with >50% level of confidence; 2. 2°C with 83% level of confidence Source: IPCC AR6 Synthesis Report; ADNOC Analysis 22 Powering Possible 21

The potential for partnership between the energy and technology sectors Greater collaboration between the energy and technology sectors has the potential to unlock progress across this global agenda. Fortunately, AI has become a higher priority over the last three years for 66% of energy companies, according to our survey. The following three chapters cover three specific areas: Chapter 3 explores the potential for AI’s emergent and potential abilities to support the operation of the energy systems of the future. Chapter 4 explores the small but growing energy demand for AI, and the role energy and technology companies can play in meeting that demand with carbon-free supply. Chapter 2 explores the potential for AI’s current and emergent abilities to support a just, orderly, and equitable energy transformation toward net-zero. SURVEY INSIGHT of leaders believe AI can increase the speed at which energy companies become more sustainable 82% SURVEY INSIGHT of energy companies over the last 3 years 66% AI has become a higher priority for 24 Powering Possible 23

02 THE POTENTIAL FOR AI to accelerate the energy transformation

SURVEY INSIGHT The potential for AI to accelerate the energy transformation Last year in Dubai, UAE, world leaders aligned on the pathway to decarbonizing the current energy system and building a system for the future. The COP28 UAE Consensus requires, amongst other things, tripling renewable energy capacity and doubling the rate of energy efficiency by 2030, as well as accelerating the deployment of new carbon-free technologies (see Figure 6). AI’s emergent abilities could make a significant contribution to delivering on these challenging but critical commitments. According to the ADNOC global survey, more than 90% of business leaders believe AI will have a transformational impact on the energy transformation by 2030, and even more believe it will by 2050 (see Figure 7). AI could play a significant role both in decarbonizing the current energy system and building the energy system of the future. 82% of leaders from the same survey believe AI can increase the speed at which energy companies become more sustainable. Figure 7 Leaders’ confidence in AI driving innovative energy breakthroughs for emerging energy solutions like carbon capture and SMRs quadruples in the long term Figure 6 COP28 delivered major commitments across renewable energy, energy efficiency, and more — laying the groundwork for the net-zero system of the future… The world requires a just, orderly, and equitable transition to a net-zero energy system. Note:1% of respondents responded 'Don't know' are included; Source: ADNOC AI Survey; Respondents were asked in which area of energy production do they feel artificial intelligence (AI) will have the most transformational impact by 2030 and by 2050. … but bold commitments need to be supported by bold action. 92% 2050 2030 28. Further recognizes the need for deep, rapid and sustained reductions in greenhouse gas emissions in line with 1.5 °C pathways and calls on Parties to contribute to the following global efforts, in a nationally determined manner, taking into account the Paris Agreement and their different national circumstances, pathways and approaches: (a) Tripling renewable energy capacity globally and doubling the global average annual rate of energy efficiency improvements by 2030; (b) Accelerating efforts towards the phase-down of unabated coal power; (c) Accelerating efforts globally towards net zero emission energy systems. utilizing zero- and low-carbon fuels well before or by around mid-century; (d) Transitioning away from fossil fuels in energy systems, in a just, orderly and equitable manner, accelerating action in this critical decade, so as to achieve net zero by 2050 in keeping with the science; (e) Accelerating zero- and low-emission technologies, including, inter alia, renewables, nuclear, abatement and removal technologies such as carbon capture and utilization and storage, particularly in hard-to-abate sectors, and low-carbon hydrogen production; (f) Accelerating and substantially reducing non-carbon-dioxide emissions globally, including in particular methane emissions by 2030; (g) Accelerating the reduction of emissions from road transport on a range of pathways, including through development of infrastructure and rapid deployment of zero-and low-emission vehicles; (h) Phasing out inefficient fossil fuel subsidies that do not address energy poverty or just transitions, as soon as possible; 10% Traditional oil & gas 11% Renewable energy 28% Emerging energy solutions (e.g., carbon capture,biofuels, small modular nuclear reactors (SMRs), etc.) 8% AI will not have a transformational impact in this time period1 9% Traditional oil & gas 16% Renewable energy 28% Energy distribution (distributing energy more effectively, lessening cost/environmental impact) 32% Optimizing energy efficiency (creating more energy with less environmental impact) 19% Energy distribution (distributing energy more effectively, lessening cost/environmental impact) 26% Optimizing energy efficiency (creating more energy with less environmental impact) 7% Emerging energy solutions (e.g., carbon capture,biofuels, small modular nuclear reactors (SMRs), etc.) 3% AI will not have a transformational impact in this time period1 97% 92% 28 Powering Possible 27

Figure 8 Fossil fuels, coupled with CCS wherever possible, will continue to remain a critical fuel in the future across different scenarios Total fossil fuel supply (% of total supply) Renewables Fossil Fuels 2022 2030 Stated Policies Scenario (STEPS) Net Zero Scenario (NZE) 2050 2022 2030 2050 80 74 56 100% 0% 50% 80 63 14 Source: IEA, Net Zero Roadmap: A Global Pathway to Keep the 1.5˚C Goal in Reach; IEA, WEO 2024 global oil production, to achieve near-zero methane emissions by 2030 from their upstream operations.41 One challenge is that methane leaks are notoriously difficult to detect and manage: methane is invisible to the naked eye and leaks intermittently from a range of point sources.42 Once leaked, methane disperses rapidly and often escapes detection.43 This leads to potential under- reporting.44 The IEA notes that methane emissions from the energy sector could be up to 70% greater than government estimates.45 Decarbonizing the current energy system According to the International Energy Agency (IEA), demand for coal, oil, and gas will peak this decade based on current policies and fall to 56% by 2050. Even under its Net Zero Scenario, fossil fuels supply would remain 14% of energy supply by 2050, though coal would decline to near zero.36 In short, oil and gas coupled with carbon capture and storage (CCS) wherever possible, will continue to be a significant proportion of the energy system through the middle of the century, and with significant regional variation in the pace and scale of decline. The use of oil and gas in the energy system will need to be decarbonized to achieve the goals of the Paris Agreement. The emissions associated with both the production and use of oil and gas must be abated. Methane abatement and CCS are priorities.37 Early developments suggest that AI has the potential to play a valuable but supporting role in both methane abatement and CCS. The world must rapidly decarbonize the current energy system to meet its goals for 2030. Methane is approximately 80 times more potent than carbon dioxide (CO2) at trapping heat in the atmosphere over a 20-year period. It is the second-most abundant anthropogenic greenhouse gas after CO2, accounting for about 20% of global emissions.38, 39 The Global Methane Pledge commits 111 countries that account for 45% of the world’s methane emissions to reduce their emissions by at least 30% by 2030.40 The Oil and Gas Decarbonization Charter, launched at COP28, commits 30 national oil companies and 20 international oil companies, or more than 43% of Role of AI in reducing methane emissions Darryl Willis Vice President, Energy Industry, Microsoft “ As we collectively navigate the multidimensional journey towards net zero emissions, AI has transformative capabilities for optimizing energy use, reducing emissions, and innovating carbon free energy solutions. We are committed to fostering strong partnerships between the energy and technology sectors to simultaneously address economic growth, energy access, and climate change. Together, we can harness the power of AI to create a more secure, equitable and sustainable future.” 30 Powering Possible 29

Figure 9 AI-generated insights AI-generated insights from geospatial, meteorological, and historical leak rate data, used in conjunction with atmospheric dispersion models, can optimize sensor placement for maximum coverage and the timely detection of leak emissions.46 Automated generation Automatic generation of incident reports and the identification of available technicians and materials can expedite repair work. 1. Methane Detection and Monitoring Pilot ADNOC has launched a pilot project using methane detection technology for enhanced environmental management with high accuracy compared to the industry standard. The initiative uses passive FTIR spectroscopy, computer vision, and deep learning to monitor large areas (in combination with sensor networks) The technology provides real-time observation and remote operations. The system offers reliable alerts, 24/7 detection, and validation of methane, CO, CO2, TVOC, and H2S. It supports LIDAR efforts, identifies emission sources, and uses AI for plume modeling and heat mapping. 2. Real-Time Flare Combustion Monitoring ADNOC has implemented an AI-based solution for real-time flare combustion monitoring Using cost-effective CCTV cameras and deep learning, ADNOC assesses combustion efficiency (CE) with Temporal Standard Deviation (TSD) and flare event detection networks. The system provides reliable real-time data, supporting emission control strategies, aligning with environmental regulations, and enhancing sustainability by reducing emissions cost-effectively. Recent developments suggest that AI’s emergent abilities could support oil and gas players’ endeavors to measure, predict, and optimize complex systems to improve methane management. Potential opportunities include: Progress is being made. For example, Oxford University has developed an AI tool that scans geospatial data to detect leaks 20% more accurately than legacy tools.47 The model was trained on large volumes of data from NASA satellites. The researchers have made the base data and code open-source so that the tool is available to others. ADNOC is also developing and deploying a range of AI-based tools for managing methane that have shown early promise (see Figure 9). Methane detection Flaring management 3. Lab-Scale Flare Stack System ADNOC has developed a lab-scale flare stack system for studying and optimizing combustion, emissions, and safety protocols in industrial processes. This prototype simulates real-world conditions for systematic data collection and AI-driven flare management testing. Detailed monitoring and manipulation of combustion parameters generate high-quality data for AI models. This initiative can enhance the understanding of flare dynamics, reduce environmental impacts, and offer actionable insights for optimizing flare operations. ADNOC AI use cases for methane emissions and flaring reduction have shown early promise Algorithm analysis Algorithms that analyze data from internet of things (IoT) sensors can detect and even quantify leaks in near real time. 32 Powering Possible 31

Taufik Tengku President & GCEO, Petronas “ As a solution, CCS represents a measurable undertaking that can help move us towards a decarbonised energy system. Through increased adoption of AI, we can significantly improve CO2 capture efficiency, rapidly identify and map optimal storage sites and closely monitor injected CO2 behavior. Today, we are only scratching the surface of AI's potential. Energy players must continue to innovate the solutions they are offering through technological partnerships in order to meet the energy needs of the present without compromising future generations as part of a just and responsible energy transition.” Role of AI in improving CCS AI has the potential to support innovators in building scale and improving the efficiency of CCS projects in both the capture and sequestration phases: CCS is an essential tool for decarbonizing sectors where emissions are hard to abate otherwise (such as cement, steel, and chemicals) and for producing carbon-free hydrogen from fossil fuels. To reach net zero by 2050, CCS deployment must increase significantly. By 2030, global CCS capacity needs to be more than 40 times larger than it is today.48 Greater scale and increased efficiency will be critical.49 The IEA highlights the need for continued innovation in CCS to improve its effectiveness (for example, increasing capture rates from 90% to 98%) and reducing costs. This requires advancements in capture technologies, transportation methods, and storage solutions. Moreover, CCS needs to be integrated across various sectors, including power generation, industry, and fuel transformation, with a particular focus on scaling in regions with large industrial bases.50 Storage AI might also provide operational support in the sequestration phase of deployments.53 It can simulate pressure levels during carbon storage, aiding in the identification of optimal injection rates and sites for carbon sequestration. Stanford University, California Institute of Technology, Purdue University, and NVIDIA have developed an AI-based tool that doubles accuracy in certain simulation tasks.54 Capture AI could help identify new materials that will support higher capture rates.51 Material properties related to CO2 binding and kinetics determine the performance of CCS hardware. Using AI and supercomputers to design materials with optimal carbon capacity, researchers at Argonne National Laboratory identified 120,000 promising new material candidates in only 30 minutes.52 AI can be used for rapid screening of potential CCS site locations and assessing trap and seal integrity. For example, Geoteric is using AI to address the challenge of identifying suitable storage sites for carbon sequestration projects.55 AI tools can support comprehensive CCS planning by identifying the optimal pathways between sequestration sites and emission sources, considering factors like distance, transportation options, and volume. 34 Powering Possible 33

Global under construction and planned solar and wind projects as of February 2023, in GW Under construction1 PPA signed/FID2 Bidding process3 Authroized4 Submitted5 Announced6 89% of projects under planning, mostly announced and authorized (89%) 1,553 166 (11%) 144 (9%) 195 (13%) 325 (21%) 121 (8%) 601 (39%) 1. All consents received, preliminary work has commenced, and placing main contracts has occurred and/or project under synchronization. 2. The project company has signed an offtake agreement, the project company has made the final investment decision. 3. Authorized project with an order has been placed for main equipment and/or major site work. 4. Project has received public/statutory consents by national authorities. 5. Project has been submitted to national authorities. 6. Project announced by company or planned in a national development plan. Source: Enerdata February 2023. Building the energy system of the future The energy system of the future will be built from many technologies available today—such as wind, solar, and batteries—and some that require further development—such as advanced geothermal, small, modular nuclear reactors, advanced nuclear, and green hydrogen. To be successful, we must continue pushing these technologies down the cost curve and accelerating the speed of their deployment. AI is not a silver bullet in this, but it has a role to play. The energy system of the future will be built from development and deployment of new technologies. AI has a supporting role to play. Site selection When identifying a site for new renewable energy projects, developers optimize for a range of factors, including weather patterns, topography, grid connectivity, and transmission congestion.57 These factors are vital to consider because they influence deployment lead time and cost effectiveness, both factors in the return on investment. AI-based optimization tools could prove helpful.58 They can assess weather patterns, analyze benchmark data from other projects in similar environments, and streamline interconnection studies to understand how rapidly and efficiently a project could be linked to the grid. For example, the World Bank’s REZoning tool is an open-source platform that uses advanced analytics to identify optimal locations for solar and wind installations.59 Permitting The build-out of renewables has slowed because of the permitting process. All told, the permitting and approval of renewable energy projects can take seven to 12 years, creating a major bottleneck in the planning stage. Today, nearly 90% of renewable energy projects worldwide are somewhere in the planning stage—construction is still in the future (see Figure 10). Figure 10 Accelerated permitting is critical to near-term deployment According to the IRENA, the rate of deployment for renewables must accelerate to meet the target of tripling renewable energy capacity by 2030, as required by the COP28 UAE Consensus. Specifically, annual rate of wind deployment needs to increase by 16.9% and solar-PV deployment by 18.4%.56 AI can contribute to this effort, including by accelerating site selection and permitting. Role of AI in renewables build-out AI and GenAI can expedite various parts of the permitting process. For example, AI can analyze satellite imagery and geospatial data for environmental impact assessments to support in expediting the process by up to 40%, from five years to three years,60 while GenAI’s language processing and text classification capabilities can help streamline documentation processes.61 36 Powering Possible 35

Role of AI in reducing downtime Once renewable projects are operational, it is crucial to maximize both emissions reduction and business value. Investments in renewables, grids, and battery storage need to double through 2030 to meet the COP28 target of tripling capacity.62 One attractive aspect of the investment proposition for renewables is that, after the initial capital outlay, projects have relatively low operating costs.63 AI applications to reduce downtime in variable renewables AI helps maintain low operating costs for renewable projects by minimizing the amount of downtime due to planned or unplanned maintenance. AI-driven predictive maintenance can reduce downtime by 10% to 20% at the asset level, significantly lowering maintenance costs.64 Reduced downtime also allows projects to supply more clean power to the grid, potentially replacing energy from emission-intensive sources. For example, a 10% to 20% reduction in downtime across the asset base of a 1 GW wind farm could translate to emissions savings of approximately 5T CO2 per year (see Figure 11). Presight, a leading big data analytics company powered by GenAI, has developed an asset management tool for its renewable energy projects around the globe. Figure 11 ~10-20% AI-enabled reduction in unplanned downtime for solar and wind ~95% AI-enabled reduction of emissions attributable to downtime optimization AI-driven energy loss prevention potential Fossil fuels VRE 0.03 0.07 0.02 0.01 0.06 0.01 Estimated unplanned VRE downtime (TWh) Emissions corresponding to additional VRE generation vs. equivalent fossil fuel generation (MT CO2) 0.11 0.03 4.91 Solar Wind 1.84 -94% -99% Solar Wind Note: Methodology used 1GW as representative of an average solar/wind farm capacity, potential impact is estimated based on the real-world deployment of 10-20% downtime reduction with the upper end of the range being used in the analysis on the page. Source: NREL, Life Cycle Greenhouse Gas Emissions from Electricity Generation: Update 2021; IPCC - Emission Factor Database (2023) – with minor processing by Our World in Data; Expert interviews; ADNOC analysis AI-driven predictive maintenance capability can reduce downtime by at the asset level 64 10% – 20% 38 Powering Possible 37

Role of AI in green- technology innovation Figure 12 Several technologies must climb the technological readiness curve toward maturity before they can be deployed at a global scale Continued innovation will be needed to develop and commercialize new technologies and eliminate the “green premium,” which is the cost difference between legacy carbon-intensive and newer, low-carbon technologies. New technologies that are not yet at scale will account for 35% of the emissions reductions needed to achieve net zero by 2050.65 Bringing these technologies to scale is achievable but will require concerted effort. Innovation stage Demonstration (TRL = 7 or 8) Prototype (TRL < 6) Early adoption (TRL = 9 or 10) Maturity (TRL = 11) Technology readiness level (TRL) TRL is determined during an assessment that examines program concepts, technology requirements, and demonstrated technology capabilities. TRLs are calculated on a scale from 1 (least mature) to 11 (most mature) Low-carbon hydrogen and synfuels Bioenergy Energy storage Carbon removal (incl. DAC) Green factory technologies Distributed energy Key challenges to technological maturity include: Funding End-system integration Supply chain Talent sourcing Factory setup Product development Patrick Pouyanné Chairman and CEO, TotalEnergies “ Electricity and renewable energies are digital ready: AI enables us to accelerate and optimize their integration along the entire electricity chain. AI technology will play a key role in the transition to a net-zero energy system!” The IEA has highlighted hydrogen electrolyzers, direct air capture (DAC), and next-generation batteries as three of the most critical innovations along the path to net zero.66 However, these innovations, along with other nascent technologies, face a steep climb to technological maturity, reducing costs, and scalability (see Figure 12). Source: BCG, Fast-Tracking Green Tech: It Takes an Ecosystem, 2023. 40 Powering Possible 39

Direct Air Capture has the potential to help meet the world's net-zero target Green hydrogen Developing membrane-less technology for hydrogen production is critical for the technology’s ability to scale. Researchers from Harvard, EPFL, and HPE have developed AI-enabled digital twins for membrane-less electrolyzers, with the potential to boost efficiency by up to 20% and cutting costs by up to 25%.67 Direct Air Capture (DAC) Researchers from Microsoft, MIT, and UC Berkeley are developing an AI model to accelerate the discovery of new materials for cost- effective carbon dioxide removal.68 The AI-driven model aims to identify materials capable of capturing large amounts of CO2 from the atmosphere while minimizing the energy required to release it for storage or reuse. Next-gen battery materials Battery storage faces challenges due to rising costs and limited supply of critical minerals like lithium. AI is helping to address these constraints. Microsoft and Pacific Northwest National Laboratories used AI to discover new battery materials with reduced lithium dependence in weeks instead of years.69 Small modular nuclear reactors (SMRs) Researchers at Purdue University and Argonne National Laboratory have developed an AI algorithm that predicts changes in SMR performance with 99% accuracy, potentially lowering costs and improving reactor efficiency, which would make nuclear energy more viable and easier to manage.70 These efficiency improvements and cost reductions are anticipated to make these technologies more economically viable, potentially attracting more capital and hence more adoption. Nuclear fusion At Princeton Plasma Physics Laboratory, AI is accelerating nuclear fusion development by predicting and mitigating plasma instabilities, optimizing complex computational tasks, and reducing computation time from tens of seconds to milliseconds.71 These advancements are key to making fusion a scalable and reliable energy source. Storage Octopus Energy’s Kraken platform is a powerful deep- tech solution designed to optimize the management of energy resources, particularly for energy storage systems. It leverages advanced analytics, real-time data monitoring, and machine learning to balance supply and demand, especially with the integration of renewable energy sources. AI’s game-changing ability to accelerate innovation shows particular promise in six key areas: AI has the potential to support a just and equitable transition to a net-zero energy system. To maximize its impact, AI must develop its capabilities and overcome obstacles to implementation. For example, the integration of AI into permitting processes will require improvements in the reliability and transparency of GenAI, significant process changes and the upskilling of personnel. Moreover, AI is not a silver bullet. Alone, it cannot address all the challenges associated with the Conclusion energy transformation. For example, any new materials for direct air capture will still need to be commercialized and funded at scale. Nonetheless, given the scale and urgency of the climate challenge, it would be remiss not to comprehensively explore and fully exploit the potential of AI. 42 Powering Possible 41

03 FUTURE ENERGY SYSTEM AI and the

The energy system of the future and AI AI can play a valuable role at all three levels, as its capabilities become more refined and barriers to implementation are overcome. For the world to deliver a just, orderly, and equitable energy transformation, the energy system of the future will need to look very different than it does today. It will need to be fundamentally transformed on three levels (see Figure 13 overleaf). At the systems level, energy must become more electric, distributed, and variable in both supply and demand. At the operational level, energy must become more efficient, resilient, and supported by sufficient and robust infrastructure. And at the global level, the Global South will need to play a far greater role, as it is expected to be responsible for nearly 80% of the new electricity demand between now and 2050.72 Anima Anandkumar Bren Professor of Computing and Mathematical Sciences, Caltech “ AI is transforming our understanding of the physical world, and it will have a transformative impact on how science and engineering are done.” 46 Powering Possible 45

System Operator Dynamically balancing supply and demand Grid scale storage Long-duration storage that is flexible and readily dispatchable O & G Production Abated through CCUS, methane reduction, and other Scope 1+2 levers Renewable and low-carbon production Large -scale wind, solar, nuclear, geothermal, and other low-carbon production Generation High-efficiency generation that works dynamically with conversion, storage, and the grid Domestic supply Domestic user Multi-agent interaction Commercial and Industrial users Transmission High-voltage bulk carrying power across regions CCUS at point of energy generation Conversion Electrons converted into highest value form (e.g. Power-to-X) CCUS at point of energy generation Domestic user Distribution Lower voltage network Figure 13 A vision of the future energy system Three transformations that will characterize the future energy system At the systems level, the future of energy system will become more electric, distributed, and variable. At the operational level, it will need to be efficient, resilient, and supported by the right infrastructure. At the global level, EMDCs have the potential to play a central role in the energy system of the future. 3. 2. 1. AI can help to orchestrate the future energy system, leveraging its abilities for complex analysis and optimization 48 Powering Possible 47

Energy will become more electric, distributed, and variable in supply and demand Electricity will be the core energy vector of the net-zero system, accounting for approximately 50% of total energy demand by 2050, up from 20% today.73 However, the primary energy mix behind electricity must diversify, with nearly 90% coming from renewable sources.74 Seven different energy sources (solar, wind, solid bioenergy, nuclear, oil, hydro, natural gas with CCS) will provide at least 20 exajoules (EJ) of energy each, compared with six (oil, unabated coal, unabated natural gas, solid bioenergy, nuclear, hydro) providing that level today.75 With the uptake of weather- dependent energy sources, the variability of energy generation will significantly grow. Nearly 70% of global electricity generation in 2050 is expected to come from variable solar and wind.76 SYSTEMS LEVEL The geographic dispersion of energy generation will increase as well, and distributed power generation will play a major role. The IEA estimates that distributed solar PV generation could increase more than 20 times by 2050, going from 320 TWh in 2020 to around 7,500 TWh.77 That would account for roughly one-third of today’s total global electricity demand.78 With the introduction of two-way power flows, distributed generation will have implications for the shape of the energy system. In some instances, bulk power—large, centralized generators—may be augmented with smaller, variable generators.79 In the residential sector, the source of power will primarily be solar PV. In the industrial sector, sources will include solar PV, wind, and combined heat and power systems.80 Energy systems and regulation will need to evolve to support two-way power flows. Already today, leading distribution networks, such as those in South Australia and Queensland, are becoming net exporters to the energy system because of the high penetration of rooftop solar.81 These systems have introduced dynamic operating envelopes to ensure they can maintain grid stability while maximizing solar output. AI may have a valuable role in prediction, simulation, and optimization of this new system. Nearly of global electricity generation in 2050 is expected to come from variable solar and wind 76 70% 50 Powering Possible 49

Badr Jafar CEO, Crescent Enterprises “ AI promises transformational advances across industries, societies, and the environment, but its rapid growth and energy demands strain an already overstretched energy system. Closer collaboration between the AI and energy sectors, alongside governments and civil society, is crucial to unlock sustainable solutions. A multistakeholder approach will accelerate progress, ensuring AI’s full potential is realised in a way that creates inclusive value and leaves no one behind.” Individuals and organizations with distributed energy resources, storage, and flexible demand (including EVs) will continue to go from being electricity customers to being active participants in grid dynamics and energy markets, using millions of digital devices to connect their distributed energy resources to the grid. These devices will enable them to generate, sell, store, and optimize their own electricity profiles.82 On the demand side, up to 1 billion households and 11 billion appliances could contribute to flexibility via demand response and multi-agent systems.83 Balancing supply and demand in these future systems will be far more complex than today. On the supply side, variable sources like solar and wind are intermittent, both regularly (seasonal and diurnal) and irregularly (weather). On the demand side, intraday load variability will be amplified by changes in the location and timing of EV charging. As the amount of dispatchable firm power in energy systems declines, the ability to predict generation and load will be critical to maintaining the stability of energy systems.84 Flexibility management—the ability to balance intermittent supply with variable demand—will also be critical. By shifting intraday load profiles—via storage and demand response, for example—grid operators can reduce peak demand and ensure the optimal load to maintain stability and power quality. The IEA estimates that the energy system’s flexibility must quadruple in a net-zero system.85 This would have the beneficial effect of making power systems more economical, because it would limit the need for additional grid investments and generation sources.86 “ AI is revolutionising energy, just like smartphones changed communication. Our Kraken system manages over 1 GW of flexible demand via 200,000 connected EVs, demonstrating how we can optimise the grid in real time. AI-driven flexibility could save the UK £10 billion a year by 2050, directly lowering energy costs for consumers. This isn’t just innovation; it’s a game-changer for affordable and resilient energy.” Greg Jackson Founder and CEO, Octopus Energy Up to 1 billion 11 billion households and appliances could contribute to flexibility via demand response 83 52 Powering Possible 51

AI has already demonstrated game-changing abilities to predict, simulate, and optimize highly complex systems in several fields. As these abilities continue to develop, there is significant potential for them to build on the value of existing information technology solutions in the energy sector. Of course, as with any new technology, there will be implementation challenges. AI’s capabilities in prediction, simulation, and optimization Simulation AI could simulate the impact of potential scenarios across different components of a future energy system. Digital twins, currently deployed across several industries, could be a major enabler for these simulation use cases to flourish, with AI being used to predict how energy systems will behave under different conditions. By integrating and leveraging data from smart meters and other sources, AI could help gain detailed insights into the energy networks, right down to local distribution networks. This capability would support several operations, including the definition of location- specific, dynamic operating envelopes for distributed energy resources. Collecting and processing myriad data in near real time, running simulations of likely outcomes down to local distribution networks and customer resources, and suggesting optimal demand response interventions to support grid stability and power quality—all of these will be impossible without new AI abilities.89 Enerjisa Üretim, Turkey’s leading power generator, leveraged Microsoft's cloud and AI capabilities to create 3D digital twins of its power plants to assess plant health (e.g., turbines, heat pumps) and risks in real time.90 Prediction Improved forecasting could address many challenges related to energy system variability. Anticipating changes on the supply side (such as the impact of weather on wind output) and the demand side (such as changes in consumption at five-to- ten-minute intervals) could provide significant value for generators, grid operators, and consumers. AI-based weather models are improving rapidly and—in combination with physics-based models—have the potential to issue more precise and more reliable forecasts of renewable generation and load demand.87 For example, Amperon, an analytics platform that partners with Microsoft, provides reliably accurate forecasts related to grid demand, meter demand, and asset-level renewable generation forecasts, with the latter able to forecast at sub-hourly intervals.88 Optimization AI could help play an important role in optimizing and integrating distributed energy resources into virtual power plants (VPPs), collecting many distributed energy resources, including solar PV, EVs, and chargers, energy storage, and smart buildings. Like traditional power plants, VPPs can provide utility-scale grid services, so they will be increasingly important as demand increases and transmission is constrained. AI could help integrate energy resources into VPPs in a range of ways, including identifying potential participants and facilitating enrollment. Additionally, AI could enhance VPP performance by analyzing the vast and fragmented data generated from extensive portfolios of distributed energy resources.91 Uplight is one example of a VPP platform in the U.S. that is using AI’s powers for optimization to enhance VPP performance and user engagement.92 Case in point: The airline industry operates within a highly complex and demanding environment. There is a need to optimize across many dimensions in this industry, including flight schedules, aircraft maintenance, weather conditions, and air traffic control—all while ensuring safety, regulatory compliance, and customer satisfaction. Recently, leading airlines have started using AI-based platforms to optimize flight routes, reduce fuel consumption, predict delays and technical issues, and provide 24/7 support services to customers.93 Solar PV could grow to by 2023, from 25 million today 85 100 million Data complexity As we have already discussed, the structure of the future energy system will be significantly more complex. To cite a few examples: Smart power meters worldwide exceeded 1 billion in 2022, a tenfold increase over the previous decade; connected devices in power systems currently number approximately 13 billion, 13 times more than in the last decade; and grids have approximately 320 million distribution sensors deployed globally today, a number that will likely increase. In the IEA’s Net Zero Scenario, the number of homes relying on solar PV could grow from 25 million today to 100 million by 2030. All these pieces of hardware generate and rely on data. This structural and data complexity could increase volatility in the energy system. Without effective management, we could see price fluctuations, increased outages, and a heavy strain on energy infrastructure. How AI’s optimization and analytic capabilities can help Advanced AI-based management tools could mitigate these challenges. For example, AI algorithms have the potential to significantly enhance grid management by assessing real-time weather data, forecasting supply and demand, and automating control systems. U.S.-based LineVision uses AI and computational fluid dynamics to integrate weather data with real-time sensor measurements, providing dynamic line rating. This capability helps power companies maximize the carrying capacity of grid infrastructure and integrate renewable energy supplies. Using LineVision tools, in just one instance, the UK’s National Grid has unlocked up to 600 MW of offshore wind capacity, increasing line capacity by up to 60%.94 Markus Krebber CEO, RWE AG “ The build out of AI needs close cooperation with the energy sector. The data centres need reliable power infrastructure and clean power supply. However, AI will also play a crucial role to accelerate the energy transition. It will help to integrate more renewables into the grid, will help to make the grids more efficient and stable. AI will also improve the prediction of renewable generation profiles as well as the respective response from storage, flexible generation and the full utilization of demand side response.” 54 Powering Possible 53

Jensen Huang Founder and CEO, NVIDIA “ We are at the beginning of a new industrial revolution, one that will capitalize on the twin technologies of accelerated computing and artificial intelligence. Artificial intelligence is already beginning to transform the energy grid, driving improved grid reliability and energy efficiency, and also supporting integration of renewable energy sources. AI is a critical solution for modern energy production and delivery, enabling greater productivity and economic growth.” The energy system will need to be more efficient and resilient, and supported by the right infrastructure OPERATIONAL LEVEL The efficiency premium Every unit of energy must be used (for example, on-site data center), distributed (for example, over the grid), converted to storage (for example, long-duration batteries), or converted to a new vector (for example, hydrogen). Distribution and conversion cost energy—with some conversions being much less energy-efficient than others. A more complex energy system will place a premium on efficiency. There is significant potential. The IEA estimates that more efficient systems could reduce energy losses in transmission and distribution in today’s energy system to 5% worldwide, compared with as much as 18% in some regions today.95 This improvement could reduce energy- related emissions by over 400 MT,96 roughly equivalent to Australia’s total annual emissions today. Identifying and ensuring the next- most-efficient source for a unit of energy is highly complex. The optimal solution in each instance must account for the level and location of current and future demand and supply, the energy efficiency of the alternative options, multi-commodity energy prices and market conditions, grid stability, and carbon efficiency. All these considerations imply an immense complexity with a very large decision space for how to operate our energy system. AI can play an important role in building resilience and efficiency of energy infrastructure. How AI could help improve efficiency AI’s ability to drive efficiency through power-market optimization is improving. In the future, AI’s new and improved abilities could help market participants optimize consumption, costs, and revenues as system complexity rises. For example, industrial players could use AI to navigate multiple commodity markets and optimize participation, while smaller players like small and medium-sized enterprises (SMEs) and households would benefit from enhanced consumption management and domestic supply (e.g., distributed solar electricity generation) export. Additionally, AI could play a central role in market-making, facilitating coordination and negotiation to meet individual participants’ needs while optimizing overall network performance. The data platforms that optimize these systems will be critical, enhancing operational efficiency and providing clear price signals. Initiatives like Denmark’s DataHub, Germany’s SMARD platform, and frameworks proposed by the UK’s Energy Data Taskforce demonstrate the potential for information technology to improve efficiency.97 Of course, the potential for AI to improve efficiency will be constrained, at least in part, by the underlying infrastructure and market structures and where the incentives of players are well aligned. These must be improved in parallel. Adopting more efficient systems can reduce energy- emissions by over roughly equivalent to the total annual emissions of Australia today 96 400 MT 56 Powering Possible 55

The infrastructure requirements As the energy system becomes increasingly electrified and interconnected, it will require a robust infrastructure capable of handling greater loads, enhanced digital connectivity, and a growing number of supply and demand points. How AI can help with planning AI-driven forecasting, simulation, optimization, and visualization tools could support better and quicker decision-making for the next generation of infrastructure planning. This is based on AI’s proven ability to simulate multiple scenarios and intervention options despite high uncertainty, across different time horizons, and with fine geographical detail. Developing and building energy systems requires complex planning and a wide set of stakeholders, so strategic decision-making will be essential. Additionally, grid upgrades will be needed to accommodate EV charging, data centers, and other new electricity demands. One approach is minimizing network upgrades through market signals to locate demand near generation and storage, and vice versa, where possible, as well as flexibility management. EV charging station In the U.S., wildfires over the last half century have cost power companies an average of per year 100 $1 billion The resilience imperative According to an S&P study cited by the United Nations Environment Programme (UNEP), in a moderate climate-change scenario, utilities will be the most exposed economic sector in 2050.98 The energy sector is exposed to both chronic risks like rising temperatures and water scarcity and acute risks like storms and wildfires. Rising temperatures can How AI can help improve resiliency AI’s capabilities for risk and prediction analytics will be vital both for anticipating events before they occur and responding dynamically after they do. AI-powered risk tools can integrate company- or region-specific parameters with larger climate and weather models. Algorithms then construct risk portfolios by approximating the eventuality of different scenarios—for example, estimating if and where an extreme weather event will occur.101 In Canada, Alberta Wildfire is using a tool built by AltaML and powered by Microsoft Azure Machine Learning to help duty officers make strategic decisions and allocate resources to combat wildfires. The AI tool can predict wildfires with 80% accuracy, providing vital intelligence in an increasingly severe environment.102 impact capacity for power plants, which experience reductions of up to 10% on days hotter than 27°C.99 Meanwhile, wildfires can create massive damages for utilities, with trees and debris collapsing onto infrastructure like transmission lines and substations. In the U.S., wildfires over the last half century have cost power companies an average of $1 billion per year.100 Clearly, the energy system will need to become more resilient. 58 Powering Possible 57

Omar Mir International Board Member, WWT “ The promise of AI in energy is about deliberate and informed technological advancement to create a more equitable and sustainable world. By investing in scalable AI solutions tailored to the Global South - such as the power of AI for distributed energy systems and microgrids, we can bring reliable, clean power to hundreds of millions of people across the Global South - powering dreams, fueling innovation, and ultimately unlocking economic potential.” Today, EMDCs, excluding China, account for 38% (160 EJ) of the total final global energy demand. By 2050, according to IEA’s Stated Policy Scenario, they will account for 50% (252 EJ) of primary global energy demand.103 Meeting that demand with reliable, affordable, and sustainable energy is a defining challenge for the energy industry and the global economy. In parallel, EMDCs have the potential to play a central role in the energy value chains of the future, including critical materials, green tech manufacturing, renewable generation, transmission, and storage. Realizing that potential will give EMDCs a greater stake in the energy transformation and have the potential to reduce energy costs for all. For example, Chile holds a third of the world’s supply of lithium and is the largest producer of copper. Both materials are critical to green technologies like solar PV and battery storage, and they position Chile in the global energy transformation.104 Africa holds 60% of the world’s solar resources but currently accounts for just 1% of solar power generation.105 India has the potential to generate 750 GW of solar capacity, roughly nine times what it currently generates.106 In fact, renewables account for nearly two-thirds of all new power capacity additions in EMDCs (again, excluding China) in the IEA’s STEPS by 2030107 (see Figure 14). Similarly, EMDCs will have a critical role to play in energy storage and transmission. Energy value chains are expected to be more fragmented and more regional than those of today. In the Middle East, green hydrogen capacity doubled year-on-year from 2022 to 2023.108 Oman, the UAE, and Saudi Arabia are positioning themselves as global leaders in hydrogen production. Emerging markets and developing countries will play a far greater role in demand and value chains GLOBAL LEVEL Meeting the energy demand of emerging markets and developing countries (EMDCs), and realizing their potential in energy value chains will be critical to global growth and prosperity. EMDCs will account for 50% today 103 38% primary global energy demand by 2050 up from 60 Powering Possible 59

The Global South is abundant with renewables Figure 14 SUPERABUNDANT RE potential >1,000x of total energy demand ABUNDANT RE potential 100–1,000x of total energy demand SUFFICIENT RE potential 10–100x of total energy demand STRETCHED RE potential 1,000 times as much renewable potential as energy demand 62 Powering Possible 61

By 2050 of firms are expected to adopt AI systems and workers are more optimistic and more experienced than their peers in the Global North.109 85% EMDCs will need to overcome significant challenges to meet their demand and achieve their potential. For instance, the extraction and processing of critical minerals can be inefficient and damaging to the local environment. Transitioning away from fossil fuels poses particular challenges for workers and consumers in EMDCs without social safety nets and the fiscal capacity to support them. Moreover, EMDCs are more likely to lack access to the capital and technology needed for such intensive transitions (see Figure 15). The high cost of capital makes financing the energy transformation harder for developing economies Figure 15 How AI Can Help AI cannot solve all the challenges faced by EMDCs, but it can play as important a supporting role for them as in advanced economies. In fact, leaders based in EMDCs are even more optimistic about the potential for AI in the energy system than their advanced economy counterparts (see Figure 16). As per a global survey of 13,100 employees conducted by BCG, by 2050, 85% of Global South firms are expected to adopt AI systems, and workers in these locations are more optimistic and more frequent users than their peers in the Global North.109 Together, this points to huge potential for leadership and skills. Indeed, technology pioneers in the Global South are already using AI in distributed generation, microgrids, and more. SURVEY INSIGHT Two-thirds of EMDC leaders believe AI can optimize energy efficiency across the value chain, and they highlight talent and training along with technology infrastructure as critical for success Figure 16 Respondents from Global South considering AI adoption and use to have an impact on (%)1 Respondents considering importance of different enablers (%)2 66% Optimizing efficiency across energy value chain (from generation/production to end use) 40% 25% 22% 19% 18% 15% 1% 20% 0% Others Investments & funding Energy infrastructure Data access & resources Talent & training Technological infrastructure do not go over red part (this figure specifically has a reduced width) do not go over red part (this figure specifically has a reduced width) >$45,000 GDP per capita $20,000-$45,000 $5,000-$20,000

Pioneers have already started driving the energy transformation in the Global South Figure 17 SOLShare Winner of the Zayed Sustainability Prize 2022 in 'Energy' Connects homes with distributed generation to microgrids and uses AI to optimize peer-to-peer energy trading SOLBox, the company's core innovation, is a bi-directional energy meter that connects homes to the grid; it also links to an app and online platform that use advanced analytics to monitor generation and facilitate trading Headquartered in Dhaka, Bangladesh; has established over 50 microgrids that connect over 6.5K households Husk Power Systems Leveraging AI to support clean microgrids in the Global South Builds and operates mini-grids that supply clean, reliable electricity in remote and rural areas. AI supports grid layout and predictive maintenance, renewable energy forecasting, demand forecasting, and more. Currently has major operations in India, Nigeria, and Tanzania and is poised to expand further in the Global South. SunCulture Using IoT and advanced analytics to optimize solar-powered irrigation systems for farmers Develops and deploys distributed solar systems that connect to battery- powered water pumps, used for irrigation. IoT-enabled equipment collects data related to solar generation patterns and water usage, supported by analytics. Based in Nairobi, Kenya, with products currently used in Ethiopia, Zambia, Togo, Cote D'Ivoire, and more. G42 in partnership with global and local tech players Collaborating with Microsoft and local innovators to build AI capacity in Kenya Joint investment with Microsoft to build comprehensive AI ecosystem in Kenya, totaling $1B. Plans to build state-of-the-art data center campus in Olkaria, Kenya, run entirely on geothermal energy. Partnering with local Kenya organizations to build local-language AI models, and E. Africa innovation hub and more. A digital investment between partners Microsoft and UAE-based G42 of in Kenya to build innovative AI-enhanced system 110 $1 billion Zola Energy Storage and Management Managing hybrid solar and energy storage systems with the help of AI Zola manages smart microgrids that combine solar and battery storage to provide stable power to remote areas. Leverages AI and advanced analytics to predict energy consumption and optimize battery capacity levels. Based in Arusha, Tanzania, with operations in Rwanda, Ghana, Cote D'Ivoire, and Nigeria. Partnerships will be essential to ensure AI is available in EMDCs. For instance, as part of an initiative with the Republic of Kenya’s Ministry of Information, Communications and the Digital Economy, Microsoft and UAE-based G42 recently announced a $1 billion digital investment in Kenya. With the help of local partners, they plan to build a cutting-edge data center, create a digital innovation lab, and develop AI models in local languages (see Figure 17 for further examples).110 “ Access to AI and other advanced technologies that support cleaner energy and lower-carbon growth are essential to levelling the playing field and ensuring a just and equitable energy transition.” Proscovia Nabbanja CEO, Uganda National Oil Company Sources: IEA 66 Powering Possible 65

As it continues to evolve with new and improved abilities, AI has the potential to play an even more valuable role in the energy system of the future. Beyond the development and deployment of AI’s current abilities, entirely new abilities are on the horizon. One example is agentic operation: autonomous or semiautonomous systems designed to make decisions and accomplish goals—on behalf of an individual— within specific environments.111 Technical research in the field of agentic systems has advanced greatly in recent years, and they are significantly more sophisticated than assistive systems, such as ChatGPT, that rely on human guidance to retrieve and process information. With the help of agentic AI, autonomous systems are rapidly advancing in areas like self-driving vehicles, complex games, and research assistance.112, 113, 114 With appropriate safeguards in place, agentic AI has the potential to unlock value in the vastly complex future energy system, particularly in grid management and demand response. In these areas, decisions must be While agentic AI has the potential to provide many opportunities, the potential applications in the energy sector are still in their infancy. More research and learning is needed to understand and refine its potential. For agentic AI systems to scale, especially in complex and technical fields like energy, they must undergo robust testing and be equipped with strong safeguards. made rapidly and accurately based on vast amounts of dynamic and interconnected data at a scale not possible today, even with assistance from AI. For example, in grid management, today’s AI can predict demand across various loads and forecast power availability based on factors like weather and grid congestion. Agentic systems could enable going one step further, assisting humans to automatically activate alternative power sources or storing surplus energy for future use. When it comes to demand response, today’s AI can gather data from devices like EVs, appliances, and smart meters to help human operators allocate power more effectively. Agentic systems, by contrast, could be instructed by humans on when to directly control devices, curbing energy use during demand spikes or charging batteries when energy is abundant. The EU Artificial Intelligence Act, which regulates AI models and systems, sets forth specific requirements for “high risk” AI systems, which include those used in the management and operation of safety components of critical infrastructure, like the energy sector. These requirements include obligations around risk assessments, data governance, transparency, and human oversight.115 In the energy sector, therefore, it is possible that certain agentic AI systems may need some level of human oversight, making them semi-autonomous rather than fully autonomous. 68 Powering Possible 67 AI has the potential to develop entirely new abilities that—if realized—would position it to play an even greater role in the energy system of the future. AI's potential to evolve more advanced abilities

04 THE ELECTRICITY NEEDS OF AI with carbon-free energy Meeting

Data center electricty demand growth rate is expected to be per year between 2023 and 2030 124 3 – 20% Data center electricty demand growth is the global electricity demand growth rate 124 1 – 6 times AI has been under development for decades, but relatively recent improvements in its performance and the associated emergence of more practical and general applications have triggered rapid growth. Breakthrough applications have emerged in fields as diverse as medical diagnostics, robotics, mobility, interactive personal assistance, and finance. The continued development and growth of AI will rely on several enablers, including processor power, algorithmic innovation, high-quality data, data centers, and the availability of carbon-free electricity. While the electricity consumption of GPU chips has been increasing on a per unit basis, it has been decreasing on a per compute basis, which is expected to continue (see Figure 19 and Figure 20). In the short term, electricity demand for data centers is expected to grow 8–23% per year through 2026, doubling its share of global electricity demand to 2.65%.123 Looking forward to 2030, based on IEA and Goldman Sachs data, we have projected that electricity demand for data centers will potentially grow 3–20% per year, equivalent to 1–6 times the global electricity demand growth rate over 2023-2030.124 While AI is emerging as a new driver for data center electricity demand, its current impact remains relatively small, accounting for about 0.02% of global electricity demand in AI’s energy demand Figure 18 AI and data center proportion of global electricity demand Source: IEA, Electricity 2024: Analysis and Forecast to 2026, 2024 Methodology: The pie chart was develop based on data from the IEA Electricity Report 2024 (see figure on page 35), AI-related energy consumption in 2022 is estimated at approximately 5 TWh, and at around 73 TWh by 2026. The other data presented in the pie chart are either explicitly provided in the report or calculated based on these two AI energy consumption estimates. Traditional data center Dedicated AI data center Total data center Cryptocurrency 1.1% (5) 23.9% (110) 19.8% (160) 2.6% (810) 97.4% (29,791) 98.3% (26,620) 1.7% (460) 71.2% (577) 9.0% (73) 2026 2022 75.0% (345) Figures in parentheses are electricity demand in TWh World minus data centers 2022. At current rates of growth, AI is expected to account for 0.24% of global electricity demand in 2026 (see Figure 18).125, 126 Data centers are key infrastructure for hosting AI. Despite the rapid growth in data center operations and AI applications, AI’s energy demand is likely to remain minor on a global scale in the near term. However, this demand growth for continuous and reliable power is beginning to strain certain local grids where data centers are concentrated, making it crucial to plan how energy supply will meet this demand. Global electricity demand is expected to rise at 3–4% per year through 2030, which is approximately equivalent to the total annual electricity demand of Japan.116, 117 The growth of electricity demand is largely driven by increased electrification in households and commercial buildings as well as in transport and industry. Nearly 80% of this demand growth is from EMDCs.118 Electrification is a critical part of the energy transformation because it makes it easier to power sectors like transport, heating, and industry with carbon-free energy. According to the IEA STEPS, global electricity capacity will need to increase 1.7 times by 2030, as compared to 2023.119 In addition to the significant financial investment in power production to meet this rapid rise in demand, upgrades to transmission infrastructure will be necessary to reliably deliver electricity. Data centers, critical infrastructure for AI, are driving part of this growth. Today, data centers account for approximately 1–1.3% of global electricity demand—about 1.5 times the total annual electricity demand of Spain120—crypto adds another 0.4% (see Figure 18). As the infrastructure needed to support AI expands, electricity demand from data centers will rise, though the extent of the increase remains uncertain.121 History suggests that innovation can slow that demand. Between 2010 and 2020, global data center workloads increased by approximately 840%, while data center electricity demand increased by only 10%.122 72 Powering Possible 71

Figure 19 Energy intensity of AI computer chips has been improving historically 1000 100 10 1 0.1 2022 2020 2018 2016 2014 2012 2010 2008 Energy intensity Index of energy intensity of Al computer chips (2008-100, log scale) Chip model release date Source: IEA, What the data center and AI boom could mean for the energy sector, October 2024 Figure 20 Energy intensity of compute is decreasing even as energy consumption of AI chips is increasing on a per unit basis 7 10 14 1.30 0.32 0.20 0.0 0.5 1.0 1.5 0 5 10 15 20 Max Power (kW) Energy intensity of compute (kW/pFLOPs) DGX A100 DGX H100 DGX B200 More advanced GPU generations Max Power (kW) Energy intensity of compute (kW/pFLOPs) Source: Nvidia, DGX Station A100 Hardware Specifications; Nvidia, H100 Tensor Core GPU Datasheet; Nvidia, DGX B200 Datasheet; ADNOC Analysis. 74 Powering Possible 73

Data center electricity demand is expected to account for of U.S. electricty demand by 2030 130 5 – 9.1% Reliably forecasting AI electricity demand beyond a few years is extremely difficult.127 It will depend on how AI models are developed and operated, how technologies advance, and how models evolve. Innovations will be critical to increase energy efficiency in data center design and operations. But according to IEA projections, data centers will remain a relatively small driver of overall electricity demand growth at the global level in the decade to come (see Figure 21). In certain regions, the speed and scale of AI and data center growth has the potential to put more pressure on local electric supplies due to rapid connection requirements that outpace historic planning timelines. The high demand for reliable carbon- free power further intensifies these challenges, requiring careful coordination and strategic planning to ensure that both AI and data centers can effectively integrate into the evolving energy landscape. Figure 22 Data center electricity consumption and growth across markets with high data center concentration For example, in regions where data centers are concentrated, they can account for a significant percentage of the total electricity demand load (see Figure 22). In the EU, electricity demand for data centers is expected to increase at ~9% per year due to digitalization, including AI, and could exceed 5% of the EU electricity demand by 2026. In the U.S.—the largest data center market globally—data centers represent about 4% of the nation’s electricity demand; that is expected to rise to nearly 6% by 2026.128 These rates of growth can be challenging in regions with aging transmission infrastructure and growing competing demands for carbon-free generation. In certain markets, that growth will potentially translate into a large share of grid demand. In the U.S., electricity demand has stayed relatively flat since the turn of the century but is now expected to grow at 1–2% per year, resulting in an overall increase of 15–20% from 2023 to 2030.129 According to the Electric Power Research Institute (EPRI), data center electricity demand is expected to grow 5–15% annually and can potentially account for 5.0–9.1% of U.S. electricity demand by 2030, up from nearly 4% in 2022.130 Power producers and technology companies are building generation capacity in response. American Electric Power is adding 20 GW over the next decade, 15 GW of which is to meet demand from data centers.131 Technology companies are partnering in building capacity with utilities—for example, in Nevada with advanced geothermal and in Virginia with SMR.132 2022 2026 Share of total data center demand in the region (right axis) European Union United States 5% 6% 4% 3% 2% 0% 1% Data center electricity demand (TWh) Share of total data center demand in the region (%) 275 330 220 165 110 0 55 Ireland Denmark 35% 42% 28% 21% 14% 0% 7% 15 18 12 9 6 0 3 Data center electricity demand (TWh) Share of total data center demand in the region (%) Source: IEA, Electricity 2024. Figure 21 While electricity demand increases strongly in IEA’s Stated Policy Scenario (STEPS), uncertainties could push demand further, by up to 15%. 36,000 40,000 44,000 TWh 20,000 28,000 32,000 24,000 2 3 4 6 5 7 8 More heat waves Faster EV growth Faster data centre growth Lower appliance efficiency STEPS Slower EV growth Slower data centre growth Higher appliance efficiency 1 2023 2030 2015 2035 Source: IEA, WEO 2024 76 Powering Possible 75

“ As AI continues to drive growth in data centers, it's clear that the future of energy must be carbon-free. At Quantum Switch, we're committed to ensuring our infrastructure supports this shift. AI demands stable, scalable energy, and investing in carbon-free power sources—like wind, solar, and storage—will not only help meet this demand but also ensure that our data centers remain efficient, sustainable, and resilient in a net-zero world.” Timothy Bawtree Chairman, Quantum Switch Group Total U.S. data center electricity demand could grow between 30–90 GW between 2023 and 2030.133, 134 Estimates suggest that addressing this incremental demand could require $4–$9 billion in investment per year from 2023 to 2030.135 In the EU, data centers could account for about 17% of incremental electricity demand from 2024 through 2026.136 In certain markets, that growth would translate into a large share of national demand. For example, the IEA forecasts that data centers could account for 28% of Ireland’s demand by 2026 and approximately 20% of Denmark’s by 2026.137 In some countries, the possibility to develop data centers further has been limited due to concerns on grid and clean energy capacity availability. Keeping up with this growing demand for carbon-free energy in the regions that support data centers will be critical to ensure that AI growth is aligned with, and helps to accelerate, the global race to net zero. Growing renewable generation can deliver a significant share of this rising demand, but the intermittency of renewables will need to be addressed with storage or reliable non-intermittent energy sources. Further developing such solutions is critical to meet the growing electricity demand, including those of AI with carbon-free energy. We see two main areas of opportunity for energy players to work with technology players: improved data center efficiency and new sources of carbon-free electricity. 78 Powering Possible 77

Improving the efficiency of AI and data centers The IEA has called for energy efficiency to be the “first fuel of choice.” The COP28 UAE Consensus calls for the annual rate of energy efficiency to double by 2030.138 Meanwhile, AI’s energy efficiency has improved, and there is potential for further improvement in these areas: Hardware Investing in the latest energy- efficient processors and storage devices can greatly reduce the energy required to perform computational tasks. One strategy is to optimize the hardware mix, using high- power GPUs for intensive computations and low-power CPUs for other tasks. A future possibility is to use quantum computers, which have the potential to perform some compute-intensive tasks more efficiently.139 While the absolute electricity demand per GPUs is increasing, they are becoming more efficient, meaning they use less electricity per unit of processing power basis, and this trend is expected to continue.140 Data Electricity use can be minimized by reducing the amount of data that needs to be processed or stored through compression techniques and by optimizing storage systems. It is possible, for example, to reduce the amount of “dark data” held on servers, such as obsolete data like unused backups. Another strategy is edge computing, which moves some processing tasks closer to the source of data. Edge computing is designed to reduce the amount of data transmitted to and from data centers, thus lowering energy usage by both networks and data centers. Smaller models Small language models (SLMs) are designed to perform simpler tasks than large language models (LLMs). They can be more accessible to organizations with limited resources and may be more easily fine-tuned to specific needs.141 SLMs are also more energy-efficient because of their lower compute requirements, thereby providing a viable option for users that do not need the power of an LLM.142 Energy management AI can help monitor and optimize energy usage in real time, adjusting cooling systems, workload distribution, and other parameters dynamically to minimize energy consumption. AI systems can also predict when equipment is likely to fail or require maintenance, allowing for proactive intervention that both ensures systems run at peak efficiency and reduces energy waste due to malfunctioning equipment. Demand response This term refers to the practice of balancing power grids by incentivizing customers to shift demand to times when supply is available or utilizing electricity generation or storage on-site to reduce demand from the grid at certain times.143 Technology companies have launched pilot programs, with promising early results. And select Microsoft and Google data centers in the EU used demand response to support the grid during periods of energy scarcity in the winter of 2022 to 2023.144 More research is being done on opportunities for grid operators to optimize grids with greater granularity and efficiency offered by AI. Cooling Replacing traditional air cooling with liquid cooling systems, where coolants are circulated directly to hot components like CPUs and GPUs, can significantly reduce energy use for cooling. Liquid cooling is more efficient at transferring heat, reducing the need for extensive air conditioning. While some liquid cooling systems use a considerable amount of water, many do not. A focus on minimizing both water and energy use is essential, as there is often a trade-off between these resources in cooling strategies. Water-free cooling systems can be a strong option, especially when surplus carbon-free energy is available to power them. Leveraging environmental conditions, such as using cool outside air, can further reduce reliance on energy-intensive mechanical cooling. Segregating hot and cold air streams within the data center to prevent mixing can also improve cooling efficiency. 80 Powering Possible 79

Developing new sources of carbon-free electricity Although AI is projected to be about 0.24% of total global electricity use by 2026, the broader cross- sectoral load growth necessitates the build-out of carbon-free electricity and additional sources of carbon-free energy generation. The technology and energy industries must work to meet the evolving demands on the world’s grids and this build-out of carbon-free energy sources through a combination of direct investment and green procurement (e.g., power purchase agreements). In addition, new sources of energy will need to be connected to new sources of demand through increased transmission and distribution capacity. The integration of software that increases grid transparency and accuracy can help. SURVEY INSIGHT of leaders in energy companies and of leaders in technology companies surveyed are willing to pay more for carbon-free energy 83% 75% 82 Powering Possible 81

Cooling towers of a nuclear plant Wind and solar Wind and solar are expected to provide about 30%–41% of future generation by 2030, according to the IEA.145 Power purchasing agreements (PPAs) with renewable energy providers can signal demand but it will need support from energy industry participants and regulators to ensure expedited build-out. The technology companies that build and run AI models are among the top six corporate buyers of solar and wind PPAs in 2021.146 Masdar has a number of PPAs supplying electricity generated from offshore and onshore wind with hyperscalers in the U.S., Germany, and Spain. Large consumers have the potential to expand the use of PPAs via novel engagements, but grid operators, utilities, utility regulators and others in the electricity industry must also work to bring more carbon- free electricity online to meet the needs of the future grid. In particular, it is crucial to ensure that new capacity is developed in EMDCs to increase electricity access for local communities. Bioenergy and gas with CCS In the short term, new large-scale natural gas and bioenergy plants will potentially be necessary in some markets to meet rapidly growing demand. These plants should be equipped or retrofitted with CCS technologies whenever feasible. Geothermal Geothermal power plants have a high-capacity factor of 90% or more and can provide firm, dispatchable electricity all year long.148 Microsoft and UAE-based G42 have made plans to set up a geothermal- powered data center in Kenya with a potential capacity of 1 GW.149 Similarly, Sage Geosystems will provide Meta’s data centers with 150 MW of geothermal power by 2027.150 Nuclear fusion Another potential source of sustainable energy is nuclear fusion.154 In 2022, a team at U.S.-based Lawrence Livermore National Laboratory’s National Ignition Facility conducted the first controlled fusion experiment that produced more energy from fusion than the laser energy used to drive it. The path to market-ready fusion energy will depend on partnerships and the ability to mobilize resources for scaling.155 In 2023, Microsoft formed the world’s first nuclear fusion power purchase agreement with Helion to buy 50 MW of fusion power, with deployment starting in 2028, when the plant becomes operational.156 Long-duration energy storage Long-duration energy storage (i. e., for eight hours or more) is particularly well-suited to complementing renewables and can be cheaper than increasing solar capacity to cover mornings and evenings. Microsoft and Google recently collaborated with Nucor to accelerate development of carbon-free energy solutions such as long-duration energy storage.147 Nuclear fission Nuclear fission generates carbon-free firm electricity. Microsoft and Constellation Energy have entered into a power purchase agreement to restart the Three Mile Island nuclear power plant. This plant will supply power to data centers for 20 years, starting in 2028, pending regulatory approvals.151 Additionally, there has been progress in developing next-generation nuclear technologies, including small modular reactors with Kairos power and X-energy.152, 153 “ With renewed momentum and support for deploying nuclear plants across countries, financial institutions and the international civil nuclear sector, driven by the strong market signals from technology companies urgently requiring large amounts of clean, baseload electricity, there is a clear opportunity for nuclear energy to sustainably power digital transformation globally.” Mohamed Al Hammadi MD & CEO, ENEC 84 Powering Possible 83

05 REALIZING AI'S POTENTIAL Recommendations for for the energy transformation

Build capacity in the workforce to leverage AI for the energy transformation. 01 Increase collaboration between technology and energy companies to deploy more carbon-free energy while making it more available and more affordable for all. 03 04 02 Invest in AI for the energy transformation with a focus in four key areas: tripling the availability of renewable energy, building a resilient grid, reducing methane emissions, and utilizing carbon capture and storage. 05 Develop AI with and for emerging economies to meet their unique needs. 06 Establish data standards and protocols for AI to better support the energy sector. 07 Advance policy and governance for responsible, sustainable AI and a secure and inclusive transition transformation to a net-zero energy system. Expand and enhance grid capacity and increase availability of carbon-free electricity, especially in locally stressed grids or regions—while continuing to innovate to increase energy efficiency. Recommendations for realizing AI's potential for the energy transformation The potential for the energy and technology sectors to accelerate a just, orderly, and equitable energy transition to net-zero whilst minimizing the emissions footprint of AI is huge. But success is not inevitable: It requires collaboration between both sectors, along with academia, governments, and others to develop the abilities of AI and effectively deploy them in the energy sector. Seven priority areas include: 88 Powering Possible 87

Associations Collaborations and consortiums can provide a structured and durable mechanism for companies to leverage complementary strengths in AI and energy for common objectives. For example, Microsoft has invested $1.5 billion in Abu Dhabi’s G42 to gain G42’s regional expertise while Microsoft’s global reach will help drive innovations across multiple sectors in the region, including energy.157 Commercial engagements Energy and technology companies can partner on commercial engagements with societal co-benefits— for example, working together on investments in new carbon-free energy generation, power-purchase agreements for carbon-free electricity for data centers, and funding research and development. Sectoral initiatives Energy and technology players can participate in existing sectoral initiatives to foster collaboration or create new ones. The First Movers Coalition and the Oil & Gas Decarbonization Charter are just two of the many examples. SURVEY INSIGHT 94% of leaders surveyed agree on the greater need for collaboration at the intersection of energy, AI, and climate. 01 Increase collaboration between technology and energy companies to deploy more carbon-free energy while making it more available and more affordable for all We see three potential mechanisms for collaboration: 90 Powering Possible 89

The rate of new energy patents has halved since the 1970s.158 In addition, private investment in AI for energy declined by over 30% in the last three years.159 02 Invest in AI for the energy transformation with a focus in four key areas: tripling the availability of renewable energy, building a resilient grid, reducing methane emissions, and utilizing carbon capture and storage Technology and energy collaboration can lean in to assist in three areas. R&D Energy and technology companies should openly collaborate in mapping emergent AI abilities to the four key focus areas named above, and then partner in developing practical solutions that will accelerate their development and deployment for the energy transition. For example, energy companies can use their R&D resources to provide opportunities for real-world testing of new materials for carbon capture and storage. Companies can include targets for their company’s particular investment in related R&D in their annual budgets. Scaling Licensing new technologies can accelerate adoption and lower costs for all participants. Open-source sharing is particularly suitable for large problems that require market-wide collaboration to accelerate innovation. This can include technologies with significant societal co-benefits, but also commercial endeavors that demand a shared technical foundation on top of which companies can compete. Commercial- ization Energy and technology companies should collaborate on the commercialization of proven AI-applications for the energy transformation to make them more broadly available in the market. 92 Powering Possible 91

Energy and technology companies are well placed to collaborate on reducing the carbon footprint of data centers. Two priorities stand out: Energy efficiency Technology players should continue to invest in the energy efficiency of data centers and their constituent technologies to minimize the energy demand of AI and its associated emissions. Carbon-free electricity Energy and technology players should collaborate on increasing the availability of carbon-free electricity to data centers and increasing the reliability of the grid. For example, joint investment and power-purchasing agreements in collaboration with grid operators and regulators can be effective means to accelerate build-out of generation capacity and transmission infrastructure. 03 Expand and enhance grid capacity and increase availability of carbon-free electricity, especially in locally stressed grids or regions—while continuing to innovate to increase energy efficiency 94 Powering Possible 93

Energy companies cannot benefit from AI unless they have the skills to identify and implement solutions. Technology companies cannot deliver AI solutions without a better understanding of energy systems and the energy transformation. Joint training and workforce development could accelerate this agenda. 04 Build capacity in the workforce to leverage AI for the energy transformation AI training for energy companies Energy companies have long prioritized technology training. Training on the potential for AI in the energy transformation should be available to a wide range of personnel in both technical and nontechnical functions. By mainstreaming AI, leaders can build a culture of innovation within their organizations—key for attracting and retaining talent. Approximately 80% of AI talent leaves organizations because they want a more dynamic role or do not see opportunities for advancement.160 Energy leaders that build AI-ready organizations will be more successful when it comes to retaining talent and driving their organizations into the future. Leading institutions such as the UAE’s Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) can serve as hubs for advancing AI knowledge and skills within the energy sector. Energy training for technology companies This training should start from the top at technology companies, with leaders establishing carbon-free electricity as the “fourth pillar” of AI development, along with computing, data, and algorithmic innovation. At the product development level, technology companies need to understand AI’s potential for applications in the energy transformation. This will require training in the key challenges of the energy transformation, as well as in the operating environment of energy companies (such as technical standards for efficiency, product quality, and safety). SURVEY INSIGHT 78% of leaders consider talent and training a challenge to adopting and using AI. 96 Powering Possible 95

Approximately 80% of electricity demand growth through 2050 is expected to come from EMDCs, who are also expected to play a key role in new green energy value chains.161 AI solutions must be designed with and for these markets. The areas to concentrate on are: 05 Develop AI with and for emerging economies to meet their unique needs Data centers Today, approximately 85% of data centers are outside EMDCs. This disparity will grow unless corrective action is taken: By 2030, it is expected that the developed economies will add 1,800 data centers, and the EMDCs only 300.162 There has been concentrated data center development in just a few markets. Expanding into emerging markets will enhance sustainability and foster innovation in these new regions. Talent As highlighted in Chapter 3, talent is one of the most important enablers for AI development in EMDCs. It can be fostered through talent exchange programs, networks and mentorship, and the sharing of best practices.163 Both energy and technology companies will find that EMDCs contain pools of ambitious talent. Survey analysis found that the five countries with the highest confidence in the future of AI are all located in EMDCs.164 Training programs should incorporate talent from EMDCs. Models The vast majority of AI models are developed in the Global North, but local data is required to ensure they can be adapted to local realities, which can be highly constrained.165 Technology leaders should partner with EMDCs to design and deploy AI models and solutions that meet the needs and reflect the circumstances of EMDCs.166 Approximately of electricity demand growth through 2050 is expected to come from EMDCs 161 80% 98 Powering Possible 97

Unified data formats Establishing standardized data formats for energy sector-related data (such as consumption patterns, generation data, and grid status) will be critical. Such formats ensure that AI systems can effectively analyze and share data across different platforms, devices, and segments of the value chain. 06 Establish data standards and protocols for AI to better support the energy sector Communication protocols Standardized Application Programming Interfaces (APIs) and communication protocols will be needed to facilitate seamless data exchange between AI systems, smart devices, grid operators, and other stakeholders in the energy ecosystem. This includes protocols like MQTT or OPC-UA for industrial IoT devices. Interoperability frameworks Creating frameworks that allow for the interoperability and integration of disparate types and sources of data (such as weather data, market data, and grid status data) into AI models will enable more comprehensive analysis and decision-making. Such frameworks will be critical for grid operators that sit at the nexus of supply and demand and receive data from many sources. Data standards and protocols are necessary for the efficient flow of data across the energy system. Protocols should focus on pre-competitive zones of the data layer,167 ensuring that proprietary data remains protected. Some sectors have succeeded in balancing data mutualization and competition. Take, for example, software’s Linux open-source operating system, weather forecasting’s public weather databases, and commercial air travel’s open data for aviation safety.168 100 Powering Possible 99

Governments have an opportunity to accelerate the energy transition by crafting standards and policies that harness the capabilities of AI and help ensure alignment with sustainability outcomes as well as its safe, secure and responsible use. 07 Advance policy and governance for responsible, sustainable AI and a secure and inclusive transformation to a net-zero energy system Enhanced reporting protocols Policies that encourage and leverage AI and digital tools for energy and environmental-related reporting can enhance accuracy, reduce costs and quickly adapt to shifts and differences in metrics and accounting. Frameworks and policies for AI in the energy sector It will be important to develop thoughtful standards and policy frameworks that govern the use of AI in the energy sector, ensuring compliance with existing energy laws, safety standards, and environmental regulations, following industry best practices for AI and implement government-led AI safety standards and policies. Greater collaboration Facilitate collaboration and partnerships among companies, AI researchers, scientists, and policymakers by creating a regulatory sandbox to address sustainability challenges collaboratively and by funding interdisciplinary projects that combine AI expertise with policy, sustainability, and social justice. 102 Powering Possible 101

United Nations, Millennium Development Goals Report, 2015. World Bank, Life Expectancy at Birth (Years), available here (https://data. worldbank.org/indicator/SP.DYN.LE00. IN). World Bank, Literacy Rate, Adult Total (% of People Ages 15 and Above), available here (https://data.worldbank. org/indicator/SE.ADT.LITR.ZS). IEA, World Energy Outlook, 2024. Our World in Data, Global primary energy consumption by source (citing data from: Energy Institute—Statistical Review of World Energy (2024); Smil (2017)), January 2024. 197 countries and the European Union. COP28 UAE Presidency, The UAE Consensus, COP28 Dubai, 2023, available at here (https://www.cop28. com). IEA, World Energy Outlook 2024, 2024. IEA, Electricity 2024, January 2024. Stanford University, AI Index Report 2024, Human-Centered AI Institute, 2024, available at here (https://aiindex. stanford.edu). IRENA, Renewable energy statistics 2024, 2024. ORE Catapult, Accelerating Offshore Wind: The Role of Innovative Technology in Decision-making and Faster Consenting, 2023, report available at here (https://ore.catapult. org.uk). IEA, Electricity Grids and Secure Energy Transitions, October 2023. Payton, “How AI Can Help Power Energy Systems in the Age of Renewables,” Reuters, 18 December 2023, available here (http://reuters. com); Byrne, “How artificial intelligence is helping Ghana plan for a renewable energy future,” Nature Careers Podcast, 7 May 2024, available here (https://doi.org/10.1038/d41586-024- 01316-w). Global Methane Pledge, Global Methane Pledge Declaration, launched at COP26, Glasgow, 2023, available here (https://www. globalmethanepledge.org). Includes U.S., China, Russia, Iran, India, Kazakhstan, Turkmenistan, Indonesia, Pakistan, Nigeria. IEA, Methane Tracker, available here (https://www. iea.org/data-and-statistics/data-tools/ methane-tracker-data-explorer). ADNOC, Survey of Global Leaders on AI and Energy, with approximately 400 participants. Bolgar, Discoveries in weeks, not years: How AI and high-performance computing are speeding up scientific discovery, Microsoft, 9 January 2024, available here (http://news.microsoft. com). Sandalow et al., Artificial Intelligence for Climate Change Mitigation Roadmap, December 2023, available here (https://www.icef.go.jp/roadmap). Goldman Sachs, AI Investment Forecast to Approach $200B globally by 2025, 1 August 2023, available here (https://www.goldmansachs.com/ insights). IEA, Electricity 2024, January 2024. Ibid. Ibid. Our World in Data, World Bank, World Bank World Development Indicators, 2023; Bolt and van Zanden - Maddison Project Database 2023, University of Gronigen; Maddison Database 2010; Oxford Economics (GDP data from IMF Data Portal-1980 onwards); ADNOC Analysis. IEA, World Energy Outlook 2024, 2024. Acemoglu, The Simple Macroeconomics of AI, 2024, National Bureau of Economic Research, May 2024; Briggs & Kodnani, Generative AI Could Raise Global GDP by 7%, Goldman Sachs, available here (https:// www.goldmansachs.com/insights). Smith & Nakagawa, Accelerating Sustainability with AI: A Playbook, Microsoft, 16 November 2023, available here (https://blogs.microsoft.com). Lam et al., “Learning skillful medium- range global weather forecasting,” Science, 14 Nov 2023, Vol 382, Issue 6677, pp. 1416-1421. Kurth et al., “FourCastNet: Accelerating Global High-Resolution Weather Forecasting Using Adaptive Fourier Neural Operators,” Association for Computing Machinery, June 2023. Bolgar, Discoveries in weeks, not years: How AI and high-performance computing are speeding up scientific discovery, Microsoft, 9 January 2024, available here (http://news.microsoft. com). Hu, “ChatGPT Sets Record for Fastest-growing User Base,” Reuters, 2 February 2023. Beauchene et al., “AI at Work 2024: Friend and Foe,” BCG Insights, 26 June 2024, available here (https://www.bcg. com/publications). Our World in Data, Average temperature anomaly, Global (citing data from: Met Office Hadley Centre (2024) – “Lower” [dataset]. Met Office Hadley Centre, “HadCRUT5 HadCRUT.5.0.2.0” [original data]), October 2024. IPCC, Climate Change 2023: AR6 Synthesis Report, IPCC, 2023, available here (https://doi. org/10.59327/IPCC/AR6- 9789291691647). Berns et al., “A Blueprint for the Energy Transition”, BCG, 2023, available here (https://www.bcg.com/publications). IEA, World Energy Outlook 2024, 2024. OGDC, Oil & Gas Decarbonization Charter, launched at COP28, Dubai, 2023, available here (https://www. ogdc.org). Clean Air Task Force, Reducing Methane Emissions Is the Fastest Way to Slow Global Warming, 2024, available here (https://www.catf.us). United States Environmental Protection Agency, Global Methane Initiative: Importance of Methane, 2024, available here (https://www.epa.gov/gmi). Global Methane Pledge, Global Methane Pledge Declaration, launched at COP26, Glasgow, 2023, available here (https://www. globalmethanepledge.org). OGDC, Oil & Gas Decarbonization Charter, launched at COP28, Dubai, 2023, available here (https://www. ogdc.org/). Kessel & Tabuchi, “It’s a Vast, Invisible Climate Menace. We Made It Visible,” The New York Times, 8 November 2019. The European Gas Research Group (Gerg.EU), Methane Emissions, available here (https://www.gerg.eu/ projects/methane-emissions). Shen et al., “National quantifications of methane emissions from fuel exploitation using high resolution inversions of satellite observations” Nature Communications, 14, 4948, 2023. IEA, Global Methane Tracker 2022, 2022. McGovern, Microsoft and Accenture partner to tackle methane emissions with AI technology, Microsoft, 23 August 2023, available here (https:// azure.microsoft.com/de-de/blog). Růžička et al., Semantic segmentation of methane plumes with hyperspectral machine learning models. Sci Rep 13, 19999 (2023) IEA, Net Zero by 2050, October 2021. IEA, CCUS in Clean Energy Transitions, September 2020. IEA, Carbon Capture, Utilization and Storage, July 2023. Sandalow et al., Artificial Intelligence for Climate Change Mitigation Roadmap, December 2023, available here (https://www.icef.go.jp/roadmap). Argonne National Laboratory, “Argonne Scientists Use AI to Identify New Materials for Carbon Capture.”, February 2024, available here (https:// www.anl.gov/article/argonne- scientists-use-ai-to-identify-new- materials-for-carbon-capture). IEA, Carbon Capture, Utilization and Storage, available here (https://www. iea.org/energy-system/carbon- capture-utilisation-and-storage). Koch el al., “Using Carbon Capture and Storage Digital Twins for Net Zero Strategies”, NVIDIA Developer, 6 April 2023, available here (https://developer. nvidia.com/blog). Geotric, “The Role of AI in Accelerating CCUS to Support Net-Zero”, Geoteric Case Study, available here (https:// www.geoteric.com). IRENA, Renewable energy statistics 2024, 2024. Sandalow et al., Artificial Intelligence for Climate Change Mitigation Roadmap, December 2023, available here (https://www.icef.go.jp/roadmap). U.S. Department of Energy, AI for Energy: Opportunities for a Modern Grid and Clean Energy Economy, April 2024. World Bank, REZoning, The Renewable Energy Zoning Tool, available here (https://rezoning.energydata.info). ORE Catapult, Accelerating Offshore Wind: The Role of Innovative Technology in Decision-making and Faster Consenting, 2023, report available here (https://ore.catapult.org. uk). U.S. Department of Energy, AI for Energy: Opportunities for a Modern Grid and Clean Energy Economy, April 2024, available here (https://www. energy.gov/sites/default/files/2024-04/ AI%20EO%20Report%20Section%20 5.2g%28i%29_043024.pdf). IEA, World Energy Investment 2024, 2024. IEA, Renewables 2020, 2020. NREL, Life Cycle Greenhouse Gas Emissions from Electricity Generation: Update 2021; IPCC - Emission Factor Database (2023) – with minor processing by Our World in Data; Expert interviews; ADNOC analysis IEA, Net Zero Roadmap: A Global Pathway to Keep the 1.5C Goal in Reach, 2023 Update, 2023. Ibid. Koumoutsakos et al., Digital twins accelerate a green hydrogen future, Hewlett Packard Enterprise, 2023, available here (https://www.hpe.com/ psnow/doc/a50007626enw). 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 Endnotes 1 – 67 104 Powering Possible 103

Smith & Nakagawa, Accelerating Sustainability with AI: A Playbook, Microsoft, 16 November 2023, available here (https://blogs.microsoft. com). Bolgar, Discoveries in weeks, not years: How AI and high-performance computing are speeding up scientific discovery, Microsoft, 9 January 2024, available here (http://news.microsoft. com). Prantikos et al., Physics-informed neural network with transfer learning (TL-PINN) based on domain similarity measure for prediction of nuclear reactor transients. Sci Rep 13, 16840 2023. Seo et al., Avoiding fusion plasma tearing instability with deep reinforcement learning. Nature 626, 746–751, 2024. IEA, World Energy Outlook 2024, 2024 IEA, Net Zero Roadmap: A Global Pathway to Keep the 1.5 °C Goal in Reach, 2023. IEA, Net Zero by 2050, October 2021. IEA, Net Zero Roadmap: A Global Pathway to Keep the 1.5 °C Goal in Reach, 2023. IEA, Net Zero by 2050, October 2021. Ibid. IEA, Electricity Information, July 2024. U.S. Department of Energy, AI for Energy: Opportunities for a Modern Grid and Clean Energy Economy, April 2024, available here (https://www. energy.gov/sites/default/files/2024-04/ AI%20EO%20Report%20Section%20 5.2g%28i%29_043024.pdf). U.S. Environmental Protection Agency, Distributed Generation of Electricity and Its Environmental Impacts, April 2024, available here (https://www.epa. gov/energy). SA Power Networks, SA Network Sets a New Mark for Negative Demand, 31 October 2022. U.S. Department of Energy, AI for Energy: Opportunities for a Modern Grid and Clean Energy Economy, April 2024, available here (https://www. energy.gov/sites/default/files/2024-04/ AI%20EO%20Report%20Section%20 5.2g%28i%29_043024.pdf). Bouckaert, The Clean Energy Transition Requires Action on Electricity Demand. IEA Commentary, 10 January 2018, available here (https://www.iea.org/commentaries). U.S. Department of Energy, AI for Energy: Opportunities for a Modern Grid and Clean Energy Economy, April 2024, available here (https://www. energy.gov/sites/default/files/2024-04/ AI%20EO%20Report%20Section%20 5.2g%28i%29_043024.pdf). IEA, Net Zero by 2050, October 2021. U.S. Department of Energy, AI for Energy: Opportunities for a Modern Grid and Clean Energy Economy, April 2024, available here (https://www. energy.gov/sites/default/files/2024-04/ AI%20EO%20Report%20Section%20 5.2g%28i%29_043024.pdf). Ibid. Schafer, Amperon’s Industry-leading Forecasts Now Available to Transact on the Microsoft Azure Marketplace, Amperon Blog, 6 August 2024, available here (https://www.amperon.co/blog). Rapoport et al., “Digital Twins Offer Plant Operators Relief from Energy Crisis”, Engie Impact, 20 October 2022, available here (https://www. engieimpact.com/insights). Enerjisa Üretim, Enerjisa Üretim revolutionizes sustainable energy management with Microsoft Azure and AI, 8 December 2023, available here (https://ms-f1-sites-03-ea. azurewebsites.net/en-us/ story/1711956075760053884- enerjisauretim-azure-energy-en- turkiye) U.S. Department of Energy, AI for Energy: Opportunities for a Modern Grid and Clean Energy Economy, April 2024, available here (https://www. energy.gov/sites/default/files/2024-04/ AI%20EO%20Report%20Section%20 5.2g%28i%29_043024.pdf). Hartnett, “Powering VPPs: The (Optimization) Engine Under the Hood”, Uplight Blog, 26 April 2024, available here (https://uplight.com/blog). Sahota, “Navigating the Skies with AI”, Forbes, 29 March 2024; Airbus, Skywise, available here (https://aircraft. airbus.com/en/services/enhance/ skywise). National Grid, National Grid Trials New Technology Which Allows More Renewable Power to Flow Through Existing Power Lines, Press Release, 20 October 2022. IEA, Sustainable Recovery, June 2020. Ibid. IEA, Introduction to System Integration of Renewables, March 2020; Energinet. DK, What is DataHub?, available at: https://en.energinet.dk/energy-data/ datahub/; UK Government, Energy Data Taskforce, available here (http://gov.uk). UNEP Finance Initiative, Climate Risks in the Power Generation Sector, May 2024. Coffel & Mankin, “Thermal Power Generation Is Disadvantaged in a Warming World”, Environmental Research Letters, 2021; UNEP. UNEP Finance Initiative, Climate Risks in the Power Generation Sector, May 2024. World Economic Forum, Innovation and Adaptation in the Climate Crisis, Insight Report, 16 January 2024. Ho, Fighting Canada’s Record Wildfires with a Combination of AI and Intuition, Microsoft News, 19 September 2023, available here (https://news.microsoft. com). IEA, World Energy Outlook 2024, 2024. Jobet, Chile’s Critical Minerals, Columbia Energy Exchange Podcast, 25 June 2024 Toesland, “Start-ups Powering Up Africa’s Solare Energy Ecosystem”, Africa Renewal, 25 January 2024. ET Bureau, Shining Away: India’s Solar Potential, The Economic Times of India. 27 May 2024. IEA, World Energy Outlook 2021, 2021. Singh, “Middle East’s Clean Hydrogen Capacity Plans Double Year on Year”, S&P Global, 25 September 2023, available here (https://www.spglobal.com). Bauchene et al., “AI at Work 2024: Friend and Foe,” BCG Insights, 26 June 2024, available here (https://www.bcg. com/publications). Microsoft Source, Microsoft and G42 Announce $1B Comprehensive Digital Ecosystem Initiative in Kenya, Microsoft News, 22 May 2024, available here (https://news.microsoft.com). Stanford University, AI Index Report 2024, Human-Centered AI Institute, 2024, available here (https://aiindex. stanford.edu). Ibid. Ibid. The Tesla Team, The Bigger Picture on Autopilot Safety, Tesla Blog, 9 January 2024, available here (https://www.tesla. com/blog). European Commission, Proposal for a Regulation Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act), COM(2021) 206 final, 21 April 2021, available here (https:// eur-lex.europa.eu). US Energy Information Administration, Electricity 2022, 2022. IEA, World Energy Outlook 2024, 2024. Ibid. Ibid. US Energy Information Administration, Electricity 2022, 2022. IEA, World Energy Outlook 2024, 2024. IEA, Global trends in internet traffic, data centres workloads and data centre energy use, 2010–2020, 4 November 2021. IEA, Electricity 2024, January 2024. ADNOC analysis based on data center electricity demand estimates data available in IEA’s Electricity 2024: Analysis and Forecast to 2026, World Energy Outlook 2024, and Goldman Sachs’, Generational Growth: AI/data centers’ global power surge and the Sustainability impact 2024. IEA, Electricity 2024: Analysis and Forecast to 2026, February 2, 2024. According to the IEA Electricity Report 2024, AI-related energy consumption in 2022 is estimated at approximately 5 TWh, and at around 73 TWh by 2026. The other data presented in the pie chart are either explicitly provided in the report or calculated based on these two AI energy consumption estimates. Luers et al., “Will AI accelerate or delay the race to net-zero emissions?”, Nature Comment, 22 April 2024. IEA, Electricity 2024, 2024. IEA, Electricity 2024, 2024. EPRI, Powering Intelligence: Analyzing Artificial Intelligence and Data Center Energy Consumption, 2024. American Electric Power, 2024 Q2 Quarterly Earnings Call, 2024. Reuters, Google partners with Nevada utility for geothermal to power data centers, 13 June 2024; NBC News, Amazon goes nuclear, plans to invest more than $500 million to develop small modular reactors, 16 October 2024. Lee, U.S. Data Center Power Outlook: Balancing competing power consumption needs, BCG LinkedIn Article, 28 June 2024, available here (https://www.linkedin.com/pulse/ us-data-center-power-outlook- balancing-competing-consumption- lee-iz4pe/; EPRI 2024). Based on analysis performed on findings presented in EPRI’s ‘Powering Intelligence 2024’ (30–58 GW based on 80% utilization) and BCG’s ‘U.S. Data Center Power Outlook 2024’ (60–90 GW). 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 Endnotes 68 – 134 106 Powering Possible 105

Davenport, “Generational Growth: AI, data centers and the coming US power demand surge”, Goldman Sachs Research, 29 April 2024. IEA, Electricity 2024, 2024. Ibid. IEA, Greece 2023: Energy Policy Review, April 2023. Nayak, “Microsoft achieves first milestone towards a quantum supercomputer”, Microsoft News, 21 June 2023, available here (https:// azure.microsoft.com/en-us/blog). Nvidia, Product specification sheets (e.g., DGX A100, H100); Smith, “Challengers Are Coming for Nvidia’s Crown”, IEEE Spectrum, 61.10, 4 October 2024. Beatty, “Tiny But Mighty: The Phi-3 Small Language Models with Big Potential”, Microsoft News, 23 April 2024, available here (https://news. microsoft.com). UNESCO, Small Language Models (SLMs): A Cheaper, Greener Route into AI, 18 March 2024. IEA, Demand Response, available here (https://www.iea.org/energy-system/ energy-efficiency-and-demand/ demand-response). CEE Multi-Country News Center, "We are ready to do our part in reducing energy consumption in Europe", Microsoft News, 13 December 2022.;Roach, "Microsoft datacenter batteries to support growth of renewables on the power grid", Microsoft News, 7 July 2022; Mehra, “Supporting Power Grids with Demand Response at Google Data Centers”, Google News, 4 October 2023. IEA, World Energy Outlook 2024, 2024. Bloomberg NEF, Corporate Clean Energy Buying Tops 30GW Mark in Record Year, 31 January 2022. Nucor, Google, Microsoft, and Nucor announce a new initiative to aggregate demand to scale the adoption of advanced clean electricity technologies, 2024, available here (https://nucor.com/newsroom/ google-microsoft-and-nucor- announce-initiative). IRENA, Geothermal energy, available here (https://www.irena.org/Energy- Transition/Technology/Geothermal- energy). Microsoft Source, “Microsoft and G42 announce $1 billion comprehensive digital ecosystem initiative for Kenya,” 22 May 2024, available here (https://news.microsoft.com). Microsoft, Gulf News, Data Center Dynamics, and Meta press releases. Constellation press releases, Bloomberg, WSJ, et al, September 2024. Meta, “New Geothermal Energy Project to Support Our Data Centers”, 26 August 2024, available here (https://about.fb.com/news/2024/08/ new-geothermal-energy-project-to- support-our-data-centers). X-Energy, “Amazon Invests in X-energy to Support Advanced Small Modular Nuclear Reactors and Expand Carbon”, Financing Update, 16 October 2024. IAEA, “What Is Nuclear Fusion?”. IAEA News, 3 August 2023. Ibid. Helion and Microsoft press releases, 2023. 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 Endnotes 135 – 168 157 158 159 160 161 162 163 164 165 166 167 168 Microsoft, “Microsoft Invests $1.5B in G42 to Accelerate AI Development and Global Expansion”, Microsoft Source, 15 April 2024. European Patent Office—Global Patent Index Database, ADNOC Analysis. Stanford University, AI Index Report 2024, Human-Centered AI Institute, 2024, available here (https://aiindex. stanford.edu). Beauchene, “How to Attract, Develop, and Retain AI Talent,” BCG Article, 16 May 2023, available here (https://www. bcg.com/publications). IEA, Net Zero by 2050, October 2021. datacenterHawk; BCG Geoanalytics. Data centers in China not considered in the calculation. BCG & AI for the Planet Alliance, How AI Can Be a Powerful Tool in the Fight Against Climate Change, July 2022. Beauchene et al., “AI at Work 2024: Friend and Foe,” BCG Insights, 26 June 2024, available here (https://www.bcg. com/publications). World Economic Forum, The “AI Divide” Between the Global North and the Global South, 16 January 2024. World Economic Forum, Innovation and Adaptation in the Climate Crisis, Insight Report, 16 January 2024. World Economic Forum, Innovation and Adaptation in the Climate Crisis, Insight Report, 16 January 2024.; The Linux Foundation, OS-Climate. Red Hat, What Is Linux?; NOAA, National Digital Forecast Database; National Business Aviation Association, The Value of Data Sharing, May/June 2021. 107