Energy-AI Nexus: Powering the Next Great Leap for Human Progress

ENERGY-AI NEXUS: POWERING THE NEXT GREAT LEAP FOR HUMAN PROGRESS INTEGRATED ROADMAP FOR SYSTEM-WIDE SOLUTIONS June 2025

3

4 5 TABLE OF CONTENTS 4 The data and evidence in this report was developed by ADNOC and Atlantic Council in advance of ENACT DC to inform the discussion. All quotes from the event are unattributed. 1. Introduction 2. Executive summary 3. Responding to the short-term surge Maximizing the output of existing power generation capacity Raising the load and reliability of transmission infrastructure Enabling demand management from all users Delivering affordable energy for all 4. Delivering a smarter build-out for the long-term Optimally siting the next wave of infrastructure Building and funding a 21st century grid Creating a smart energy economy Accelerating next generation energy technologies 5. Solutions Appendix References 6 8 12 14 16 20 22 24 28 30 34 36 38 40 44 ENACT: A CATALYST FOR ENERGY ACTION ENACT convenes global leaders from energy, investment, and technology with the mission to catalyze energy action for the advancement of sustainable prosperity and human progress. It is a future-focused platform to solve global challenges, build bridges across sectors and geographies and seize new value creation opportunities. Rooted in Emirati identity and principles of inclusivity, openness and mutual respect, it convenes diverse voices to align action plans and accelerate solutions. The inaugural ENACT Majlis was held in Abu Dhabi in November 2024, where over 80 cross-sector leaders defined a five-point action agenda to help realize the transformative potential of AI.

6 7 Energy powers progress. AI unlocks possibility. Together, they represent the most transformative force of our time. We are at a pivotal juncture - where the convergence of artificial intelligence and energy systems will define the next era of human advancement. AI has the potential to accelerate discovery, drive productivity, reshape industries, and change the fabric of everyday life. Yet without abundant and reliable energy, it remains just that - potential. AI doesn’t run on code alone — it runs on gigawatts. While photons are free and abundant, electrons are expensive, meaning that the cost of intelligence will eventually be determined by the cost of energy. In the short term, the energy demand surge from AI will be met by using the existing generating capacity more effectively, through increasing utilization, reducing curtailment, delaying retirements, and connecting new capacity to the grid faster. We need more connected, intelligent and sustainable energy infrastructure, not a patchwork of fixes or incremental upgrades, but bold, integrated, system- wide solutions - designed to scale, built to last, and ready to power the future. Today’s grid is the largest machine in the world but has been underfunded and undervalued and has become overstretched and overregulated. We need to get more out of the existing grid today by using tools such as dynamic line-rating, grid topology, and the strategic siting and use of available storage. Shifting demand from peak hours can help to utilize today’s capacity in a smarter way, leveraging the capabilities of AI to help us while ensuring that we sustain affordable and reliable energy, allowing everyone to prosper. 7 No single energy source will be enough to meet these rapidly growing energy needs. We will need an “and, and” approach, combining cheap renewables with flexible gas and clean nuclear, backed up with storage and supported by a 21st century grid with a “backbone” to move electrons over large distances, batteries that time-shift demand, and synchronized condensers and modern inverters that manage frequency stability. Today’s AI can accelerate the next generation of technology and materials for an optimum future low-carbon energy system. Over time, we have been able to allow our energy system to evolve - yet today we are going through a complete metamorphosis. It is complicated and there are no easy answers. Energy and AI are twin organs, connected to fast-forward the future and accelerate human progress. This is the moment to shape the future and realize this historic opportunity for humanity. ENACT is building a cross-sector community and providing leadership to advance an integrated roadmap for action to realize the full transformative potential of AI. It’s convening cross-sector leaders, bridging disconnects, and catalyzing the solutions that will ultimately deliver the bold, integrated, system-wide solutions that are designed to scale, built to last, and ready to power the next leap for human progress. We need to invest today for tomorrow’s intelligence - tomorrow’s arteries and organs - that will need an estimated $1.5 trillion of annual global spend to 2050. Today’s finance system may not be ready to support the huge surge of capital required over the next 30 years, requiring us to unlock new sources of asset-backed credit to leverage future capacity. Tomorrow’s data centers will be like dropping a city the size of Denver on the map in one go, which will need to be placed not only where the electrons live but also where there is water, land, labor, low exposure to weather events and supportive regulation. However, as with Rome, Denver was not built in a day - its population rose gradually over 70 years to three million people, allowing its energy system to grow with prosperity over time. Data centers will need all that growth at the flip of a switch - and yet when a circuit breaker trips in a house, no one notices, but when a data center load changes, all our lights flicker. INTRODUCTION Energy is the beating heart of the global economy and AI is its brains. Connecting the heart and brain is a growing network of arteries - the power grid - essential to avoid the system blacking out. Unattributed quote The AI-Energy Nexus is not tomorrow’s issue, it is today’s imperative. Unattributed quote

8 9 Today, AI promises to unlock the next wave of economic growth - but without energy for the data centers on which it relies, AI will remain no more than potential. In June 2025, leaders from energy, technology, finance and government convened to consider what it will take to deliver energy for the growth of AI in the US and how AI can enhance the efficiency, resilience, and performance of the energy system. In the near term, energy infrastructure build-out is struggling to keep pace with growing demand from data centers, with compute capacity expected to grow 45GW and be attributable for 40% of new power demand by 2030. This is equivalent to the power demand of 30 million US households requiring an additional 50GW of installed gas generating capacity or 150GW of new renewables. Projections show that almost 85% of new generation capacity installed by 2030 will come from renewables, which require up to three to four times as much transmission capacity as conventional dispatchable generation. This puts pressure on an aging grid, where 50% of grid transformers are nearing end-of-life, resulting in a number of high-profile system black outs, such as on the Iberian Peninsula in May, 2025. Utilizable generation capacity growth is expected to fall 30GW short by 2030, while permitting, connection queues and long lead times mean new capacity takes 3-7 years to deliver. Grid connection queues stand at 2,600GW, whilst consumers face price pressure and reliability issues in congested regions. More should be done with the existing supply and demand capacity to ensure the US manage data center growth and seize the opportunity of artificial intelligence. ENACT identified four areas of opportunity to meet the short-term surge: 1. Maximizing the output of existing generation capacity: Delaying retirements and re-rating assets, and increasing utilization through reduced curtailments and speeding up grid connections. 2. Raise the load and reliability of the grid: Technologies such as Dynamic Line Ratings and advanced conductors, and battery storage can unlock gigawatts of underused grid capacity. 3. Enable and incentivize demand management: Data centers and consumers can adopt flexible compute strategies, shifting non-urgent workloads to off-peak times or lower-stress regions. 4. Ensure reliable and affordable energy for all: With electricity prices up 26% for US households since 2020, public support for data center-related grid upgrades is at risk. “Beneficiary pays” models and flexible Power Purchase Agreements (PPAs) can shield consumers while accelerating upgrades. 15% Median CAGR (2024-2030) 2% Median CAGR (2024-2030) 6-13% Min-Max CAGR (2030-2050) 1-6% Min-Max CAGR (2030-2050) Data Center compute capacity (IT), US, 2024-2050 (GW) Electricity demand, US, 2024-2050 (TWh) 0 400 200 800 1000 600 2024 2030 2050 2040 Long-term Short-term 0 5000 2024 15000 20000 10000 2030 2050 2040 Long-term Short-term Low-High Range Base Case For generations, energy has been a foundation of progress and prosperity. In the United States, AI is contributing to a surge in the construction of data centers, which are emerging as a new class of energy consumer. In response, more electricity generation capacity is being added than ever before, after several decades of slow growth, and on top of an ageing grid. However, long-term investment, estimated at over $700 billion per year, is being stifled by uncertainty around energy demand requirements and the availability of supply. In both the short and long terms, an integrated approach is needed across generation, transmission, and consumption, to ultimately ensure prosperity for all. The Situation: Responding to the Surge 2025-2030 Exhibit 1: Data center capacity is projected to grow 15% per year to 2030, contributing to growing electricity demand. Beyond 2030, compute capacity and power demand will continue to grow but with increasing uncertainty. EXECUTIVE SUMMARY Percentage of transformers (%) Age (years) 3.5 3 4 2.5 2 1.5 1 0.5 0.0 5 10 15 20 25 30 40 45 50 55 60 35 Exhibit 2: The majority of in-service transformers are between 25 and 45 years old, indicating potential near-term replacement needs.

10 11 ENACT DC drew on the diverse experience and deep expertise of its participants to identify the key areas that should inform a roadmap for action on energy and AI. The opening plenary discussion focused on maximizing the contribution of existing energy infrastructure - specifically, repowering and extending the lifespans of older power plants to bridge the generation gap in the short-term. AI-enabled digital twins and grid-enhancing technologies were identified as effective means, within a pool of options, to address near-term grid constraints. On the demand side, participants discussed implementing flexible compute and incentivizing consumers to manage their energy usage. The importance of public support for data center build-out was well-recognized, and ensuring energy prices remained affordable through transparent beneficiary-pays models. The second plenary shifted focus to the long-term, starting with a discussion of the factors that might determine the location of the next wave of data centers, and moving to the need for more innovative financial mechanisms to more effectively and equitably allocate risks and increase capital flows. The need for smarter regulation that reduces risk and enables faster decision- making was highlighted throughout the day. Roadmap for action 11 EXECUTIVE SUMMARY Investing in the Long Term 2030-2050 Participants emphasized that AI is as crucial for energy as energy is for AI. They focused on AI’s potential to maximize existing capacity, streamline regulatory processes, and help government employees develop skills for an increasingly complex environment. High- quality data emerged as essential across all applications - both for training and deploying specialized AI in the energy sector and for accelerating next-generation energy technologies that operate within physical principles. In short, AI requires a new energy leadership approach. Participants were clear: we must move faster, build smarter, and work more creatively than ever before. The opportunity is enormous, but our window to act is narrow. We must move faster, build smarter and collaborate more creatively than ever before. Unattributed quote We’re not here to talk - we’re here to enact. Unattributed quote 3. Developing a smart energy economy: Buildings, transport, and industry will drive 60-70% of new energy demand to 2050 as the economy electrifies. A smarter energy system - including much greater reliance on AI-enabled demand-response - would unlock capacity for high-value energy users to drive prosperity. 4. Accelerating next-generation technologies: AI has the potential to accelerate advances in material science, carbon management, future generation, and storage technologies. AI can catalyze next-generation technologies, potentially through greater use of open-source research, resulting in a more efficient, less carbon- intensive energy system that powers progress. 1. Optimally siting the next wave of infrastructure: 50% of new data centers are being built in areas with significant constraints. Locating future data centers outside existing clusters would better solve for energy and other resource factors, whilst providing an opportunity to optimize the build-out of increasingly complex power grids and having a positive impact on job creation and carbon emissions. 2. Building a 21st century grid: A step-change in design, investment, and operation is required to deliver an affordable, reliable and sustainable power grid that is connected to reliable sources, efficiently moves electrons, and uses data (photons) as a low-cost way of moving processed energy. In the longer term, power demand will continue to grow but the pace and nature of that growth becomes increasingly uncertain beyond 2030, particularly for AI-driven data center compute, given underlying uncertainties such as compute capacity, its efficiency, utilization, and location. The successful build-out of data centers and AI-related energy infrastructure will require a holistic, data-driven, location-based approach that takes “the path of least resistance”, in conjunction with an economy-wide energy policy and support for next generation technologies. ENACT identified four priority areas for the long-term:

12 13 13 As of late 2024, the number of hyperscale data centers operated by major providers increased to over 1,130, more than doubling over the past five years, with the US accounting for 60% of global data center capacity. Nowhere is this expansion more visible than in Northern Virginia, often referred to as “Data Center Alley,” which hosts the largest share of the global data center market. This growth shows no signs of slowing. This report forecasts that over 45GW of additional compute capacity could be deployed in the US by 2030. This will require up to 150GW of new installed generation capacity, enough to power 30 million additional US households. Increasing data center capacity is growing in tandem with wider economic activity that carries with it significant additional electricity demand. New semiconductor manufacturing investments in the US will add plants with an estimated power need of 200-1,200MW, with nine such plants announced adding up to a total of 4GW of additional load with a total investment of over $300 As of 2024, more than 2,600GW of energy projects were stuck in interconnection queues across the country, a five-fold increase from the average of the last decade, with delays of two to four years before connection. Markets like CAISO and PJM are already experiencing grid congestion and permitting delays that are stalling critical upgrades. In August 2024, CAISO reported that its interconnection queue contains more than three times the capacity needed to meet the state’s 2045 clean energy goals. New generation and transmission capacity, meanwhile, is exposed to supply chain bottlenecks for generation and distribution. Projects face three to seven year lead times due to investment constraints, delays in permitting, interconnection queues, shortages of transformers and other equipment, and overall policy uncertainty. As a consequence, the delivery of electricity in the US is increasingly unreliable and is beginning to raise household electricity prices, risking data centers’ social license to operate. Residential electricity prices have sharply increased since 2020, surpassing 15 cents per billion. Large battery factories - so-called “gigafactories” - can also reach 260MW in demand. These sectors are bringing a new class of highly energy- intensive consumers, similar to aluminum smelters, which typically required dedicated power generation. Switching on a 1GW data center adds a similar energy footprint to that of three million individuals - the rough equivalent of adding the city of Denver to the electricity grid overnight. The US energy system is ill-equipped to absorb this demand surge in the short term, 85% of new installed capacity will come from renewables by 2030. These resources require three to four times the grid capacity of conventional thermal power due to large swings in power generation and low load factors. This is creating a similar spike in capacity growth as seen at the turn of the century when the US switched from coal to gas. Exhibit 3: More generation capacity is being added to the grid than ever before, with the greatest contribution coming from renewables. Transmission capacity is surging to connect new sources of supply and demand and reflecting the transmission intensity of more distributed generation. Exhibit 4: Total generation capacity in interconnection queues increased nearly five times from the 2010-2019 average to 2023, driven predominantly by renewables and storage projects. 1950 1960 1970 1980 1990 2000 2010 2020 2030 Renewables Natural Gas 1950s -1980 1990s 2000s 2010s 2020 -2024 2025 -2030 Coal Nuclear 18 6 25 18 39 70 2 2 5 9 6 4 11 33 60 10 6 7 21 Annualized gross generation capacity additions, 1950-2030 (GW) Relative transmission capacity (Indexed to 1950, 1950-2030) kilowatt hour (KWh) for the past three years. A recent national survey found that approximately 31% of Americans reported difficulty paying their electricity bills last year, up 25% from the previous year. Another study found that $64.4 billion worth of data center projects in the PJM/Northern Virginia region have been blocked or delayed as a result of public opposition. The US must move quickly to respond to the short-term surge in energy demand resulting from the growing demand for computing power. Stakeholders will need to embark on a strategy that optimizes available supply, deploys new tools to manage growing demand, and empowers the estimated $2 trillion of capital needed by 2030 (with $0.3 trillion dedicated just to generation and transmission infrastructure) to limit a degradation of the reliability and affordability of energy that is a foundation of national prosperity. RESPONDING TO THE SHORT-TERM SURGE 3 x5 533 2,592 2023 0 500 1,000 1,500 2,000 2,500 3,000 2010-2019 av. Gas Renewables & storage Other (incl. nuclear, hydro, geothermal) Source: Energy Markets & Policy, Berkeley Lab, Authors calculations x5 533 2,592 2023 0 500 1,000 1,500 2,000 2,500 3,000 2010-2019 av. Gas Renewables & storage Other (incl. nuclear, hydro, geothermal) The opportunity is enormous but the window to act is really pretty narrow. Unattributed quote

14 15 Maximizing the output of existing power generation capacity 3.1 SHORT-TERM SURGE Key Takeaways: • Despite new domestic generation assets coming online, the US may have an equivalent utilizable energy shortfall of ~30GW by 2030. • Without the ability to build new capacity in this time frame due to supply chain constraints, this shortfall will require an optimization of the existing or marginal generation base. • Aging, retiring, or recently disconnected generation assets should be reviewed to bring back available sources of supply, enabled by planning for the capital needed for upgrading, addressing project economics of older generation, and managing operation and maintenance costs of old assets to set public and private investment in ‘asset recovery’ up for success. Many data centers are exploring opportunities for behind- the-meter generation. However, this could intensify strain on the electricity grid as grid operators lose visibility into forecast loads, and introduce risks of synchronization issues across the grid during periods of peak demand or outages, increasing volatility (and therefore prices) for consumers. To meet rising electricity demand and empower data center build-out, the US will need to assess how to close a projected 30GW gap in utilizable generation capacity by 2030. Although 160GW of new utilizable capacity is planned by 2030, 43GW will retire, 16GW is likely to be curtailed, and 30GW will probably experience some delays in completion or connection. While new generation capacity is growing, additional new resources have at least a three to seven year lead time, leaving a short- term gap in supply amidst the forthcoming surge in electricity demand. Managing retirements will require a number of tools. Upgrades, especially for retiring natural gas assets, provide a solution that can boost output and increase efficiency, with timelines stretching anywhere from six to 24 months, depending on supply chain constraints and permitting. For example, 29% of the currently planned natural gas retirements are simple-cycle combustion turbines managed by the Tennessee Valley Authority, which are being replaced with 500MW of aeroderivative turbines. Significantly, combined-cycle power plants with duct burners enjoy higher load factors than those without. Repowering existing solar and wind plants with new PV cells and wind turbines planned for other projects could help increase the generation capacity within grid constraints, with the old PV cells and turbines being repurposed for other projects that do not require the full capacity of the new equipment and are more cost- sensitive. This is akin to second-hand cars, whereby the U.S. buys new, more efficient cars, and the second-hand cars support prosperity in developing countries; however, in this instance, it brings everyone lower-cost and lower- carbon energy. Planned power generation falls short of requirements by 2025-2030 ~160 Additional generation by 2030 Total available ~100 ~30 Required capacity by 2030 0 50 100 150 Curtailment Connection queues Retirement by 2030 Equivalent changes in utilizable generation capacity (GW) Exhibit 5: Although 160GW of new utilizable generation is planned by 2030, delays in connection, curtailment, and retirements could reduce usable capacity to just 100GW, leaving a 30GW gap to projected requirements. Maximizing the utilization of existing generation infrastructure will therefore be essential to bridge the supply gap. Recent generation capacity in the US has been concentrated in solar, wind, and batteries. In 2024, 93% of incremental US electricity generation capacity hailed from those three sources, which despite having competitive levelized cost of electricity (LCOE), zero fuel costs, and rapid construction times, have low load factors, high installation costs, and rising PPA prices amid elevated interest rates, and intermittency. Natural gas, meanwhile, remains essential for flexibility. In 2024, US natural gas generation increased more than solar (58.8 vs. 53 TWh, respectively), yet, gas also faces constraints: competition with LNG exports, slowing Permian output, and turbine shortages. To meet short-term needs, domestic energy planners will likely have to reexamine the pace of retirements in the electricity generation fleet. Though a ten-year low of 7.5GW of capacity was retired in 2024, 12.3GW of assets are planned for retirement in 2025 - a 62% increase, including 2.6GW of natural gas and 8.1GW of coal. For coal in particular, 9.8GW of capacity has been retired in each of the last 10 years - a consequence of increased competition with cheap natural gas and the deployment of renewables, as well as asset age and high operating and maintenance costs. Planned coal retirements in PJM and SERC comprise more than 10% of peak electricity demand, while retirements in the MISO grid near 25%. More can be done to upgrade sites that are pulled out of retirement. Unattributed quote

16 17 Raising the load and reliability of transmission infrastructure Estimated capacity unlocks (GW) Exhibit 6: US Department of Energy estimates that grid-enhancing technologies could unlock up to 300GW of transmission capacity. These technologies have the potential to accommodate additional generation capacity without additional transmission build-out. Advanced Conductors High-performance materials that increase transmission line efficiency and capacity Wires System optimization Dynamic Line Rating Smart systems that monitor conditions on transmission lines in real-time and increasing capacity based on the optimization of grid topology Energy Storage Solutions that store electricity for later use, improving grid stability and enabling renewables integration Virtual Power Plants Digitally managed networks of distributed energy resources (DER), operated collectively as a single source In Scotland, by monitoring the temperature on transmission lines, higher power loads from wind, turbines could be added to existing lines over 95% of the time as they usually coincided with higher wind, which cooled the lines enabling them at times to operate at double their line ratings. In MISO, a company called Pearl Street Technology has used AI to reduce the computation grid simulation analysis needed to re-rate lines from two years to just 10 days. In addition to DLR, advanced conductors, energy storage, and distributed energy resources (DERs) offer readily available, cost-effective opportunities to quickly unlock additional grid capacity over the next five years. DERs (otherwise known as virtual power plants) can increase the capacity and capability of a system by shifting load to accommodate more growth. A recent Wood Mackenzie study showed that up to 217GW of capacity could be added to the US grid from 2019-2028, through a number of DERs, including solar utilization, distributed battery storage, smart thermostats, and heat pumps. This would be equivalent to adding over 10% to the generation capacity in US today. In addition to data center and power generation, implementing these solutions will require more electricians for which the US forecasts a shortage of 800,000 by 2030. Key Takeaways: • To optimize the existing resource base, the transmission and distribution infrastructure capacity requires upgrading, both to improve the efficiency of the grid and to increase the utilization of domestic electricity generation assets. • Several of these opportunities are readily available, including Dynamic Line Ratings, and distributed energy resources such as heat pumps and smart metering. Policy is a critical tool to incentivize the deployment of these technologies. • Battery storage, when sited strategically alongside key demand hotspots or used creatively to support consumer energy reliability, can both alleviate grid stress and improve reliability for data centers and household consumers. Optimizing the energy resource base is dependent on actions that can quickly expand the capacity and reliability of the existing electricity grid. Dynamic Line Ratings (DLR), which varies capacity (line rating) based on ambient temperatures and other conditions can be deployed rapidly and cheaply, serving as a bridging solution by enabling existing assets to increase their utilization, reduce their curtailment, or both. Other companies are using grid topology optimization (which is the WAZE for power flows), which helped reduce the level of curtailment at a wind plant in the US by 77%. In another pilot program, a utility identified $40 million in savings, through the application of simple switches or of a new technology called Smartwise, which changes the resistance in the wires to shift the flow of electrons. Grid -enhancement technologies can help upgrade line ratings - such as when AES installed 40-line sensors to enable and accelerate its plant’s start-up. By easing congestion, they also incentivize deployment of other resources. 3.2 SHORT-TERM SURGE In Scotland, implementation of DLR have allowed transmission lines to operate beyond their static ratings for 95% of the time, unlocking an additional 30% of capacity. 2500 2000 1500 1000 Capability (MVA) 500 0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Cumulative Frequency (Percent of Time) Dynamic Line Rating Static Line Rating The grid was struggling even before the advent of AI. Unattributed quote

18 19 Raising the load and reliability of transmission infrastructure This $21.8 billion, 24-project initiative is targeted to go into service from 2032 to 2034, connecting up to 120GW of new capacity. US construction of high-voltage transmission lines has significantly decreased recently. While COVID undoubtedly played a role, US high-voltage direct current (HVDC) electricity lines will be a critical long-term solution for meeting US electricity needs. In its assessment of transmission corridors, the US Department of Energy estimates interregional transfer capacity needs of up to 700GW, with HVDC lines comprising the bulk of the long-distance capacity. Utility-scale battery storage capacity reached 26GW in 2024. Across the aggregate generation fleet, battery storage will be critical to mitigate grid constraints during periods of peak demand, and more strategic siting of batteries to provide back-up storage and support on-site generation can further address transmission capacity challenges. These systems, often composed of diesel generators, lithium-ion batteries, iron-air, long-duration battery energy storage, or even natural gas microturbines, allow facilities to continue operating during outages or peak grid conditions. On-site systems can enable grid-aligned curtailment without compromising uptime, managing the intermittency associated with 59.5GW per year of anticipated renewable projects between 2025 and 2030. Though these solutions are readily available to alleviate grid constraints today, appropriate coordination with regulators is critical to ensure reliable forecasting of peak loads and grid-management needs. Upgrades to existing infrastructure can be supported by both policy and the utilization of already available capital pools. FERC order 881, for example, requires all providers to adopt ambient-adjusted ratings, paving the way for broader DLR investment and integration. Blends of policy and subsidy can support the feasibility of virtual power plants (VPPs), whether from their treatment as aggregated resources (as designated by FERC order 2222), investment tax credits (ITCs) or public-led pilot projects such as California’s Self- Generation Incentive Program (SGIP), which offers a $1,000/kWh storage incentive for participating customers. Demand management strategies are particularly important for markets like Virginia’s PJM, which currently boasts the largest collection of data centers in the country but has, as a consequence, struggled with demand saturation. As an initial step, Virginia launched a grid management effort in 2025 called the Virginia Grid Reliability Program (VGRIP). VGRIP offers a $11.5 billion fund-matching facility to support financing for demand planning and grid modernization, including the integration of localized solar-plus-storage systems and smart sensors (see case study in appendix.) In addition to improved infrastructure management, grid expansion and interregional transmission are very useful tools. For example, in 2024, MISO approved a grid expansion and long-range transmission plan encompassing 3,631 miles across 15 states, including a 765 kV high-voltage backbone. 18 19 3.2 SHORT-TERM SURGE Exhibit 7: Annual construction of high-voltage transmission lines has fallen from 9,100 miles in 2010-14 to 3,900 miles in 2020-24, despite rising electricity demand. 2010 - 2014 2020 - 2024 2015 - 2019 0 2000 4000 6000 8000 10,000 US construction of high-voltage transmission (345kV and higher) circuit miles laid, by period A digital twin of Chile’s grid allows millions of simulations to optimize dispatch and design. Unattributed quote

20 21 Enabling demand management from all users Increasing energy efficiency and lowering overall grid demand, where possible, will unlock greater headroom for AI as data center demand is estimated to reach 9% of total electricity demand by 2030. Data centers will have an important role to play in the management of demand. Not all data center energy demand - for instance, for inference, training, or experimentation - is alike. High-energy tasks, such as video generation, can demand over 1,000 times the energy of a simple ChatGPT query, which consumes just 0.34Wh. Building mechanisms to consume energy as efficiently as possible based on these features may in some cases allow for improved demand management, through the application of flexible compute architectures that allow for regional task-shifting. Task-shifting dynamically reallocates latency-tolerant workloads, such as model training or batch analytics, to regions with surplus capacity or lower electricity prices. This “follow-the-energy” approach can optimize energy usage while strengthening reliability. Crucially, it allows operators to scale capacity without compounding local grid constraints, making it a viable strategy as peak demand events increase. Managing data center demand can also be supported through consumer choice and AI Energy Scores. Increasing model efficiency over the past year opens opportunities for consumers to select models that are less energy-intensive and are being communicated for public use through the AI Energy Score Project. For example, for simpler text generation or search tasks, consumers can opt for smaller, task-specific models, such as Hugging Face’s DistilBERT, which consume minimal power, avoiding large, more energy-intense models like ChatGPT-4 for simple tasks, which is like “turning on stadium floodlights to look for your keys”. Demand management can also be facilitated by reducing demand adjacent to data centers. This will include the installation of rooftop solar - and, ideally, batteries - at nearby commercial, industrial, and residential sites, easing grid needs. Furthermore, improving energy efficiency near data centers via battery storage, HVAC, heat pumps, or other technologies can ease local transmission bottlenecks, enhance societal buy-in, and improve overall reliability. For example, Vermont-based grid operator Green Mountain Power launched a ‘Powerwall’ program, which allows residential consumers to install household battery storage systems, improving grid resilience much more quickly than by building out new grid infrastructure. Exhibit 8: Energy intensity of queries will depend on the query size, words generated and type of model used. Making 30 ChatGPT queries is not worse than leaving a single LED light bulb on for an hour. Vehicle to Grid (V2G) buses charging at a dedicated depot. Dominion Energy’s V2G program enables these depots to act as batteries to support grid stability given that most school buses are not in use during periods of peak demand. Demand-response (DR) programs can help reduce data center power use by up to 25% during peak times, contributing to grid stability and efficiency. Automated DR (Auto-DR) can make data centers flexible grid assets. PJM allows curtailment bids and capacity payments, while PG&E offers monthly incentives for load readiness and smart system upgrades. Efforts to address rising demand must consider patterns of peak aggregated electricity demand from businesses and households. For example, annual peak demand periods in the summer and winter are driven by local requirements for heating and cooling. Weekly load distribution is concentrated during working days (Monday - Friday) but tends to increase in the evenings as families return home and household energy consumption rises. Tailoring demand management around these peak demand patterns can alleviate grid stress more effectively, benefiting all consumers. Since 2019, Dominion has also piloted a vehicle-to-grid (V2G) program with Fairfax County schools to further support new models for demand management, reflecting the implementation of demand management strategies in response to peak demand patterns. The V2G system used allows electricity to flow back into the grid from EV batteries via charging stations, allowing for increased efficiency and flexibility for local utilities. Exhibit 9: The electricity grid is built to meet peak demand, even though those extreme hours represent only a small share of total usage. Reducing these peaks - including through flexible compute - can free up capacity. Exhibit 10: Hourly load distribution in the US shows a consistent daily pattern, peaking in the evening across all days of the week. These patterns offer opportunities for targeted demand management and system planning. May Jun Jul Aug Sep 2023 2022 600 500 400 300 600 500 400 300 Midnight 06:00 12:00 18:00 22:00 Hourly load distribution in US, average for May 2024-May 2025 (GW) Peak hourly electricity demand each day, Eastern Interconnection, 2022, 2023 (GWh) The new electricity inputs can be used to aid load management for variable renewable resources, acting akin to a behind-the-meter energy source. Insights collected in 2023 from the pilot indicated that, 95% of the time, participating buses had nearly 3MWh of capacity for grid services, with even higher capacity at night and in the early morning. 3.3 SHORT-TERM SURGE Key Takeaways: • Effective management of data center demand begins with creative approaches to the most intensive parts of AI energy consumption. Models for regional task-shifting, which allow data centers to respond to peak-demand periods, are useful tools. • Implementing demand-response (DR) programs is another sector-agnostic tool that rewards the efficient commercial management of energy demand during peak periods, though policy incentives are required to ensure these efforts are successful. • New technologies - such as vehicle-to-grid (V2G) - can empower more effective demand management across an electrified economy, monetizing underutilized assets. 0.29 0.3 4.9 40 35 30 25 Wh Google Search Typical ChatGPT Query* Long-input Query ChatGPT consumes between 0.3Wh and 40Wh, dependent on the query, prompt and task 20 15 10 5 10.0 20.0 32.0 40.0 10W LED Light Bulb for 1hr Mobile Phone Charge Microwaving for 2 mins Maximum input Query Source: ADNOC, Epoch, IEA

22 23 Delivering affordable energy for all The increase in domestic electricity demand driven by data center build-outs is both rapid and unprecedented, and follows a half-century of decline and stagnation. Between 2025 and 2030, data center power consumption is projected to rise significantly, by about 2% per year, outpacing every other commercial and industrial segment. Though efforts to bring on new supply and expand the grid both aim to alleviate this challenge, the $300 billion of investment needed to meet data center energy needs alone by 2030 ($0.3 trillion of which will be needed just for the generation and transmission infrastructure) begs the question of how to ensure that these costs are not passed on to consumers. One solution is to reevaluate models for upgrade costs. In consultation across utilities, ratepayers, and data centers, stakeholders can assess how grid improvement costs can be shared or pooled, and if artificial intelligence requires a fundamental reassessment of how costs are allocated. Implementing transparent and equitable cost-benefit analyses are crucial for the fair allocation of expenses related to grid upgrades. Inefficient or opaque cost allocation frameworks can inflate total grid upgrade costs by 20-25%, often by encouraging redundant or misaligned infrastructure investments. Infrastructure upgrades should start with fair cost allocation. A “beneficiary pays” model ensures that those who benefit most from infrastructure investments, such as data centers, bear a proportionate share of the costs. This approach can unlock additional capital from the balance sheets of data center operators and aligns with Exhibit 11: In Arizona, extra high-load users - including data centers - are projected to outgrow all other segments - fundamentally shifting the user profile of the grid. Exhibit 12: Electricity prices have risen for all sectors since 2020, but residential customers continue to pay the highest prices despite contributing a lower proportion of load growth. The energy consumption of data centers has also disrupted the traditional patterns of growth in electricity demand. An individual data center can represent a 1GW addition of demand to the electricity grid. That means that - unlike in the past several decades, where demand growth has been incremental - current increases in demand weigh heavily on both prices and reliability. Some estimates, such as the Arizona Public Service Commission, show a potentially massive surge in demand, which simultaneously increases the cost of electricity for data centers and raises energy prices for every other sector too, including household consumers. Since 2020, electricity prices have increased by 22% for industrial users, and by 26% for residential users in the same period. Increasing prices are particularly challenging for residential energy consumers. Their retail prices have outpaced inflation for three consecutive years, even before artificial intelligence-driven demand hits the market. 73% of Americans are now concerned their electricity costs will rise this year, and only 38% say their state government protects their interests when it comes to regulating local electrical or gas utilities. FERC guidance, which mandates that transmission costs are allocated in a manner that reflects estimated benefits. FERC also mandates a six-month cost allocation dialogue with state regulators before finalizing regional transmission plans, ensuring fair cost allocation, preventing cross-subsidies and avoiding the inefficient deployment of capital. Several of these strategies are already in practice. For example, the Midcontinent Independent System Operator (MISO) adopted a “beneficiary pays” approach for its Multi-Value Projects, which distributed $5.2 billion in transmission costs across the load zones that would benefit. In Texas, Senate Bill 6 proposes assigning the costs of new infrastructure to large-scale electricity users — such as data centers and industrial facilities — instead of residential customers. The bill is supported by the creation of a $10 billion Texas Energy Fund, which provides low-interest loans for new capacity while supporting cost recovery directly from major commercial beneficiaries. 2008 5 10 15 20 0 2010 2012 2014 2016 2018 2020 2022 2024 22 Finally, Power Purchase Agreements (PPAs) can support grid flexibility when structured to bundle generation and incorporate curtailment provisions. Traditional fixed- price, fixed-shape Power Purchase Agreements (PPAs) were effective for scaling renewables but now conflict with modern grid needs. They lack the flexibility for grid operators to manage real-time supply-demand imbalances, and often ignore locational constraints, dispatchability, or curtailment risks. This often leads to unnecessary curtailment, overbuilt capacity, and added congestion challenges for data centers and utilities alike. 3.4 SHORT-TERM SURGE Key Takeaways: • Organic growth in electricity demand has been replaced by abrupt increases in energy demand as data centers plug into the grid. • Precipitous and unexpected increases in electricity demand are putting stress on consumer energy prices - with households particularly exposed to affordability concerns. These risks undermine data centers’ social license. • The scale of capital needed to respond to the data center surge does not have to add to the burden on consumers by passing through investment costs through energy prices. “Beneficiary pays” models and innovative PPAs can limit this effect, while also de-risking investment into system upgrades. 5,000 2022 2024 2026 2028 2030 2032 10,000 15,000 20,000 0 Residential Comm + Ind 3 MW Comm + Ind XHLF Electric Vehicles Arizona Public Service Energy Demand Forecast (GWh) Consumers are paying the cost and bearing the risk of connecting data centers. Unattributed quote

24 25 Power demand will continue to grow beyond 2030. But the pace and shape of that growth is uncertain, particularly given the uncertainty surrounding data center build-out, utilization and compute trends. Understanding the number, location, utilization and energy efficiency of new data centers is essential for optimizing the energy system, and in particular the policy and investment decisions needed to begin that process in earnest. For example, the efficiency of data centers - known as power usage effectiveness (PUE) - is expected to improve, including through re-racking or repurposing of existing data centers, but the potential is unclear. In parallel, the models themselves are expected to become more efficient - but the impact will depend, in part, on users. Moreover, the increase in energy efficiency may incentivize increased usage and even higher power demand - known as the Jevons’ Paradox. Compute capacity is projected to grow, but the associated impact on power demand is uncertain. It will depend on a complex interplay between total compute capacity, server utilization rates, and power usage effectiveness, emphasizing the need for smart siting, hardware optimization, and energy-aware design. The composition of data center tasks over the next several decades - distributed between experimentation, inference, and training - also shapes the demand outlook. Exhibit 14: Several factors drive the uncertainty in power demand projections for data centers, including the scale and mix of compute capacity, future of utilization, and infrastructure efficiency. Exhibit 15: The shape and scale of energy infrastructure build-out is also uncertain given uncertainty about the generation mix and how cost, emissions and load factors play into transmission requirements. Exhibit 13: Considerable uncertainty in total power demand by 2050 complicates planning the future energy mix with significant gas and nuclear required in the high case. 3,750 6,450 7,400 8,900 Low scenario Equivalent Transmission Capacity (TW) Base scenario High scenario Solar & wind Gas Others 2.1 2.9 5.1 6,000 17,600 1,500 1,850 5,550 1,300 1,150 3,550 2050 US Power Generation By Source (TWh) Moreover, the shift from training models to active inference tasks, will alter the power demand profiles of data centers. Inference tasks currently account for about 60% of data center energy use, as compared with 30% for training, but the balance may change with increased adoption and more efficient training. Training-focused data centers benefit from large clusters of cutting-edge chips but do not require proximity to end-users or low-latency connections. In contrast, inference workloads can run on older chips and smaller clusters, but often must be located closer to users to meet real-time latency requirements. Data Center Growth Data Center capacity • Total compute required • Mix of Data Center types • Location of future clusters • Pace of build-out • Re-racking existing Data Centers Server utilization • Mix of workload e.g., inference vs training • Utilization rates for different workloads • Growth of edge computing • Chip performance Power usage efficiency • Size of data centers • Power efficiency improvements • Location of future clusters 220 - 951 GW (IT) 20 - 83% 1.05 - 1.79 Energy Infrastructure Growth Generation mix • Relative cost of different power sources • Commercialization of advanced energy tech • State-by-state policy environment Load factor • Mix of generation • Technology & economics may change load factors Transmission • Grid-specific energy mix and load factors • Location of sources of supply and demand • Deployment of grid management technologies 0.7-10 $B/GW 2.1 - 5.1 TW 90% 75% 75% 50% 35% 20% Nuclear Gas Coal Offshore wind Onshore wind Solar DELIVERING A SMARTER BUILD-OUT FOR THE LONG-TERM 4 Range by 2050 Range by 2050 We can’t build 50-year assets with 15-year off-take agreements. Unattributed quote

26 27 The shape and scale of the energy build-out is also uncertain. For example, the generation mix is becoming more diverse, but the final mix and aggregate cost are unknown. Moreover, load factors vary widely across sources — from 90% for nuclear to just 20% for solar — which dictates transmission requirements. Deploying capital against this uncertainty is a challenge. Ongoing capital investments exceeding $260 billion annually are required to build the energy system (on top of the $450 billion per year needed for data centers) but requires de-risking in the context of uncertain policy, technology and demand environments. Investments today might be stranded in 20 years. 27 An optimized energy system must anticipate an AI-driven future, where AI has shifted from being a challenge to being an enabler of energy security, prosperity, and sustainability. Even today, basic smart systems support grid management and offer scalable opportunities that should be fully leveraged. AI can enhance the management, growth, and efficiency of the energy system. In permitting and compliance, AI - and particularly natural language processing (NLP) - can ease staff constraints through automated document review and regulatory interpretation. Future applications may include forecasting and managing wholesale market dynamics and interregional transmission. Finally, AI might be the catalyst for the breakthrough energy technologies in the future energy system - whether that be battery storage, carbon management, SMRs or even fusion - through physical AI models that can tackle advancements in materials science, supporting engineering and design, as well as modeling and analyzing diagnostic data. DELIVERING A SMARTER BUILD-OUT FOR THE LONG-TERM 4 The road to AI is paved with great data. Cutting red tape is not enough; government needs to reduce the cost of capital through offtake guarantees. Unattributed quote Unattributed quote Resolving this uncertainty requires a commitment to optimize future energy systems. This must be governed by an approach to building out energy supply and infrastructure that is constructed around paths of least resistance to accommodate local, regional, and inter-regional supply, transmission, and demand management realities.

29 28 Atlanta Northern Columbus Chicago Des Moines Omaha North Dakota Dallas Phoenix Cluster IT load (GW) Existing clusters 4

30 31 Building and funding a 21st century grid Stakeholder decision-making regarding the siting of data centers will need to be accompanied by a recommitment to building the grid to enable an electrified AI economy to continue to grow. Though the makeup of domestic power generation varies in terms of mix, load factor, and deployment, the ability to efficiently and reliably transport electrons locally, regionally, and interregionally will be critical to limiting disruption and ensuring affordability. That variability also affects the distribution of finance needed across the energy-AI value chain, with incentives needed to build sufficient pools of capital in supply, generation, transmission, distribution, and data centers’ own infrastructure, without overweighting investment in one segment versus another. Future-proofing the domestic electricity grid, however, must acknowledge the age of its existing infrastructure. Specifically, the distribution transformer fleet is aging; approximately 55% of units in service are over 33 years old and approaching end-of-life, while the fleet’s distribution transformer capacity might need to triple by 2050 compared with 2021. This challenge is made only more acute by demand for transformers from data centers (which often require multiple transformer units to convert electricity across voltage levels) alongside the need for grid expansion to support overall capacity demand. This can also manifest as a risk to the security of deliverable electricity. A 2025 power station fire at London Heathrow Airport - which resulted in a widespread electricity outage - was caused by the failure of a 1968 transformer. Exhibit 19: The necessary holistic approach to building the infrastructure required across the energy, power and AI value chain, must account for the varying costs of moving molecules, electrons and photons. Equipment failures of natural gas infrastructure due to a lack of winterization highlight the vulnerability of critical energy systems to extreme cold — a key reliability challenge underscored during winter storms like Uri. Molecules ($$$$) Electrons ($$$) Photons ($) Energy Generation Transmission Demand Fiber optics Similar resilience measures need to be adopted to assure security of supply. While optimized data center placement might take advantage of regional natural gas resources, the utility of these local resources will only go so far if resilience is not considered early on. Texas’ experience during Winter Storm Uri in 2021 is one such example - when the available gas-fired generation declined significantly due to limited winterization, accelerating an electricity crisis in the absence of what was previously considered an uninterruptible baseload. Supply resilience is also critical as more renewables come online. Although Spain has experienced highly successful renewable energy integration, insufficient management of these interconnections may have contributed to the major power outage in May 2025 due to a combination of low-frequency oscillation and voltage fluctuation events. Finally, market connectivity can be a crucial way to ensure that competitive pricing for supply supports resilience and security, exemplified by Ukraine’s decision to resynchronize its grid with Europe’s grid frequency to disconnect its reliance on the Russian grid. Addressing the reliability risk of existing infrastructure can be complemented by wider transmission planning and development. In a high-growth scenario, capacity for regional and interregional transmission will need to grow by 128% and 412% (respectively) from 2020’s installed capacity, according to the US Department of Energy. 31 Key Takeaways: • Realizing AI and data center potential will require a modernized approach to the energy supply chain - one that addresses the age of the existing grid and prioritizes resilience. • Embracing a “build, build, build” approach to transmission infrastructure will be key. Interregional transmission can be a useful tool in ensuring electricity can be accessed reliably and economically throughout the country, regardless of where the demand resides. • Financing these efforts necessitates building pools of capital across the energy value chain to meet the $700 billion of investment needed annually to realize an AI-Energy economy in the US alone. Innovative investment models for finance will be particularly important. Exhibit 20: Data centers are projected to drive 19-25% of total US power sector investments — representing a major shift in energy planning priorities, with a minimum of $49 billion per year directed toward generation, transmission, and distribution to support compute demand. In order for each of these tools to be successful, and with data centers expected to drive 19-25% of all investment in the power sector, stakeholders will need to build new pools of capital to support the estimated $260 billion of annual investment required to build an energy system to support growth and enable AI. Funding a 21st century grid US Power Sector Investments (Billion USD, Annual Avg.) Generation Battery Storage Transmission Distribution Low High 239 264 2025-2030 2030-2050 2030-2050 303 450 60 61 2025-2030 2030-2050 2025-2030 64 79 47 20 118 42 18 115 Data Centers Investments (Billion USD, Annual Avg.) Power Sector Investments related to Data Centers (Billion USD, Annual Avg.) US Power Sector Investments (Billion USD, Annual Avg.) Generation Battery Storage Transmission Distribution Low High 239 264 2025-2030 2030-2050 2030-2050 303 450 2025-2030 2030-2050 2025-2030 64 79 47 20 118 42 18 115 Data Centers Investments (Billion USD, Annual Avg.) Power Sector Investments related to Data Centers (Billion USD, Annual Avg.) US Power Sector Investments (Billion USD, Annual Avg.) Generation Battery Storage Transmission Distribution Low High 239 264 2025-2030 2030-2050 2030-2050 303 450 60 61 2025-2030 2030-2050 2025-2030 64 79 47 20 118 42 18 115 Data Centers Investments (Billion USD, Annual Avg.) Exhibit 21: Annual US power sector investments supporting data center build-outs are projected to total over $230 billion through mid-century, with sustained spending across generation, transmission, storage, and distribution. While generation remains the largest category, to support a more flexible and resilient system, investments in grid infrastructure and battery storage must rise notably after 2030. US Power Sector Investments (Billion USD, Annual Avg.) Generation Battery Stora Transmission Distribution 239 264 2025-2030 2030-2050 64 79 47 20 118 42 18 115 Generation Battery Storage Transmission Distribution Low Hig 239 264 2025-2030 2030-2050 2030-2050 303 450 60 61 2025-2030 2030-2050 2025-2030 64 79 47 20 118 42 18 115 Data Centers Investments (Billion USD, Annual Avg.) Power Sector Investments related to Data Centers (Billion USD, Annual Avg.) US Power Sector Investments (Billion USD, Annual Avg.) US Power Sector Investments (Billion USD, Annual Avg.) Generation Battery Storage Transmission Distribution Low High 239 264 2025-2030 2030-2050 2030-2050 303 450 60 61 2030-2050 2025-2030 64 79 47 20 118 42 18 115 Data Centers Investments (Billion USD, Annual Avg.) Power Sector Investments related to Data Centers (Billion USD, Annual Avg.) US Power Sector Investments (Billion USD, Annual Avg.) Generation Battery Storage Transmission Distribution Low High 239 264 2025-2030 2030-2050 2030-2050 303 450 60 61 2025-2030 2030-2050 2025-2030 64 79 47 20 118 42 18 115 Data Centers Investments (Billion USD, Annual Avg.) Power Sector Investments related to Data Centers (Billion USD, Annual Avg.) US Power Sector Investments (Billion USD, Annual Avg.) Generation Battery Storage Transmission Distribution Low High 239 264 2025-2030 2030-2050 2030-2050 303 450 60 61 2025-2030 2030-2050 2025-2030 64 79 47 20 118 42 18 115 Data Centers Investments (Billion USD, Annual Avg.) Power Sector Investments related to Data Centers (Billion USD, Annual Avg.) 4.2 LONG-TERM BUILD-OUT At our current pace, we won’t have the transmission infrastructure needed for 2035 until 2235. Unattributed quote

32 33 Building and funding a 21st century grid At the core of this challenge are the high costs of capital associated with the development of new scaled energy resources and infrastructure. The cost of rebuilding the grid and adding capacity is substantial (for US, estimated at $50 billion p.a.). Power generation, battery storage, transmission and generation from 2030 to 2050 will require investment of more than $250 billion in the US per year, compared to the $450 billion going into data centers. New economic models will be required to incentivize investments and overcome the overlay of long permitting periods, extended project lifecycles, and the technical risks of advanced technologies. For example, generating capacity and transmission lines typically have a 30-year service life though contract commitments are 10-15 years at best, challenging the investment case for new projects. Establishing clear, bankable cost-recovery mechanisms that provide utilities and private developers with the financial certainty that will enable them to invest at-speed and at-scale may include: • Public private partnership (PPP) models that directly allow for cost-sharing or distributed cost- recovery; • Build-own-transfer (BOT) agreements that allow a private developer to secure initial financing and eventually pass management of new infrastructure to the utility at a pre-arranged price; • Milestone-based arrangements where funding is released incrementally upon completion of specific project phases (e.g., permitting, substation build-out, line energization). Concessional finance and capital stack design can reduce the cost of capital of high-impact, but decidedly longer- term projects, layering different sources of funding (equity, debt, grants, and guarantees) into a hierarchy that reflects each source’s risk tolerance, cost of capital, and repayment priority. Expanding access to credit for energy infrastructure by monetizing the projected cash flows over the lifecycle of an energy asset can underwrite new projects. By assigning immediate value to these future earnings, developers can raise capital up front at investment-grade terms, significantly lowering the cost of financing. In practice, this structure “monetizes US resources” by turning energy potential into tangible balance sheet assets, and secures debt or attracts equity investment. For example, a federal credit facility or public-private vehicle might underwrite a multi-state transmission line by projecting its value based on congestion reduction, market integration benefits, or contracted usage rates. Technology providers themselves are a critical piece of capital formation in new generation and transmission. Though many data center providers have looked to co-location or behind-the-meter solutions out of short- term needs, this may have some risk. In the UAE, Emirates Global Aluminum built 3 gas-fired power plants to meet its independent demand, but this proved highly inefficient and EGA later transferred the assets to the local power operator to access the full grid at lower costs. The financial strength of technology providers may be better suited to develop higher-cost solutions like nuclear or geothermal if provided the appropriate financial incentives. 4.2 LONG-TERM BUILD-OUT We need to standardize and industrialize the build-out of the grid infrastructure. Unattributed quote Unlocking low-cost credit from world wealth Eco Capital Exchange uses an interest-free, investment grade credit system that underwrites alternative energy and transmission projects by monetizing future productive capacity. Companies can unlock value from underutilized assets by converting them into a form of credit known as the ECO—an enterprise-backed, interest-free credit obligation. This model, in effect, allows US resources to be used as assets on its balance sheet to unlock investment. Modeled results show a >10% IRR uplift for infrastructure projects by using a 40- 50% ECO credit, eliminating financing costs. 33

34 35 Creating a smart energy economy Power demand for data centers will continue to rapidly increase but will face competition from buildings, transport, and industry, which will continue to constitute the larger share (60-70%) of growth as electrification continues through 2050, with demand expected to grow faster in transportation and low-carbon technologies as the US shifts to a more efficient and sustainable energy system. In the base case, demand will grow at 3% per annum between 2030 to 2050 but could be double this in a high case scenario, based on our analysis. Exhibit 22: Electricity demand could nearly triple by 2050 in a high-growth scenario, driven by data centers, industrial reshoring, and electrification. Failing to keep pace on the supply- side would force prioritization. 4,550 5,450 4,950 1,650 950 4,000 1,350 1,550 1,100 850 3,500 500 1,150 850 6,000 8,850 17,550 Drivers of demand Efficiency and Decarbonization Powering Progress (Data Centers) US Manufacturing Superpower! Keeping the lights on! Low scenario Base scenario High scenario Security Flexibility RCA Industry Data Center Transport Low carbon technology Currently, Massachusetts and Rhode Island have the most ambitious EERS in the US but require only 2.5% reduction in electricity retail sales annually. Technology deployment could enable greater ambition. Oregon, for example, requires utilities to coordinate with transportation and electrification planning processes and demand-side management plans to proactively implement energy efficiency methods, including smart grids, energy storage, and demand-response programs. Innovative business models can simplify consumer energy efficiency. Schneider Electric and Siemens are deploying Energy-as-a-Service (EaaS) offerings that allow local third-party providers to supply energy through subscription services or performance-based contracts rather than up-front capital costs. EaaS providers supply, manufacture, and build out energy resources, such as renewables, generators, or microgrids, local to the subscribing customer. The customer then pays a recurring fee based on performance or subscription agreements. Consumers can also leverage everyday assets. For example, many electric vehicles can provide Vehicle-to- Home (V2H) services, which are capable of powering a home in an emergency. A Hyundai Ioniq 5 can power the typical electricity usage for a home for 2 to 5 days. Smart controls can shift the demand of energy-intensive appliances such as dishwashers and HVAC to off peak periods. Schneider Energy is building out a Smart Power Manager that allows for demand response through an app. In order to lay the groundwork for an energy system that can sustainably accommodate the growth of data centers, energy efficiency must become a priority. According to recent reports, buildings may be able to reduce their energy intensity by up to 38%, and industrial companies can reduce energy usage by up to 15% through energy efficiency measures and electrification. A more flexible energy system geared towards maximizing the efficient use of energy amongst all energy consumers is critical to the new energy economy. Household and commercial energy efficiency rebates and grants, many of which were established in the Inflation Reduction Act, provided a foundational incentive for investment and participation in energy efficiency programs. The projected increase of energy demand in the US also necessitates a reevaluation of Energy Efficiency Resource Standards (EERS) - state-level policy tools that establish energy efficiency targets for utilities to address load growth, reduce energy waste, and provide grid benefits. Key Takeaways: • The proliferation of data centers is just one class of electricity demand growth. A balanced approach to flexibility and security across the energy system is critical as the economy becomes increasingly electrified. • Energy efficiency enabled by policy and technology can create more demand headroom for data centers as they rapidly add gigawatts demand to the grid, while also improving the sustainability of the energy system overall. • Many current approaches to energy efficiency can then be applied to data centers, increasing the opportunities for data centers to manage their own energy demand and limit their burden on the electricity grid. 4.3 LONG-TERM BUILD-OUT Data centers should set the efficiency standard for the other sectors to follow. We need to think big, act big and build big again. Policy is critical to incentivizing consumers to pursue these opportunities. Energy Star, an energy efficiency program championed by the Environmental Protection Agency (EPA), offers an incentive package for homeowners that includes smart appliances, EV chargers, and smart security measures. Data centers are ideal candidates for these efficiency strategies. New EERS’ can be tailored to incoming data centers as they come online to standardize and predict energy use. Individual contracts within EaaS systems allow data centers to tailor their energy requirements against local grid availability, increasing the capability of data centers to respond to grid constraints or support reliability through task-shifting. Unattributed quote Unattributed quote

36 37 Key Takeaways: • AI has the potential to create a positive feedback loop of energy abundance and accelerate technological innovation. • Already, companies are working to use AI as an innovation tool for materials development, decarbonization, battery efficiency, SMR design and even nuclear fusion. • Realizing these opportunities will require targeted funding strategies that encourage creative thinking and greater collaboration, learning from the success of open-sourcing in IT. Accelerating next generation technologies While existing technologies can be scaled to maintain energy supply at pace with data center development, AI can in turn be a catalyst for the next generation of less energy-intensive materials and technologies which can make the energy system fundamentally more efficient, secure, and sustainable. AI-driven materials discovery has expanded significantly through tools, such as Graph Networks for Materials Exploration (GNoME), created by Google’s DeepMind, which has already located 2.2 million new crystalline molecules, 380,000 of which can be used to power future technologies, and 48,000 of which are currently computationally stable. Exhibit 24: AI is accelerating the path to fusion energy by enhancing modeling, plasma control, superconductor development, engineering, and diagnostic analysis - helping unlock breakthroughs faster. As nuclear energy continues to emerge as a necessary energy resource for data centers, AI can, in turn, help to realize the development of fusion energy technologies. Nuclear fusion has the potential to generate four times more energy per kilogram of fuel than fission, without producing a residual waste product. High Performance Computing (HPC) and AI can accelerate the path towards reliable commercial fusion energy. In 2022, Oak Ridge National Labs debuted the Frontier supercomputer, capable of performing one quintillion floating-point operations per second (Flops). AI capabilities coupled with newly developed supercomputing technologies can enable advancements in autonomous decision-making, predictive modeling, and data analysis for fusion energy. AI: Accelerating the pathway to fusion Modeling and simulation Engineering and design Optimizing plasma control Analyzing diagnostic data Developing super- conductors Partnering with the Berkeley Lab, Google DeepMind successfully developed 42 new materials, opening the potential for AI-powered materials discovery and new- age energy technologies. AI-driven opportunities in materials science can leverage open-source approaches, widening the pool of participants and shaping a more sustainable energy system. IBM’s Foundation Model for Materials (MN4M) is a new open- source foundation model that provides public access to materials modeling. According to IBM, projects related to more efficient battery storage materials and less carbon- intensive energy materials are already underway. Incentivizing innovation Innovation, however, often comes at a cost. Enabling funding opportunities and catalyzing innovation are crucial to maintaining the development of new technologies and meeting projected energy demand. While there are currently a number of federally funded grants that are directed towards innovation in the energy sector, AI and data center development could benefit from tailored incentives to promote AI-Energy-specific research and development. Incentives for AI innovation in these spaces are lacking. While many universities sponsor grants and host research competitions, the awards are modest and limited to research faculty. The development of a federal program could catalyze significant developments in both AI and energy sectors as we look to create new technologies and scale energy supply. ‘Mega research competitions’ can be built out in an open-source fashion - for example, leveraging publicly available models from IBM, or recent grant-making opportunities from the Bezos Earth Foundation, to catalyze AI-driven innovation from nontraditional parts of academia and the private sector. A recent study suggests that AI can be used to accelerate fusion energy advancements most effectively in materials selection, the development of high-temperature superconductors (HTS), and design modeling. AI systems can predict material behavior under the harsh conditions within a commercial fusion reactor, improving component resilience and reducing the trial-and-error periods. Companies, such as Commonwealth Fusion Systems, are already utilizing AI to optimize HTS manufacturing and improve anomaly detection accuracy, which can speed up the production of HTS technology. Similarly, First Light Fusion and digiLab are implementing an AI-assisted design process to model and develop machines and laser targets. Exhibit 23: Breakthroughs in material science, catalysis, carbon management, and advanced technologies - from super-conductors and CCUS to small modular reactors (SMRs) and fusion - are essential to building a low- carbon, high-performance energy system. AI can be a crucial enabler of these breakthroughs. Polymers Super-conductors Textiles Steel ... and others ... and others Material science Reduce energy-intensity of materials or find better alternatives Carbon management Reduce CO2 impact of expansion of generation infrastructure Future technology Accelerating the development of advanced generation and storage DAC CCUS Methane capture SMRs Geothermal Fusion Batteries ... and others 4.4 LONG-TERM BUILD-OUT As AI learns and understands physics, it can simulate, design, and operate real-world systems. It’s not either or; we need conventional and renewable. Unattributed quote Unattributed quote

38 39 1. Repowering generation: Upgrading existing power plants can expand capacity with fewer permitting challenges. For example, upgrading Combined Cycle Gas Turbine (CCGT) components can increase the capacity of gas-fired plants by up to 25%. Build investment-backed programs to identify and coordinate repowering opportunities, incl. labour supply chain & permitting. 2. Delaying retirements: Extending the lifespan of older plants can bridge supply gaps with existing transmission infrastructure. Systematically identify and address retirements that are technically and commercially feasible and within policy preferences. 3. Behind-the-meter integration: Optimizing data centers’ existing assets, such as backup generators, batteries, and solar panels, and improving visibility to grid operators can unlock value for all. Pilot data-sharing protocols and grid participation models. 4. Battery placement: Installation of battery storage at optimal locations on the grid increases capacity and resilience. As curtailment has risen in recent years, California plans to increase its installed battery storage from 16 to 52GW by 2045. Encourage investment-backed program to install new or existing batteries in optimal locations to ease congestion around data center clusters. 5. Transmission lines: Implementing dynamic line ratings based on real-time conditions and dynamic modeling of the grid can increase transmission capacity. Pilot best practice AI-enabled dynamic line rating systems on key transmission corridors to enhance data center grid headroom. 6. Flexible compute: Shifting data center computing tasks in time and space can ease grid pressure and improve flexibility. Pilot compute-shifting frameworks based on grid signals, evaluating cost, emissions and policy needs. 1. New clusters: Developing future data centers away from existing overloaded clusters would reduce the strain on energy infrastructure. Curate a Data Center Readiness Index to benchmark optimal locations for data centers based on criteria such as energy availability, water resources, labor markets, infrastructure suitability, regulatory conditions, connectivity, and climate resilience. 2. Permitting: Streamlining and automating the critical but time- consuming permitting processes would accelerate the build-out of essential energy infrastructure. The US DoE has launched a pilot to develop Permit AI, a tool to expedite federal environmental reviews. Develop a GenAI tools to automate project execution, permit generation and review processes. 3. Upskilling: Regulators require enhanced training to manage increased complexity and pace of the energy build- out. The US DoE highlights an urgent need to expand training beyond traditional fields to ensure workforce is equipped with the skills required in the future. Leverage AI to plot career trajectories, upskill public servants and disseminate best practice. 1. Short-term surge 2. Long-term build-out 7. Consumer load-shifting: Pricing signals, alerts and automation are effective tools to incentivize end users to reduce or shift energy consumption during peak hours. California’s proactive alerts during a heatwave swiftly reduced consumption by 2,100MW. Test behavioral interventions and automation tools to scale demand- side flexibility. 8. Equitable cost allocation: Assigning grid upgrade costs based on share of usage ensures fairness and drives efficient siting. Data centers are expected to account for 40% of new electricity demand by 2040 per our analysis, requiring major investments in transmission infrastructure. Develop frameworks for high-load users to pay proportionate costs through pricing and rate reforms. 9. Humanizing energy: Gamification of energy use can drive and reward consumer behaviors that reduce demand during peak hours or shift it to non-peak hours. AI tools connected to consumers’ smart home devices can optimize energy consumption by adjusting the time when the devices are used based on peak times. 4. New investment models: New commercial models and financial mechanisms are essential to unlock capital for energy infrastructure projects at scale. Develop and assess alternative models to finance the build-out and operation of data center-related energy infrastructure. 5. AI digital twins: Addressing vulnerabilities in aging and complex grid systems through AI can be addressed with AI-enabled design and operations. The US incurs an average of 250 power cuts per year for an average of 5 hours. Create a database of grid disruptions along with an AI tool to identify and prevent disruptions learning from similar events around the world. 6. Predictable data center power: Data centers are a new class of energy user posing unique challenges for grid stability and energy affordability due to their intensive energy use. US data centers are projected to account for up to 18% of the country’s overall electricity usage by 2050. Standards and regulation could play an important role around data center efficiency, design and power consumption (including day-ahead nomination of usage). 7. Efficiency standards: Developing efficiency standards and tools for other sectors is crucial to ensure energy efficiency. For instance, the NREL utilizes integrated energy system simulations to co-optimize across multiple energy systems. Build an integrated energy model to help optimize the energy transformation, minimizing capital, resources, energy and carbon intensity. 8. Open innovation: AI has the potential to accelerate innovation related to energy and materials. For instance, new research suggests that High Performance Computing and AI can accelerate the path toward commercial fusion energy. Identify and develop specific areas, such as batteries, SMRs, and fusion, for investment in open- source innovation. HIGH-IMPACT SOLUTIONS AND NEXT STEPS HIGH-IMPACT SOLUTIONS AND NEXT STEPS 5 5

40 41 APPENDIX Northern Virginia, often referred to as Data Center Alley hosts the largest share of the global data centers colocation market. Meeting electricity demand in this cluster is increasingly challenging. Data centers accounted for 26% of Virginia’s statewide electricity in 2023 and could double by 2030. Commercial electricity sales have already risen 40% since 2019—driven largely by data centers. PJM, the regional grid operator, forecasts adequate generation to meet normal peak demand this summer, but cautions that reserve margins could be insufficient under extreme stress conditions as cooling days increase in response to hotter summer temperatures. This stress is reflected in PJM’s 2024 base reserve capacity auction, which saw a year-on-year increase from $28/MW-day to $269/MW-day. Virginia - meeting the surge in demand from data centers Case Study 1 Regional case studies These stresses are beginning to have effects on reliability. In 2024, an incident occurred where 60 data centers simultaneously disconnected from the grid due to a voltage fluctuation, leading to a sudden surplus of 1,500MW of electricity. To address these challenges, Virginia urgently needs new strategies for supply-demand management and steadily improved generation and transmission. This likely begins with the strategic management of demand, specifically demand at and adjacent to data centers. This includes installing rooftop solar and wherever feasible, batteries, across Northern Virginia, better equipping Data Center Alley to handle summertime peaking needs. Similarly, energy efficiency measures - such as installation of more efficient HVAC units and heat pumps, along with other building efficiency measures - could allow scarce electrons and molecules to flow to where they are needed most: at data centers. As an initial step, Virginia launched a grid management effort in 2025, the Virginia Grid Reliability Program (VGRIP). VGRIP offers a $11.5 billion fund-matching facility to support financing for demand planning and grid modernization, including the integration of localized solar-plus-storage systems and smart sensors. Solutions are also emerging from the private sector, including Dominion Energy’s Bath County pumped- storage system (3,003MW), and joint proposals from First Energy and American Electric Power to establish new transmission capacity across the PJM RTO. Since 2019, Dominion has also piloted a vehicle-to-grid (V2G) program with Fairfax County schools to further support new models for demand management. Another Virginia- based company, GridPoint, is offering energy-as-a-service solutions in the form of energy efficiency modeling and dynamic demand response tools for data centers and a wider array of commercial entities. Texas contains some of the world’s best energy resources, but its grid faces increasingly severe challenges due to surging load. Texas is already the largest electricity market, but ERCOT, the largest system operator in the state, anticipates “adjusted” load (load likely to materialize) rising to 145GW by 2031, up from 87GW in 2024. Data centers are the primary demand driver. Meeting this load while ensuring the resilience of supply will be a challenge, particularly to move bulk electrons to the grid. Texas already generates and consumes more electricity than any other state, meaning that as demand grows, the reliability of both existing and new supply will be critical. Texas - the vulnerabilities of supply disruption Case Study 2 This is especially severe given the makeup of Texas’ current generation mix, led by natural gas and renewables (predominantly wind and solar). Natural gas, for example, will be unable to meet this demand growth alone given competition for gas exports - whether for LNG or pipeline exports to Mexico - and delays in turbines for gas-fired generation. Renewables, meanwhile, face significant challenges in winter in particular. These limitations are exacerbated by Texas’ exposure to extreme weather events, with Hurricane Harvey, Winter Storm Uri and Hurricane Beryl being several examples of the systemic effects of supply disruptions to key demand centers. Winter Storm Uri is a particularly potent example. Between February 7 and February 17, 2021 (a period of seasonally low solar generation), a surge in electricity demand for heating and equipment failures for power generation caused nearly 52,000MW of power capacity to go offline. Declines in gas-fired generation as a result of a lack of appropriate winterization were particularly impactful, as was the isolation of Texas’ power grid, limiting its ability to import electricity. In response, Texas has introduced policy to harden generation assets and improve emergency protocols for grid management. Contingency approaches to upstream planning while expanding geographic optionality will further limit systemic risk, though these have yet to be implemented. Virginia electricity balances (GWh) ERCOT projected demand (GW) Virginia electricity demand by sector (TWh) Net Generation All other demand Residential Retail Sales Data Centers Commercial Imports (not accounting for direct use and estimated losses) Industrial

42 43 Because of its longstanding tech and innovation community in Silicon Valley, California seems at first glance to be a natural opportunity for data centers. The alignment of an AI-oriented workforce and considerable state-level support for innovation policy suggests that there will be persistent interest in California as a data centers destination as AI is increasingly used as a tool for innovation in the energy space and beyond. Los Angeles recently built a 33MW critical IT load data center, and more are on the way. California and the challenges of building new infrastructure quickly Case Study 3 Many reasons have caused this problem, including the challenges of land acquisition, political opposition, and - in certain circumstances - California’s challenging topography, which separates areas of viable generation assets and centers of demand. The inability to build major projects extends to California’s inability to expand its workforce, with housing shortages being one specific example. Despite long- standing housing shortages, California continues to lag virtually every state in authorizing and constructing new housing, after normalizing for population. Paired with increasingly high housing costs, California runs an increasing risk of neither having enough people to build data centers infrastructure not offering a competitive enough housing market to attract a data centers-oriented workforce. North Dakota - siting new data centers in areas with high readiness Case Study 4 North Dakota is home to some of the world’s best wind resource, enjoys substantial conventional hydrocarbon production, abundant land, and has a significant IT workforce. As such, this electricity-exporting state has the potential to become a data center cluster, with 19 data centers currently in operation and with new, much larger projects under discussion. North Dakota offers several opportunities which make it an attractive destination for new data centers. Ample real estate for both data centers and generation-distribution buildout, with easily navigable permitting for new construction. As a result, North Dakota has experienced one of the largest increases in commercial electricity demand growth in the nation between 2019 and 2023 at 37%. While as of 2024, North Dakota’s energy production was approximately six times that of its consumption, meeting this demand will begin to place stress on North Dakota’s energy system. Yet, it remains well-positioned to grow both its transmission and data centers management at the same time. Furthermore, in addition to its own abundant resources, North Dakota and neighbouring states can also call upon significant power from neighbouring Manitoba, which has significant hydropower and wind resources of its own. This can create auxiliary supply opportunities as data center demand continues to grow. The state’s population size, however, will be a limiting factor that North Dakota will need to address quickly. Given the population size and the lack of an addressable inference market, North Dakota will likely only be a suitable data center hub for applications - like training - that have low latency requirements. Furthermore, a data center buildout would quickly run into workforce shortages. The state unemployment rate is the second lowest in the country as it suffers from a shortage of workers. Meanwhile, North Dakota’s location offers several of its own advantages. As a relatively cooler climate, the energy needs for data centers cooling are negligible compared to other data centers hubs like Texas or Virginia. It’s also centrally located in the US, making it a highly attractive opportunity to encourage regional load shifting and data distribution. However, California’s significant challenges in building new supply and transmission infrastructure, and the costs of improving existing transmission infrastructure, challenges its attractiveness to new data centers. California has experienced significant renewables penetration since 2019, but as of 2024 this has plateaued, with over 1,000 projects waiting to be connected. This stagnation of new generation reflects severe gaps in transmission and distribution infrastructure that have not been resolved. For example, CAISO has also built just 920 circuit-miles of high-voltage lines, or 4% of all US construction, despite comprising about 6% of total US electricity consumption, facing grid stress, and holding decarbonization targets on paper. In-state Electric Generation by Fuel Type Source: Quarterly Fuels and Energy Reporting Regulations Cumulative High-voltage (345kV and higher) circuit-miles laid, since 2010 Unemployment rates (%) North Dakota monthly electricity generation and in-state sale (GWh) Wind Solar Thermal Solar PV Small Hydro Large Hydro Natural Gas Nuclear Geothermal Biomass Oil Waste Heat Petroleum Coke Coal Rest of U.S. Retail saled CAISO Generation Generation less retail sales

44 45 REFERENCES 1. https://psc.ky.gov/pscecf/2023-00123/mike.hornung%40lge-ku. com/04142023102139/07_-_KU-BOSK_Special_Contract_-_ Contract_FINAL_-_REDACTED.pdf 2. https://emp.lbl.gov/queues 3. https://www.caiso.com/documents/aug-1-2024-tariff- amendment-interconnection-process-enhancements-2023- er24-2671.pdf 4. https://smartenergycc.org/addressing-energy-affordability- infographic/ 5. https://www.data centerswatch.org/report 6. https://www.energy.gov/sites/default/files/2023-12/ National%20Transmission%20Needs%20Study%20-%20 Final_2023.12.1.pdf 7. https://www.utilitydive.com/news/buildings-energy-intensity- renewable-sustainability-report/704245/ 8. https://eta-publications.lbl.gov/sites/default/files/2025-01/l billionl_eers_report_v0113_final.pdf 9. https://www.latitudemedia.com/news/epri-takes-its-data-center- flexibility-project-global/ 10. https://research.ibm.com/blog/foundation-models-for-materials 11. https://firstlightfusion.com/our-technology/ai-and-simulation- technology/ 12. https://www.spglobal.com/commodity-insights/en/news- research/latest-news/electric-power/073024-pjm-power- capacity-auction-clears-at-record-high-price-of-26992mw-day-for- most-of-footprint 13. https://www.data centersdynamics.com/en/news/virginia- narrowly-avoided-power-cuts-when-60-data-centers-dropped-off- the-grid-at-once/#:~:text=Critical%20Power%20Channel- ,Virginia%20narrowly%20avoided%20power%20cuts%20when%20 60%20data,off%20the%20grid%20at%20once&text=Sixty%20 data%20centers%20in%20Northern,equipment%20fault%20on%20 the%20grid.&text=Grid%20operator%20PJM%20 Interconnection%20and,despite%20constraints%20on%20the%20 grid. 44 14. https://www.dominionenergy.com/projects-and-facilities/ hydroelectric-power-facilities-and-projects/bath-county-pumped- storage-station 15. https://www.ercot.com/files/docs/2025/04/07/8.1-Long-Term- Load-Forecast-Update-2025-2031-and-Methodology-Changes.pdf 16. https://www.data centersdynamics.com/en/news/prime- launches-33mw-data-center-in-vernon-california ; https://www. costar.com/article/1281372940/developers-look-to-new- frontiers-for-data-centers-in-the-home-region-of-ai 17. https://www.energy.ca.gov/data-reports/energy-almanac/ california-electricity-data/electric-generation-capacity-and-energy 18. https://www.caiso.com/todays-outlook/supply ; chrome- extension://efaidnbmnnnibpcajpcglclefindmkaj/https://emp.lbl. gov/sites/default/files/2024-04/Queued%20Up%202024%20 Edition_R2.pdf 19. Data from MapSearch, analyzed and visualized by the Atlantic Council 20. US Census Bureau; https://www.bloomberg.com/graphics/2019- california-housing-crisis/ ; https://www.atlanticcouncil.org/ blogs/econographics/housing-costs-are-slowing-down-the-us- climate-transition/ 21. https://www.data centersdynamics.com/en/news/two- companies-seek-to-develop-125 billionai-data-centers-in-north- dakota/; https://www.data centersmap.com/usa/north-dakota/ 22. https://www.eia.gov/state/analysis.php?sid=ND 23. https://www.bcg.com/publications/2025/breaking-barriers-data- center-growth 24. https://www.iea.org/reports/energy-and-ai 25. https://epoch.ai/gradient-updates/how-much-energy-does- chatgpt-use