《兰德公司：至2030年美国新增AI算力容量评估报告（2025）（英文版）（100页）.pdf》由会员分享，可在线阅读，更多相关《兰德公司：至2030年美国新增AI算力容量评估报告（2025）（英文版）（100页）.pdf（100页珍藏版）》请在三个皮匠报告上搜索。

1、ISMAEL ARCINIEGAS RUEDA,HENRI VAN SOEST,HYE MIN PARK,AUSTIN SMIDT,DAVID GILL,ROBIN WANG,KELLY KLIMA,AISHA NAJERAAssessing the United States Additional AI Power Capacity by 2030Estimating Short-Term Increases in Electricity Generation and the Ability to Meet Growth in Power DemandResearch ReportFor m

2、ore information on this publication,visit www.rand.org/t/RRA3845-1.About RANDRAND is a research organization that develops solutions to public policy challenges to help make communities throughout the world safer and more secure,healthier and more prosperous.RAND is nonprofit,nonpartisan,and committ

3、ed to the public interest.To learn more about RAND,visit www.rand.org.Research IntegrityOur mission to help improve policy and decisionmaking through research and analysis is enabled through our core values of quality and objectivity and our unwavering commitment to the highest level of integrity an

4、d ethical behavior.To help ensure our research and analysis are rigorous,objective,and nonpartisan,we subject our research publications to a robust and exacting quality-assurance process;avoid both the appearance and reality of financial and other conflicts of interest through staff training,project

5、 screening,and a policy of mandatory disclosure;and pursue transparency in our research engagements through our commitment to the open publication of our research findings and recommendations,disclosure of the source of funding of published research,and policies to ensure intellectual independence.F

6、or more information,visit www.rand.org/about/research-integrity.RANDs publications do not necessarily reflect the opinions of its research clients and sponsors.Published by the RAND Corporation,Santa Monica,Calif.2025 RAND Corporation is a registered trademark.Limited Print and Electronic Distributi

7、on RightsThis publication and trademark(s)contained herein are protected by law.This representation of RAND intellectual property is provided for noncommercial use only.Unauthorized posting of this publication online is prohibited;linking directly to its webpage on rand.org is encouraged.Permission

8、is required from RAND to reproduce,or reuse in another form,any of its research products for commercial purposes.For information on reprint and reuse permissions,visit www.rand.org/about/publishing/permissions.RR-A3845-1 iii About This Report An exponential increase in computational resources used f

9、or artificial intelligence(AI)training and deployment has enabled rapid advances in AI models capabilities and the widespread use of these models.The resulting exceptional demand for AI data centers is already posing challenges for U.S.data center construction,primarily because it is difficult to fi

10、nd and access reliable power grid capacity.To anticipate the supply-side constraints that could affect the powering of future AI growth,we ask the following questions:What will the United States additional power capacity be by 2030?What are the key constraints on scaling U.S.energy production to mee

11、t additional demand?What is the maximum amount of power available for AI data centers at particular sites or regions by 2030?This report focuses on the first question and provides projections of power-generation capabilities in front-of-the-meter energy systems(systems integrated with wider electric

12、ity grids)and behind-the-meter energy systems(systems located on the customer side of the utility meter)by 2030.Expected available capacity is aggregated by region,expected year the capacity comes online,resource type,and the sources of uncertainty in the estimate.We present policy recommendations a

13、imed at regulators at the federal and state levels,grid planners,and system operators to increase available supply,targeting implementation by 2030.The complete methodological approach is detailed in the accompanying appendixes.Technology and Security Policy Center RAND Global and Emerging Risks is

14、a division of RAND that delivers rigorous and objective public policy research on the most consequential challenges to civilization and global security.This work was undertaken by the divisions Technology and Security Policy Center,which explores how high-consequence,dual-use technologies change the

15、 global competition and threat environment,then develops policy and technology options to advance the security of the United States,its allies and partners,and the world.For more information,contact tasprand.org.Funding This research was independently initiated and conducted within the Technology an

16、d Security Policy Center using income from operations and gifts from philanthropic supporters,which have been made or recommended by DALHAP Investments Ltd.,Effektiv Spenden,Ergo Impact,Founders Pledge,Charlottes och Fredriks Stiftelse,Good Ventures,Jaan Tallinn,Longview,iv Open Philanthropy,and Wak

17、ing Up Foundation.A complete list of donors and funders is available at www.rand.org/TASP.RAND donors and grantors have no influence over research findings or recommendations.Acknowledgments We thank the leadership of the RAND Technology and Security Policy Center,Jeff Alstott and Casey Dugan,for th

18、eir guidance on this publication.We thank Konstantin Pilz for advice on AI power requirements and compute,Alison Hottes for managing the quality assurance process,and our reviewers Keith Crane,David Rode,and Jeff Anderson for their thoughtful feedback.We are also thankful to several industry stakeho

19、lders who provided insights on this topic.v Summary Issue Anticipated growth in artificial intelligence(AI)development requires the buildout of additional power capacity at an exceptional scale and speed.Data center operators often prefer that this additional capacity comes from connections to the g

20、rid because such connections offer improved reliability,resilience,and cost-effectiveness compared with on-site generation alternatives.However,stakeholders have identified major challenges with the current grid,especially concerning the integration of large volumes of energy resources.In its most r

21、ecent long-term reliability assessment,the North American Electric Reliability Corporation(NERC)1 warned that more than half of North America will face substantial risk of energy shortfalls in the next five to ten years,driven by increasing electricity demand from data centers,decarbonization throug

22、h electrification,and industrial growth.2 While many initiatives to increase generation capacity are in progress,the pace of these efforts may pose challenges to meeting future energy demands.In this report,to better understand power-supply resource gaps,we assess potential increases in supply resou

23、rces and propose a capacity metric that can be directly compared with load increases.We address the following question:What will the United States additional power capacity be by 2030?Approach Different stakeholders have evaluated potential power supply availability and resource gaps in 2030,and the

24、ir assessments have shown forecasted capacity additions that range from 115 to 800 gigawatts(GW).3 However,there is a lack of consistent metrics to measure the capability of supply-side resources to meet the increasing demand,because many of these new resources are variable renewables or energy-limi

25、ted storage.4 For these resources,the nameplate capacity does 1 NERC evaluates the reliability of the U.S.Bulk Electric System(the facilities and control systems necessary for operating an interconnected electric energy transmission network)through annual seasonal and long-term assessments,as well a

26、s a biennial probabilistic assessment,providing an independent examination of the projected resource adequacy metrics and associated risks to the North American bulk power system.2 NERC,2024 Long-Term Reliability Assessment.3 Bistline et al.,“Emissions and Energy Impacts of the Inflation Reduction A

27、ct”;EPRI,Powering Data Centers;International Energy Agency,Renewables 2024.4 Four-hour battery storage is an example of energy-limited storage.Variable renewables are renewable energy sources,such as wind and solar power,whose power output fluctuates over time because of such factors as weather and

28、seasonal changes.vi not accurately reflect the capability to meet the increased demand.5 We propose a new metric that estimates a potentially available capacity that uses an approach similar to how energy system planners evaluate resource adequacy.This approach allows direct comparison between forec

29、asted peak load and effective capacity to determine adequate resources and planning margins.6 Additionally,available capacity incorporates such factors as project completion uncertainties,as our study relies on interconnection queue dataa list of projects seeking to connect to the electricity grid,n

30、ot all of which reach completion.Therefore,the available capacity metric can be directly compared with the future power needs of data centers.This report presents data,findings,and recommendations focused on expectations of U.S.power capacity additions by 2030.Our approach looks at the projected U.S

31、.generation capacity additions from both front-of-the-meter(FTM)and behind-the-meter(BTM)resources.We aggregated capacities from various resource types and transformed them into a capacity metric that can serve increasing data center power demand that is assumed in this report to be inflexible and f

32、irm.Estimates of available capacity are thus interpreted in terms of technology-agnostic capacity that can support additional load with an acceptable reliability standard so that peak-hour demand is met.7 For a summary of capacity-related terminology,see the box on the next page.The results presente

33、d here estimate the United States power capability to meet increasing loads(e.g.,AI data centers)in 2030 given current capacity expansion plans for the contiguous United States.5 Nameplate capacity is the maximum theoretical output of a power-generation facility as specified by the manufacturer.Refe

34、r to the glossary at the end of this report for more information.6 Rendahl et al.,Resource Adequacy for State Utility Regulators.7 Currently,the typical U.S.customer can expect to experience two outages per year for a total of fewer than five hours(Denholm,“Top 10 Things to Know About Power Grid Rel

35、iability”).Reliable energy supply prevents failures and delays during high-performance computing and data center operations.See,for example,Grant et al.,“Metrics for Evaluating Energy Saving Techniques for Resilient HPC Systems.”Electricity providers,data center customers,and other large stakeholder

36、s expressed concerns about resource adequacy and reliability of the electricity grid to the U.S.Department of Energys Secretary of Energy Advisory Board in 2024.Reliable energy for AI training is needed for competitiveness and next-generation model development and is critical for customer-facing inf

37、erence activities.For discussion of AI energy demand and the desire for reliability,see Nland,Hjelmeland,and Korps,“Will Energy-Hungry AI Create a Baseload Power Demand Boom?”and Pilz and Heim,“Compute at Scale.”vii Supplementary Information:Understanding Capacity Terminologies This report uses thre

38、e distinct terms to describe the capacity of energy resources.While all are measured in gigawatts,they represent different aspects of a resources potential and actual contribution to the power grid:Nameplate capacity:Max theoretical output(manufacturer specification).Commonly used when discussing ov

39、erall energy supply potential.Effective capacity:A technology-agnostic reliable capacity.Determined by capacity accreditation factors and used by energy planners for resource adequacy assessment.Available capacity(developed by the authors):Our estimate for future,deliverable,and aggregated power.Est

40、imated from active and anticipated energy projects seeking interconnection to the grid.Builds on effective capacity by adding completion rate,losses,and deliverability.More information can be found in the glossary of this report.Key Findings We estimate that currently planned nameplate resource addi

41、tions will increase to 151 GW of FTM generation capacity and 149 GW of BTM generation capacity.The available capacity,in turn,will increase by approximately 82 GW.FTM additions will add approximately 33 GW of the net available capacity,while BTM additions will increase capacity by 49 GW to reduce gr

42、id peak loads.Net FTM capacity consists of 77 GW of additions minus 44 GW of planned retirements.The majority of the currently planned FTM additional capacity is in the region administered by the Electric Reliability Council of Texas(ERCOT),while the BTM additions are split among regions administere

43、d by ERCOT,regions administered by the Midcontinent Independent System Operator,and regions without an administrative authority.Recommendations on FTM Capacity Prioritize FTM Capacity That Increases the Reliability of the Grid Resource mix matters.The focus should be on increasing net effective capa

44、city,not nameplate capacity.Flexible and firm generation is more valuable than intermittent resources for reliably meeting increasing demand.Independent system operators(ISOs)and energy planners should prioritize interconnecting flexible and firm resources with higher capacity accreditation factors,

45、such as natural gas simple-and combined-cycle turbines,or storage technologies.For example,PJM,an ISO,announced that it would accelerate“shovel-ready”8 8 A shovel-ready project is one that is at an advanced stage of development for construction to begin soon.viii projects and projects with highereff

46、ective load-carrying capability(ELCC)9 resources to achieve expedited operation by 2029.10 Prioritize Solutions with Front-Loading Impact on Expediting the Integration of Additional FTM Capacity to the Grid Timing is crucial.Developing new resources,locating new sites,requesting interconnection,and

47、obtaining permits involve significant uncertainties and require long lead times.Several actions can be taken to expedite capacity integration in the short term;however,timely solutions might not be cost-effective and might not provide society-wide benefits.Implementing these solutions could have imp

48、lications for rate increases or local environmental concerns.First,energy planners and system operators should consider the benefits of extending planned retirements or retrofitting resources to address short-term grid reliability and resource shortages.There is precedent for such actions.For exampl

49、e,the California State Water Resources Control Board decided to extend the operation of once-through cooling natural gas power plants beyond their planned phaseout dates to address grid stability and reliability.11 Additionally,SB 846 authorized the extension of a nuclear power plant(with a nameplat

50、e capacity of 2.3 GW)to operate for up to five years beyond its current expiration date.12 However,planned retirement is typically associated with fossil fuelbased generation.Postponing planned retirements or retrofitting the resources can lead to decreased air quality,worsened health impacts on nea

51、rby communities,and increased emissions of greenhouse gases compared with taking these resources offline.Planned retirements are also associated with old and inefficient technologies,whose unit costs to generate electricity can be higher than those of newer and more-efficient generation assets.Secon

52、d,the Federal Energy Regulatory Commission(FERC),ISOs,and energy planners should initiate a simpler and expedited study process for utilizing existing sites and surplus interconnection rights to increase effective capacity.FERC has already encouraged the use of surplus interconnection service at exi

53、sting points.13 For example,colocating battery storage at an existing solar or wind site can significantly increase effective capacity,while using surplus interconnection capacity can shorten the implementation timeline.There is a need for a separate,expedited process that occurs outside the convent

54、ional interconnection queue for such development.Third,to address the backlog of interconnection requests,ISOs should prioritize projects with higher completion milestones among active queues.For example,the 9 Projects that have an ELCC of at least 45 percent are eligible.These include thermal resou

55、rces,energy storage,offshore wind,and landfill gas,according to Bielak,“Reliability Resource Initiative Straw Proposal.”10 Bielak,“Reliability Resource Initiative Straw Proposal.”11 California State Water Resources Control Board,“Board Adopts Amendment Extending Once-Through-Cooling Operations at Fo

56、ur Coastal Plants.”12 California State Legislature,Diablo Canyon Powerplant.13 Borgatti and Yasutake,ReSISting a Resource Shortfall.ix California ISO proposed prioritizing interconnection studies based on project readiness and competition to manage the backlog from an exceptional volume of requests.

57、14 Recommendations on BTM Capacity ISOs should encourage the buildout of BTM resources.Our forecasts show that all regions will see additions in solar photovoltaics(PV).However,some regions start from very low baselines and have large potential for BTM resources,which are currently not being used.IS

58、Os should focus on the buildout of BTM storage.Data centers are looking for reliable power sources.Unfortunately,the effective capacity of solar PV(measured by ELCC)is relatively low and set to decrease further as the marginal impacts of additional solar PV capacity decrease.The ELCC for battery sto

59、rage also decreases as more battery storage is brought online,but the decrease is significantly slower.15 Battery storage can thus provide more effective capacity than solar PV and a more helpful contribution to lowering peak demand and improving reliability.Recommendations on the Transmission Syste

60、m(Grid)Prioritize Solutions That Improve the Efficiency of Transmission Planning and Monitoring Grid planners need a clearer understanding of the availability and timelines of nuclear power plants,as substantial power output of each plant(ranging from 0.5 to 1.4 GW per plant)can significantly affect

61、 the system.The recommissioning of retired nuclear plants or the decommissioning of active plants can significantly alter capacity in the short term.However,without proper coordination,these changes could lead to adverse impacts,such as stranded assets,or drive up electricity costs by overlapping wi

62、th other infrastructure investments or capacity shortfalls.To avoid inefficiencies,grid planners should work closely with policymakers,utilities,and developers to align resource planning with projected demand.A more coordinated approach will help optimize investments,ensure grid reliability,and prev

63、ent unnecessary financial and operational risks.To maximize the effective capacity that BTM resources can provide,ISOs should improve BTM resource monitoring.This can be done through increased reporting of BTM installations.BTM resources should also be treated as resources by themselves rather than

64、as demand reductions.As BTM resources become more prevalent,ISOs should ensure that BTM resources pay their fair share for their use of the grid and contribute to its upkeep and further development.As BTM systems reduce reliance on the grid,they also decrease the revenue that utilities receive for g

65、rid upkeep,thereby potentially shifting costs to non-BTM customers.This pattern could lead to a“utility death spiral,”wherein remaining 14 Weaver et al.,“Re:California Independent System Operator Corporation Docket No.ER24-_-000.”15 Specht,“To Understand Energy Storage,You Must Understand ELCC.”x gr

66、id customers face increasingly higher costs.Despite challenges with accessing and maintaining the grid,it remains the most effective and efficient way of providing electricity.Prioritize Solutions That Are Flexible to Differences Across U.S.Grids Location matters:Although ERCOT shows the largest cap

67、acity additions across regions,data centers located outside the region administered by ERCOT have limited access to its power supply.At the same time,the concentration of large data centers in a single region can lead to increased power prices,reduced grid reliability,increased grid instability,and

68、increased risk of cascading power outages for an entire region.16 ERCOTs stakeholders should reevaluate whether the benefits of running an energy-only market compensate for the risks given the increase in data center load.17 16 NERC,“Incident Review Considering Simultaneous Voltage-Sensitive Load Re

69、ductions”;U.S.Energy Information Administration,“Data Centers and Cryptocurrency Mining in Texas Drive Strong Power Demand Growth.”17 Unlike other ISOs,ERCOT does not pay generators to have available capacity.Instead,it just pays for the energy they produce.xi Contents About This Report.iiiSummary.v

70、Figures and Tables.xiiChapter 1.Introduction.1Background and Motivation.1Key Assumptions and Limitations.5Chapter 2.Data,Approach,and Results.8Projecting FTM Available Capacities by 2030.9Projecting BTM Available Capacities by 2030.30Aggregate Results.44Chapter 3.Findings and Recommendations.46FTM G

71、eneration Levers for Unlocking Additional Supply on the U.S.Power Grid.46BTM Levers for Achieving Reliable Grid Peak-Load Reductions.49Further Challenges and Concerns.50Next Steps.51Appendix A.Detailed FTM Methodology.52Appendix B.Detailed BTM Methodology.68Abbreviations.71Glossary.72References.75 x

72、ii Figures and Tables Figures Figure 2.1.Illustration of Method to Translate Planned FTM Projects to Data Center Power Need.10Figure 2.2.Planned Energy Project Additions to U.S.Power:Nameplate Capacity by Year(20252030).12Figure 2.3.Expected Additions to U.S.Power:Nameplate Capacity by Year(20252030

73、).16Figure 2.4.Average CAFs Used in Our Analysis.17Figure 2.5.Planned Retirement by Resource Type and Region(20252030).19Figure 2.6.Available Additional Capacity(GW)and Planned Retirements(GW)by Year.20Figure 2.7.Comparison of Planned and Expected Additions to U.S.FTM Power Capacity by 2030.21Figure

74、 2.8.Average Expected Completion Rates Across the United States by Resource Type.24Figure 2.9.Overall BTM Effective Capacity Additions by Region in 2030(GW).37Figure A.1.Regions Served by RTOs and ISOs in the Electric Transmission Grid.52Figure A.2.Typical Process from Interconnection Request to Com

75、pletion,Including Barriers.61Figure A.3.Average Completion Rates and Average CAFs by Resource Type.66Tables Table 2.1.Summary of Methodological Details.8Table 2.2.Nameplate Capacity(MW),Active Projects by ISO Region and Resource Type,and Forecast Additions by Resource Type.14Table 2.3.Average Comple

76、tion Rates by Region and Resource Type(percentage).15Table 2.4.Components of Expected Available Capacity by Resource Type(20252030).20Table 2.5.Available Capacity by Region for Active Projects,Forecast Additions,Retirements,and the Net Final Estimate(20252030).22Table 2.6.Impact of a 1-Percent Incre

77、ase in Completion Rates by Resource Type,Including Targeting High-CAF Resources.25Table 2.7.Summary of Potential Levers to Unlock Additional Capacity.29Table 2.8.NREL Projections of BTM Capacity by Resource Type and Region in 2024 (GW).31Table 2.9.NREL Projections and ISO Data on BTM Growth by Resou

78、rce Type and Region in 2024(GW).31 xiii Table 2.10.BTM Solar PV Nameplate Capacity Growth by Region Out to 2030(GW).33Table 2.11.BTM Storage Nameplate Capacity Growth by Region Out to 2030(GW).33Table 2.12.ELCC by Resource Type and Region.34Table 2.13.BTM Solar PV Effective Capacity Growth by Region

79、 Out to 2030(GW).35Table 2.14.BTM Storage Effective Capacity Growth by Region Out to 2030(GW).35Table 2.15.Overall BTM Effective Capacity Growth by Region Out to 2030(MW).36Table 2.16.ELCC by Resource Type and Region for Solar,Battery Storage,and Solar PVPlus-Storage Hybrid.39Table 2.17.BTM Solar PV

80、Plus-Storage Effective Capacity Growth by Region Out to 2030,Positive Scenario(GW).39Table 2.18.BTM Storage Effective Capacity Growth by Region Out to 2030(GW).40Table 2.19.BTM Solar PV Nameplate Capacity Growth by Region Out to 2030,Negative Scenario(GW).41Table 2.20.BTM Storage Nameplate Capacity

81、Growth by Region Out to 2030,Negative Scenario(GW).42Table 2.21.BTM Solar PV Effective Capacity Growth by Region Out to 2030,Negative Scenario(GW).42Table 2.22.BTM Storage Effective Capacity Growth by Region Out to 2030,Negative Scenario(GW).43Table 2.23.Overall BTM Effective Capacity Growth by Regi

82、on Out to 2030,Negative Scenario(GW).43Table 2.24.FTM and BTM Available Capacity Additions by Region,20252030(GW).45Table A.1.Summary of CAISOs Active Projects.54Table A.2.Summary of PJMs Active Projects.55Table A.3.Percentage of Active Capacity by Interconnection Service Type(Proposed to Be Online

83、Between 2025 and 2030).58Table A.4.Completion Rates by Resource Type(percentage).62Table A.5.Capacity Accreditation Methods and Data Sources Used for This Study.64Table A.6.ELCC Factors Used for This Study,by ISO and Resource Type(percentage).64 xiv 1 Chapter 1.Introduction Anticipated growth in art

84、ificial intelligence(AI)development requires the buildout of additional power capacity at an exceptional scale and speed.There is concern that the necessary grid loads to power future AI data centers for training and inference activities may exceed the capacity of additions to the U.S.power grid.18

85、In this report,we assess potential increases in supply resources in the contiguous United States and propose a capacity metric that can be directly compared with load increases.We address the following question:What will the United States additional power capacity be by 2030?Background and Motivatio

86、n The anticipated load growth from AI is extraordinary in magnitude,speed,and geographic concentration.Developing advanced AI models requires enormous amounts of power,19 and existing facilities are already straining the U.S.electricity grids.20 As the demand for computational resources continues to

87、 rise in the coming years,electricity consumption by AI data centers is expected to grow at an exceptional rate.21 Estimates for data center demand growth in the United States by 2030 vary widely,from an additional 34 gigawatts(GW)to as much as 253 GW.This variability stems from significant uncertai

88、nties,including the pace of the growth in demand,potential efficiency improvements through innovation,and the unclear long-term expansion plans of data center operators.Differences in reporting units(e.g.,terawatt-hours TWh versus GW),scope(AI-specific versus general data center demand),and projecti

89、on timelines further contribute to the range of estimates:RAND projects that global AI data center power demand could reach 327 GW by 2030.22 For the United States alone,additional demand is estimated to be between 158 GW and 253 GW,assuming a retention rate of 50 percent to 80 percent.Lawrence Berk

90、eley National Laboratory(LBNL)estimates that total U.S.data center power demandincluding both AI-related and general usewill range between 325 18 Bistline et al.,“Emissions and Energy Impacts of the Inflation Reduction Act”;North American Electric Reliability Corporation(NERC),2024 Long-Term Reliabi

91、lity Assessment;Pilz,Mahmood,and Heim,AIs Power Requirements Under Exponential Growth.19 Srivathsan et al.,“AI Power”;Scharre,Future-Proofing Frontier AI Regulation.20 Secretary of Energy Advisory Board,“Recommendations on Powering Artificial Intelligence and Data Center Infrastructure.”21 Sevilla e

92、t al.,“Can AI Scaling Continue Through 2030?”22 Pilz,Mahmood,and Heim,AIs Power Requirements Under Exponential Growth.2 TWh and 580 TWh by 2028.23 This estimate corresponds to 74132 GW,representing an increase of 3492 GW over current levels.24 Goldman Sachs forecasts an incremental 47 GW of data cen

93、ter demand,including both AI and non-AI uses,by 2030.25 Beyond the challenge of aggregated scale,a single site expected to demand a large amount of power also poses issues related to building necessary infrastructure in a timely manner and maintaining grid stability.26 An individual facility may con

94、sume up to 5 GW by 2030,27 a substantial amount of power comparable to the 2023 net summer generation capacity of a single state,such as Idaho(5.4 GW),Maine(5.3 GW),or New Hampshire(4.5 GW).28 Historically,similar levels of capacity additions occurred in the mid-1970s and early 2000s,with increases

95、of 130160 GW over a five-year period.29 However,the current substantial load growth expectations driven by new digital industries,advanced manufacturing,and supply chain reshoring,combined with heightened challenges in constructing new infrastructure,present significant challenges to todays grid.The

96、se challenges stem from increased costs,stricter regulatory requirements,timing constraints,and the geographic concentration of large loads.30 Connecting to the grid remains the gold standard for the operation of data centers.Data centers value energy redundancy,31 and their preference for grid conn

97、ections is motivated by the need for improved reliability,resilience,sustainability,and cost-effectiveness.These benefits are not available from off-grid projects.Although it is technically possible to use only off-grid electricity in some cases,32 we did not find any indication of a substantial rol

98、lout of fully off-grid data center projects.Data centers may run off-grid for a limited time,which is referred to as using bridge power,as an interim solution to long interconnection queues that delay new generation capacity from being connected to the grid.AI labs still value power reliability and

99、23 Shehabi et al.,2024 United States Data Center Energy Usage Report.24 LNBL assumed an average capacity utilization rate of 50 percent to translate energy(TWh)to capacity(GW).25 Davenport et al.,“Generational Growth.”26 Energy Systems Integration Group,“Grid Planning and Operation of Large Loads”;N

100、ERC,“Incident Review Considering Simultaneous Voltage-Sensitive Load Reductions.”27 Sevilla et al.,“Can AI Scaling Continue Through 2030?”28 EIA,“Electricity.”The EIA reports net summer generation capacity,which is typically lower than nameplate capacitythe theoretical maximum output set by the manu

101、facturer.Net summer generation capacity is determined by performance tests during peak demand,from June 1 to September 30.29 EIA,“February 2025 Monthly Energy Review.”30 Riu et al.,Load Growth Is Here to Stay,but Are Data Centers?31 Lovett,“A Deep Dive into Data Center Redundancy.”32 For example,dat

102、a center company ECL is due to deliver an off-grid,hydrogen-powered modular data center in summer 2025(Vincent,“ECL Debuts 1 GW Off-Grid Hydrogen-Powered AI Factory Data Center on 600 Acres Near Houston”).3 redundancy for AI training to avoid model development delays.33 AI development,and its positi

103、ve impact on U.S.competitiveness,faces significant challenges from the grid connection bottleneck.Stakeholders have identified major challenges facing the current grid,especially concerning the integration of large volumes of energy resources.New power-generation development marks a three-to five-fo

104、ld increase over that for the previous decade.This equates to more than 1,570 GW of generation capacity and approximately 1,030 GW of storage capacity queued for grid interconnection as of the end of 2023.34 In addition,planned power projects face multiple regulatory approvals,permitting challenges,

105、and large network upgrade costs because of insufficient existing grid infrastructure.These obstacles increase project costs and lead to significant project delays that can result in incomplete or withdrawn projects.The Federal Energy Regulatory Commission(FERC)has stated that“interconnection queue b

106、acklogs and study delays afflicting generator interconnection service nationwide hinder the timely development of new generation.”35 Although many initiatives to increase generation capacity are in progress,the pace of these efforts may pose challenges for both decarbonization and AI needs.Timely so

107、lutions will support both ongoing decarbonization efforts and AI demands.Electricity grid operators face multiple objectives.Operators must balance interconnection backlogs,grid reliability,and the renewable energy transition with the energy needs from AI data centers and industrial,commercial,and r

108、esidential users.Within this complex environment,proposed generation projects will have marginal impacts on the electricity grid provided that the projects are completed and operational within the time frame to power future AI development.Upgrades to transmission infrastructure may have the largest

109、single impact on addressing the interconnection backlog and bringing more generation online.There is a wide range of reported future capacity additions,and there are high degrees of uncertainty and inconsistency in these estimates of future power capacity in the United States.We found that increment

110、s to capacity forecasts for 2030 in the United States vary widely,from 115 GW to 800 GW.These estimates are derived using various methods,including optimization modeling,survey data,and a hybrid approach that combines physical queue data with trend extrapolation.For instance,33 Secretary of Energy A

111、dvisory Board,“Recommendations on Powering Artificial Intelligence and Data Center Infrastructure.”AI labs prefer redundancy that extends beyond their own power solutions,such as on-site generators and backup generators.Furthermore,the grid offers its own redundancy through reserve margins and other

112、 mechanisms.34 Rand et al.,“Queued Up.”35 FERC,“Order No.2023.”4 Bistline and colleagues 2023 optimization model results in capacity increases of around 115 GW to 600 GW,mainly to meet policy goals,such as decarbonization.36 EPRI accounted for data center load growth,estimating 300 GW in additions b

113、y 2030 in a scenario with no data center load growth and up to 800 GW in additions by 2030 in a scenario with high data center load growth(18-percent year-on-year increase in data center loads).37 LBNL reports planned generation projects in interconnection queues across the contiguous United States

114、representing more than 2,600 GW of additional capacity.38 the U.S.Energy Information Administration(EIA)reports 141 GW of planned capacity additions in 20252030 and retirements of 56 GW of nameplate capacity.The substantial decrease in forecast additions is predominantly due to interconnection backl

115、ogs and delays.However,these literature estimates usually reflect the supply needs identified from optimization modeling to meet load growth and policy goals,without considering the physical projects that can be developed and interconnected within the 2030 timeline.As discussed earlier,there are var

116、ious barriers to the timely completion and grid connection of planned projects.An additional challenge lies in translating various nameplate capacity estimates in a way that allows comparison with data center power needs.Most new resources in development are intermittent renewables or energy-limited

117、 battery storage,which have limited ability to reliably meet increasing demand.This capability depends on the resource type and geographical location.There is a need for a technology-agnostic and reliability-neutral capacity metric that allows direct megawatt(MW)-to-MW comparison with power loads.As

118、 mentioned above,in this report,we assess potential increases in supply resources and propose a capacity metric that can be directly compared with load increases.We address the following question:What will the U.S additional power capacity be by 2030?Our approach looks at the projected U.S.generatio

119、n capacity additions from both front-of-the-meter(FTM)and behind-the-meter(BTM)resources.We aggregate capacities from various resources and transform them into a capacity metric that can inform the reliability and adequacy requirements to meet increasing data center loads.39 This aggregation is the

120、technology-agnostic available capacity to provide a firm load capable of meeting peak hour demand.40 We use this 36 Bistline et al.,“Emissions and Energy Impacts of the Inflation Reduction Act.”37 EPRI,Powering Data Centers.38 Rand et al.,“Queued Up.”39 Currently,a typical U.S.customer can expect to

121、 experience two outages per year for a total of fewer than five hours(Denholm,“Top 10 Things to Know About Power Grid Reliability”).40 Reliable energy supply prevents failures and delays during high-performance computing and data center operations.See,for example,Grant et al.,“Metrics for Evaluating

122、 Energy Saving Techniques for Resilient HPC Systems.”Electricity providers,data center customers,and other large stakeholders expressed concerns about resource adequacy and reliability of the electricity grid to the U.S.Department of Energys(DOEs)Secretary of Energy Advisory Board in 2024.Reliable e

123、nergy for AI training is needed for competitiveness and next-generation 5 metric to estimate the ability of the current capacity expansion plan to meet the projected 2030 loads in the contiguous United States.We estimate the available capacity under status quo conditions.Our scope for this report do

124、es not cover the limits to rapidly scaling energy generation with optimal federal government policies or the availability of additional resources for individual data centers.The U.S.capabilities and barriers to rapidly scaling AI will be discussed in a later report,and the potential for additional g

125、eneration(renewable,fossil fuel,or nuclear)connected to single AI data centers will be explored in another later report that addresses the capability to increase power supply to individual U.S.AI data centers.Key Assumptions and Limitations In this study,we assume that data center load profiles are

126、firm and inflexible,though future innovations and use cases can affect this assumption.41 The specific data centers power usage profiles are unknown,since access to private-sector planning is limited.42 Data centers can accommodate diverse use cases,leading to varying power consumption patterns and

127、sudden spikes.The flexibility of power consumption depends on the specific services and tasks.A single data center can also alter its use cases at any time based on operational decisions.AI training,which typically occurs in large and centralized data centers,requires significant amounts of electric

128、ity.43 The magnitude also depends on the computational processes(audio,video,and image computation processes can be more demanding than text processing).While individual instances of AI inference require only a small amount of energy,44 these processes still contribute significantly to overall energ

129、y use because of continuous and frequent operations.45 Additionally,data center cooling needs are based on local weather conditions,which can lead to seasonal,daily,and hourly load fluctuations.Potential future advances in energy efficiency and flexibility for training,inference,and data center cool

130、ing remain uncertain.46 Therefore,we assume that 100 percent of the load can occur at any time.model development and is critical for customer facing-inference activities.For discussion of AI energy demand and the desire for reliability,see Nland,Hjelmeland,and Korps,“Will Energy-Hungry AI Create a B

131、aseload Power Demand Boom?”and Pilz and Heim,“Compute at Scale.”41 We assume that the load can happen at any given time with the full capacity,which can also be referred to as firm load.42 Secretary of Energy Advisory Board,“Recommendations on Powering Artificial Intelligence and Data Center Infrast

132、ructure.”43 Aschenbrenner,“Racing to the Trillion-Dollar Cluster”;Mehta,“How Much Energy Do LLMs Consume?”44 This is true only in the relative sense:A single request to ChatGPT is still ten times more power intensive than a standard Google Search query(Wei,“ChatGPTs Power Consumption”).45 Luccioni,J

133、ernite,and Strubell,“Power Hungry Processing.”46 Secretary of Energy Advisory Board,“Recommendations on Powering Artificial Intelligence and Data Center Infrastructure”;Zhu et al.,“Scalable MatMul-Free Language Modeling.”6 In estimating future available capacity,we rely on available static capacity

134、accreditation factor(CAF)/effective load-carrying capability(ELCC)values.Grid planners routinely update these values to account for evolving demand,load profiles,and resource mixes.47 However,the CAF/ELCC studies used in our analysis do not include expected AI-related load growth.The large-scale dep

135、loyment of these loads could significantly alter both the load shape and its magnitude;therefore,the CAF of a resource type may increase or decrease depending on its ability to serve the load.In particular,the CAF/ELCC value for four-hour storage can diminish if high-demand periods become longer,a s

136、cenario that requires extensive modeling to determine the conditions under which it would occur.A storage unit does not generate energy on its own;it relies on external generation sources to charge and discharge as needed.A recent study suggests that the existing headroom in the U.S.power system fro

137、m planning reserves is sufficient to accommodate significant constant new loads without new generation.48 The study reveals that across regions,curtailment durations ranging from 1.7 hours(at a 0.25-percent curtailment limit)to 4.5 hours(at a 5.0-percent curtailment limit)without new generation can

138、support an increasing load up to 215 GW.This indicates that a four-hour duration battery could meet system needs with a firm load increase exceeding 215 GW when new generation is added.Consequently,we believe that it is reasonable to assume that by 2030,four-hour duration storage will have sufficien

139、t energy to charge and be adequate to meet increasing demand,and the CAF/ELCC for storage would not decrease significantly from values we used in our analysis.When estimating additional available capacity,we do not consider competing loads,such as increased demand from electrification.Therefore,our

140、results can be seen as an optimistic maximum capacity that could be available to AI loads.However,since new generation capacity may have dedicated off-takers or may be developed for other purposessuch as meeting reliability needs from increasing load or decarbonization goalsnot all capacity can be a

141、llocated for AI purposes.Because of uncertainties in the magnitude and timing of growing loads,these factors are excluded from our analysis.While we aggregate the capacity for the contiguous United States,a concentration of data centers in a certain region will result in a power deficiency.While the

142、 grid is connected to some extent,it is fragmented because of limited interregional transmission capacity.National aggregates should be used to measure the United States ability to meet aggregated AI power demands and do not indicate the net power available for a single location.This is because the

143、contiguous United States is not served by a homogeneous electricity grid.The three wide areas of synchronous U.S.power grids are the Eastern Interconnection,the Western Interconnection,and the Texas Interconnection.While electricity can be transmitted over long distances,the actual flow of electrici

144、ty is determined by demand and supply dynamics,as well as by the 47 Stenclik,Ensuring Efficient Reliability.48 Norris et al.,Rethinking Load Growth.7 physical limitations of the grid.We will investigate the maximum power available to a single data center for AI training and inference activities in a

145、 subsequent report.Available capacity is estimated with information available today.However,the resource mix and load profile can change the scale,as the contributions of the individual resources depend on interactions with other resource types,load profiles,and so on.In this report,we focus on asse

146、ssing available resource capacity,but having sufficient resources is only part of the challenge.A robust transmission and distribution(T&D)infrastructure is needed to reliably transmit the power to where it is demanded.Even with sufficient generation capacities,there will be utilization limitations.

147、In addition,energizing new large loads,such as AI data centers,brings challenges for power system stakeholders.These include building infrastructure at the needed pace in the needed place,addressing grid stability risks,and maintaining reliability standards.Our study does not account for infrastruct

148、ure needs or regulatory or process challenges for energization.For BTM resources,we base our estimates on aggregate projections rather than public project announcements.This is because,unlike FTM projects,most BTM projects are small and are not publicly announced.Data are therefore scarce,and we do

149、not have a view of projects in the pipeline.Instead,we must rely on projections made by the independent system operators(ISOs)and other relevant organizations.We assume that BTM resources will continue to be built out and that retirements will be limited.BTM resources are not yet widespread in the U

150、nited States,and there is still a lot of potential for consumers to install BTM resources.As solar and energy storage units age,they are likely to be replaced by similar or even more-efficient units.Nevertheless,rising material prices and supply chain issues may lead to the retirement rather than up

151、graded replacement of BTM solar and energy storage units.8 Chapter 2.Data,Approach,and Results This report provides an overview of our forecasts of additional capacity by quantity and technology type that will be connected to the U.S.electricity grid by 2030.We employ different datasets and approach

152、es to estimate the available capacity for FTM and BTM systems.Table 2.1 provides a summary of this reports methodology(e.g.,our datasets,approach,and key assumptions).Table 2.1.Summary of Methodological Details Component of Methodology FTM Capacity Additions BTM Capacity Additions High-level assumpt

153、ions made Recent trends in resource development activities continue.There are no major technological innovations or changes in the resource development and integration process.There are no major changes in the T&D network capacity.Recent trends in resource development activities continue.There are n

154、o major technological innovations or changes in the resource development and integration process.We used the same CAFs used for FTM.a Available capacity Final metric that accounts for the effect of completion rates,intermittent generation,and T&D losses on nameplate planned generation Final metric t

155、hat accounts for the effect of completion rates,intermittent generation,and T&D losses on nameplate planned generation Out of scope Maximum capacity that can be supplied to a single AI data center Maximum capacity that can be supplied to a single AI data center Data ISO interconnection queuesb EIA E

156、lectric Generator Inventoryc National Renewable Energy Laboratory(NREL)2023 Standard Scenarios,d complemented with 2024 ISO data on BTM resourcese Completion rates Estimated by region and resource based on historical data for interconnection queues and completion rates reported by ISOs Average compl

157、etion rates(of nameplate capacity):Fossil fuels:27 percent Renewables:13 percent Storage:13 percent Contiguous United States:14 percent Assumption that projected capacity represents realized capacity,without the need to take additional completion rates into account Resource types and assessing effec

158、tive capacity More than 90 percent of proposed projects are renewable,energy storage,or both.We assumed that this trend will continue.We used regional and resource-specific factors to evaluate the contribution of individual resource types.The capacity of variable renewables is intermittent and has a

159、 smaller BTM capacity is based on solar photovoltaics(PV)with battery storage,which has a smaller marginal contribution than planned storage or gas generation.We account for the difference between nameplate and effective capacity with capacity accreditation methods,such as ELCC.9 Component of Method

160、ology FTM Capacity Additions BTM Capacity Additions effective capacity than comparable storage or fossil fuel capacity.We account for the difference between nameplate and effective capacity with capacity accreditation methods,such as ELCC.T&D losses 5 percent 0 percentf Retirements Planned retiremen

161、ts converted to available capacity contribution,assumed to proceed on schedule Assumption of no retirements of BTM resources Uncertainties in estimates Completion rates and timing for additions Timing for retirements Estimates based on projections rather than actual project pipeline a BTM storage re

162、sources are not dispatchable by grid operators;therefore,their effective capacity can be less than the FTM.b The interconnection queue is a list of pending requests to connect new power-generation facilities,such as solar or wind farms,to the electricity grid.We used transmission-level interconnecti

163、on queue data from seven ISOs as of October 2024.More information is provided in Appendix A.c EIA,“august_generator2024.xlsx.”d Gagnon et al.,2023 Standard Scenarios Report.e ERCOT,2024 ERCOT System Planning;ISO New England,“Explainer”;McPherson,“NYCA Renewables 2023”;MISO Distributed Energy Resourc

164、es Task Force,“2024 OMS DER Survey Results”;Nyberg,“California Energy Storage System Survey”;PJM,“Distributed Solar Generation Forecast by Zone Cumulative Nameplate Capacity Includes Historical Degraded Values and HIS Forecast”;Vu,“DER Forecast Improvements”;Wilson,“Historical BTM PV and Storage Ado

165、ption Trends in California.”f BTM resources can directly reduce power consumption without T&D assets.Projecting FTM Available Capacities by 2030 Significant FTM projects with varying delivery timelines are proposed and under development across all regions.By 2030,the actual completion of these proje

166、cts and the retirement of existing resources will affect the available capacity.We consider the expected completion rate for projects being proposed and under development,then we further adjust nameplate capacity by resource type and geographical region.Comparing nameplate capacity with data center

167、power needs is a challenge when most new projects are variable renewable energy(VRE)or energy-limited storage resources.The resource capacity cannot be directly compared on an MW-to-MW basis with data center requirements.For instance,a 1-MW solar power plant cannot fully meet the energy demands of a

168、 1-MW data center because of intermittency.To achieve this comparison,we adopt a capacity accreditation method widely used by energy planners and grid operators to measure each resources contribution to meeting future increases in power demands.49 This method provides a technology-agnostic means of

169、49 Robertson,Palmer,and Aagaard,Reforming Resource Adequacy Practices and Ensuring Reliability in the Clean Energy Transition.ISOs,RTOs,and balancing authorities use this method to plan adequate supply resources to meet electricity demand.10 comparing resource contributions across different resource

170、s by region.The CAFs,50 often shown as a percentage of a resources effective capacity compared with its nameplate capacity,are used as the basis for capacity market offers or selection in competitive procurement processes.51 Further details and methodological differences by region are described in A

171、ppendix A.Finally,the term available capacity accounts for the completion of the project and the effective capacity,inclusive of T&D losses and constraints.This process is illustrated in Figure 2.1 and detailed in Appendix A.Figure 2.1.Illustration of Method to Translate Planned FTM Projects to Data

172、 Center Power Need NOTE:Values in gigawatts might not sum precisely because of rounding.a Our analysis includes only projects with interconnection types that ensure a reliable power supply to data centers,overcoming transmission line constraints,available grid capacity,and potential congestion.Furth

173、er details are available in Appendix A.FTM Results The term planned refers to projects that are either being developed or forecast to be added to the pipeline.To estimate developing projects,we primarily rely on interconnection queues with active interconnection requests from ISOs or those reported

174、to the EIA through the completion of 50 CAF is often called ELCC for measuring the resource adequacy contribution of intermittent or energy-limited resources.51 Stenclik,Ensuring Efficient Reliability.11 Form EIA-861 and its monthly supplement,EIA-861M.We apply completion rates to estimate the reali

175、zation of these projects,since only a small percentage get built and generate power.52 We then estimate the effective capacity contribution from each resource type by region to obtain technology-agnostic available capacity estimates.53 Finally,net available capacity is calculated by subtracting plan

176、ned retirements from the available capacity.For the FTM analysis,we assume no significant changes in the next five years in the technologies,capacity sizes,and technology portfolio from what is currently proposed for interconnection with the U.S.electricity grids.Additionally,we assume no major expa

177、nsion of the transmission network in and across regional boundaries.We did not explicitly account for project delays;however,our estimation of completion rates reflects historical performance and inherently includes some delay considerations.While future retirement extensions are likely due to relia

178、bility concerns from increasing load,our baseline estimate does not include them.54 The step-by-step estimation and results are summarized below,while detailed information on the data used and the methodology for this estimation are included in Appendix A.Step 1:Estimate Planned Energy Resources and

179、 Consider Deliverability Our analysis identified 1,086 GW of nameplate capacity planned to interconnect to the grid,including 71 GW of active projects that lack reliable grid access.We estimate that 1,015 GW of nameplate capacity is planned for operation by 2030.This total includes 653 GW of active

180、projects and 362 GW of forecasted future requests.Our focus is on projects with reliable interconnection rights,as most ISOs offer varying types of interconnection rights,with some providing more-reliable grid access than others.55 Our approach mirrors how the California ISO(CAISO)converts qualifyin

181、g capacity(QC)to net qualifying capacity when assessing resource adequacy contributions.The interconnection type ensures that reliable use of the existing grid network gets full credit,while other service types can use the network only when it is available,resulting in no credit or partial credit.Fu

182、rther details on the method and our data source are discussed in Appendix A.52 LBNL found that by 2023,14 percent of capacity and 19 percent of projects proposed between 2000 and 2018 were successfully built and connected(Rand et al.,“Queued Up”).53 Battery storage relies on external generation sour

183、ces for charging before it can discharge when needed.The current system has sufficient surplus generation,and charging is accounted for as a load in the effective capacity assessment.Further discussion and assumptions used in our analysis are included under Key Assumptions and Limitations.54 In 2023

184、,the California Public Utilities Commission(CPUC)extended Diablo Canyon Power Plants operation to 2030,beyond its planned retirement in 20242025,because of reliability concerns.55 Full-deliverability service types,such as the Network Resource Interconnection Service(NRIS)or Full Capacity Deliverabil

185、ity Status(FCDS),ensure reliable use of the existing grid network.In contrast,the other service types,such as Energy Resource Interconnection Service(ERIS)or Energy Only Deliverability Status(EODS),can use the grid only when it is available.12 Of the planned 1,015 GW capacity,49 percent of the reque

186、sted capacity is VRE,and an additional 44 percent is from four-hour battery storage.Figure 2.2 summarizes our analysis of new projects planned by the expected year that they will come online,along with the cumulative nameplate capacity for each year.Note that the term active projects refers to activ

187、e interconnection requests submitted by October 2024,while forecasted additions pertains to requests that will be made after this date.We forecast likely future additions by assuming that the volume of interconnection requests for the coming years will be consistent with that from the past 12 months

188、.Figure 2.2.Planned Energy Project Additions to U.S.Power:Nameplate Capacity by Year(20252030)SOURCES:Features information from CAISO,homepage;PJM,homepage;the Electric Reliability Council of Texas(ERCOT),homepage;the Midcontinent ISO(MISO),homepage;the New England ISO(NE-ISO),(ISO New England,homep

189、age);the New York ISO(NY-ISO),homepage;the Southwest Power Pool(SPP),homepage;and the EIA Electric Generator Inventory(EIA,“august_generator2024.xlsx”).NOTE:Dark navy represents active projects as of October 2024,based on interconnection queues and EIA data,while light blue indicates forecast future

190、 development.There are 653 GW of nameplate capacity in the combined ISO interconnection queues and the EIAs Electric Generator Inventory that have a planned commercialization date between 2025 and 2030(Table 2.2).Of these active projects,the majority will occur in 2025,2026,and 2027.This is because

191、many of the projects are requesting interconnection planning for operations in the next three years.To account for projects that may be added to the queue and completed prior to 2030,we use the 143 GW proposed in the last 12 months(i.e.,the interconnection requests submitted between October 1,2023,a

192、nd September 31,2024)and apply them to 13 forecast potential interconnection requests if the same interconnection requests occurred in each year from 2025 to 2030.With this approach,approximately 1,015 GW of incremental nameplate capacity will be planned with a scheduled completion by 2030.Because a

193、n average project takes five years from interconnection request to commercial operations,many of these projects may operate at a reduced capacity,suffer significant delays,or never be completed.In regions with larger interconnection backlogs,such as CAISO,delays can take more than seven years.14 Tab

194、le 2.2.Nameplate Capacity(MW),Active Projects by ISO Region and Resource Type,and Forecast Additions by Resource Type Resource Type CAISO PJM ERCOT MISO NY-ISO NE-ISO SPP Non-ISO Regionsa Total Forecast Additions Cumulative Coal 11 11 11 Gas 6,193 16,797 407 1,018 1,766 5,216 31,397 32,428 63,825 Ot

195、her 200 50 250 250 Nuclear Biomass 10 76 86 86 Solar 25 30,957 128,496 12,094 9,617 2,396 26,618 16,474 226,677 87,637 314,314 Wind 1,864 15,400 24,362 3,055 9,729 16,943 10,908 6,924 89,185 24,879 114,064 Geothermal 53 27 80 80 Hydro 23 2,075 2,098 2,098 Hybrid 21,985 12,615 2,971 3,217 944 10,988

196、52,720 15,778 68,499 Storage 15,497 30,309 138,486 1,240 15,158 27,601 15,872 6,267 250,431 201,002 451,433 Total 39,424 95,518 308,341 19,768 37,721 48,902 66,153 37,109 652,934 361,725 1,014,659 SOURCES:Features information from CAISO,homepage;EIA,“august_generator2024.xlsx”(as of November 2024);E

197、RCOT,homepage;ISO New England,homepage;MISO,homepage;NY-ISO,homepage;PJM,homepage;SPP,homepage.Forecast additions indicates forecasted future development.NOTE:Numbers in bold are the sums of all rows.a Nameplate capacity for non-ISO regions represents the sum of nameplate capacity for proposed proje

198、cts outside ISO jurisdiction as reported in the EIA Electric Generator Inventory.Non-ISO regions are the Northwest(Washington;Oregon;Idaho;and parts of Montana,Wyoming,Nevada,and Northern California),Southwest(Arizona,Utah,Colorado,and the majority of New Mexico),and Southeast(Florida;Georgia;Alabam

199、a;Tennessee;South Carolina;the majority of North Carolina;and parts of Mississippi,Missouri,and Kentucky).Refer to Figure A.1 in Appendix A.15 Step 2:Estimate Project Completion Many planned FTM generation projects will not deliver the necessary capacity at the scale and speed required to meet proje

200、cted future AI energy development needs.Completion rates by resource type and region are estimated to determine the expected nameplate capacity that will be connected to the grid by 2030.Full details of the methodology used for completion rates are in Appendix A.Completion Rates Methodology Completi

201、on rates by region and resource type are provided in Table 2.3.Empty cells indicate no proposed generators of a given resource type and region.The table reveals that most interconnection applications are ultimately withdrawn,with a completion rate of less than 30 percent based on capacity.Notably,va

202、riable renewable-resource projects have low completion rates,and energy storage projects show particularly low completion rates.56 Among ISOs,ERCOT stands out with the highest project completion rate.57 Table 2.3.Average Completion Rates by Region and Resource Type(percentage)Resource Type CAISO PJM

203、 ERCOT MISO NY-ISO NE-ISO SPP Non-ISO Regions Coal 6 Gas 19 31 30 26 17 26 Other 24 24 Nuclear Biomass 13 13 Solar 10 8 17 28 3 11 1 11 Wind 6 2 28 19 4 14 17 14 Geothermal 0 0 Hydro 30 30 Hybrid 15 9 28 3 11 2 11 Storage 3 0a 22 6 0a 5 0a 5 56 The rapid growth of these resources in the queues likel

204、y contributes to interconnection delays and high upgrade costs,leading to a higher number of projects being withdrawn.Battery storage,as a newer technology,entered the queue during an already backlogged period,which is likely one of the key factors contributing to its lowest completion rate among al

205、l other resource types.57 ERCOT employs a“connect and manage”interconnection process,which bypasses system impact studies prior to resource interconnection.This approach is known to be faster and more efficient than other ISOs approaches.16 Resource Type CAISO PJM ERCOT MISO NY-ISO NE-ISO SPP Non-IS

206、O Regions a 0 percent is from rounding;the actual values are 0.2 percent(PJM),0.2 percent(NY-ISO),and 0.1 percent(SPP).Analysis of historical interconnection queue data indicates that PJM completed 31 storage projects between 2014 and 2023,totaling 25 MW,and withdrew 437 projects of 15 GW in the sam

207、e period.Similarly,NY-ISO completed five storage projects(100 MW)and withdrew 411(41 GW),while SPP completed a project(20 MW)and withdrew 134(17 GW)in the same period.Low completion rates assume that,similar to the historical record,the vast majority of storage projects requesting interconnection in

208、 these regions will be withdrawn.Figure 2.3 shows annual expected capacity completions,differentiating between active projects and forecasted additions.Because of low completion rates,total capacity drops from 1,015 GW to 151 GWa reduction by nearly a factor of 7.Figure 2.3.Expected Additions to U.S

209、.Power:Nameplate Capacity by Year(20252030)SOURCES:Features information from CAISO,homepage;EIA,“august_generator2024.xlsx”(as of November 2024);ERCOT,homepage;ISO New England,homepage;MISO,homepage;NY-ISO,homepage;PJM,homepage;SPP,homepage.NOTE:Dark navy represents active projects as of October 202

210、4,based on interconnection queues and EIA data,while light blue indicates forecast future development.Step 3:Account for Resource-and Region-Specific Performance Consideration of resource type and region-specific attributes is crucial because fuel availability,generation variability,and marginal con

211、tributions can differ greatly by location and existing energy mix.To account for this,we use capacity accreditation or ELCC factors that 17 assess the probabilistic contributions of these resources taken from different ISOs.58 For planned generation resources in non-ISO regions,we use the average EL

212、CC factors across all ISOs for that resource type.ELCC calculations rely on complex“loss-of-load-probability”models that simulate decadal variation in load and resource conditions.During planning,grid planners aggregate the effective capacity of entire portfolios to evaluate system resource adequacy

213、.This accreditation method is evolving to address regional needs because the more diverse resource mix leads to significant variations in processes,metrics,and approaches across regions or planning entities.This approach helps planners understand resource adequacy without conducting complex analyses

214、 for future load increases.In addition,some regions assess marginal contribution instead of average contribution.For renewables,the marginal contribution of the new capacity decreases as more of a resource is added to the grid.For example,increasing solar capacity shifts the net load peak to evening

215、 hours,when solar generation decreases as the sun sets.In a grid already saturated with solar capacity,additional solar capacity may provide minimal additional value;this is reflected in a low marginal ELCC.Figure 2.4 illustrates average CAFs used in our study.Appendix A includes detailed factors by

216、 region and resource type,and Table A.6 provides the final estimates used in our analysis.Figure 2.4.Average CAFs Used in Our Analysis SOURCES:CAF estimates for ISOs are based on different time frames:2024 data(ERCOT,homepage;MISO,homepage;NY-ISO,homepage;SPP,homepage),20262027 projections(PJM,homep

217、age),and 2028 projections 58 Stenclik,Ensuring Efficient Reliability.0%10%20%30%40%50%60%70%80%90%100%CoalGasBiomassSolarWindHydroStorageHybridNuclearCapacityaccreditationfactors(%)Resourcetype 18(CAISO,homepage;ISO New England,homepage).Marginal values were prioritized when available.Details can be

218、 found in Table A.5 of Appendix A.NOTE:Use of average ELCCs may lead to capacity overestimation.For non-ISO-regions,ELCC estimates are calculated as an equally weighted average of ELCC factors from ISO regions.From the expected 151 GW of nameplate capacity additions,regional and resource variation i

219、n capacity accreditation and transmission losses reduce this capacity by almost 50 percent(75 GW).The largest decreases in capacity are the available contribution of single resource-renewable projects,with 44 GW of planned solar and wind contributing approximately 7 GW of additional capacity.The rem

220、aining 77 GW capacity increase is available for powering AI compute or other uses within the current reliability standard.Our approach is detailed in Appendix A.We assume no technological improvements in generation efficiency.Step 4:Analyze Net Additions by Considering Planned Retirements The estima

221、te of future available capacity is reduced by planned retirements of existing power plants.Approximately 55.6 GW of nameplate capacity currently connected to the grid are scheduled for retirement between 2025 and 2030.Figure 2.5 provides a breakdown by resource type and region,showing that 70 percen

222、t of this capacity comes from coal-fired plants,while 29 percent is from oil-and gas-fired plants.Most scheduled retirements are in the MISO region.59 Additionally,we analyze nuclear power plant license expirations and find no indication of short-term decommissioning that can affect the available ca

223、pacity.Detailed data and findings are provided in Appendix A.59 The most notable renewable project with a planned retirement by 2030 is the Old Gold wind farm in northern Iowa,where 80 MW will be taken out of the grid prior to a redevelopment to hybrid generation with 240 MW of wind and 80 MW of sto

224、rage.19 Figure 2.5.Planned Retirement by Resource Type and Region(20252030)SOURCE:Features information from EIA,“august_generator2024.xlsx.”NOTE:Others include solar,wind,hydro,biomass,and storage resources.Resource-and region-specific performance factors,as in Step 3,are also applied to retiring re

225、sources,reducing the 55.6 GW of retiring capacity to 44 GW of available capacity.These reductions are smaller than for planned additions because fossil fuel generators have a higher capacity value than the intermittent renewable resources.Historically,the fossil fuel resources have received almost a

226、 full capacity credit for their nameplate capacity or have been adjusted for forced outage rates only.60 Given these possible retirements,a net addition of 33 GW of net available capacity can be achieved with the retirements,and 77 GW of net available capacity could be available if all retirements a

227、re postponed until after 2030(Figure 2.6 and Table 2.4).Such a widespread delay in fossil fuel retirement may provide additional power for AI development while additional infrastructure and transmission capacity for new generation are built.In Chapter 3,we briefly discuss potential implications of d

228、elayed retirements because of their economic inefficiency or because they are not compliant with Environmental Protection Agency(EPA)regulations.60 Stenclik,Ensuring Efficient Reliability.-7113-31442250022-2468101214161820CAISOPJMERCOTMISONY-ISONE-ISOSPPNon-ISONameplatecapacity(GW)CoalOilandGasOther

229、s 20 Figure 2.6.Available Additional Capacity(GW)and Planned Retirements(GW)by Year SOURCES:Initial inputs for active projects and forecast additions are interconnection requests as reported in October 2024 from CAISO,homepage;ERCOT,homepage;ISO New England,homepage;MISO,homepage;NY-ISO,homepage;SPP

230、,homepage.Available capacity for generators requesting interconnection in non-ISO regions is as reported in EIA,“august_generator2024.xlsx.”Retirements estimated from resources with planned retirement dates in 20252030 are as reported in EIA,“august_generator2024.xlsx.”Table 2.4.Components of Expect

231、ed Available Capacity by Resource Type(20252030)Resource Type Total Planned(GW)Forecast Additions(GW)Retirements(GW)Net Changes(GW)Gas 6.5 8.0(12.3)2.1 Nuclear Coal and biomass 0.1 (31.6)(31.5)Solar 4.2 0.8 5.0 Wind 3.0 1.0 4.0 Hybrid 2.8 0.3 3.1 Other renewables 0.5 0.5 Storage 25.2 24.6 49.7 Total

232、 42.2 34.6(44.0)32.8 SOURCES:Initial inputs for active projects and forecast additions are interconnection requests as reported in October 2024 from CAISO,homepage;ERCOT,homepage;ISO New England,homepage;MISO,homepage;NY-ISO,homepage;SPP,homepage.Available capacity for generators requesting intercon

233、nection in non-ISO regions 21 Resource Type Total Planned(GW)Forecast Additions(GW)Retirements(GW)Net Changes(GW)is as reported in EIA,“august_generator2024.xlsx.”Retirements estimated from resources with planned retirement dates in 20252030 are as reported in EIA,“august_generator2024.xlsx.”Figure

234、2.7 shows the change in capacity by fuel type when the previously described steps are applied in aggregate for the contiguous United States.The most significant reduction in capacity occurs during the completion rate phase(Step 2).Following this,most capacities,aside from storage,are significantly r

235、educed because of the low ELCC factors(Step 3).The retirement of resources with higher ELCC factors further reduces available capacity in 2030.Ultimately,the net incremental available capacity amounts to 33 GW.Figure 2.7.Comparison of Planned and Expected Additions to U.S.FTM Power Capacity by 2030

236、Storage,451Storage,50Solar,314Solar,5Wind,114Wind,4Hybrid,68Hybrid,3Gas,64Gas,2Coal,-31Total1015GWTotal151GWTotal77GWTotal-44GWTotal33GW-200020040060080010001200PlannedNameplateCapacityExpectedNameplateCapacityAvailableCapacityAdditionsPlannedRetirementNetAvailableCapacityCapacityadditions(GW)22 SOU

237、RCES:Planned nameplate capacity is estimated from October 2024 interconnection queues from CAISO,homepage;ERCOT,homepage;ISO New England,homepage;MISO,homepage;NY-ISO,homepage;and SPP,homepage;and non-ISO planned capacity is as reported in EIA,“august_generator2024.xlsx.”Expected nameplate capacity

238、applies historical completion rates to planned capacity,as estimated from CAISO,NYISO,SPP,and MISO interconnection queues and self-reported by ERCOT.Non-ISO and NE-ISO completion rates are estimated as averages of completion rates across other regions.Available capacity,then,also takes transmission

239、line losses and ELCC estimates for CAISO,NYISO,ERCOT,NE-ISO,SPP,and MISO into account,an average across the regions used for non-ISO resources.Planned retirement resources are estimated by applying ELCC and transmission loss factors to nameplate capacity of resources scheduled for retirement in 2025

240、2030 that appear in EIA,“august_generator2024.xlsx.”Net available capacity for 20252030 is equal to available capacity additions minus planned retirements.The net change in FTM regional available capacity is shown in Table 2.5,alongside the breakdown of cumulative capacity by active projects,forecas

241、t additions,and retirements.Most interconnection requests are proposed in ERCOT,where approximately 87 GW of interconnections were proposed in 2024.Interconnection queue requests have been increasing in ERCOT in recent years,leading to the large volume of forecasted additions.Requests are currently

242、paused in CAISO and PJM.Net change in available capacity by region is positive for ERCOT,NY-ISO,and NE-ISO.However,retirements exceed the available capacity additions in MISO,PJM,CAISO,SPP,and the non-ISO region,resulting in net negative capacities.Table 2.5.Available Capacity by Region for Active P

243、rojects,Forecast Additions,Retirements,and the Net Final Estimate(20252030)Available Capacity(GW)CAISO PJM ERCOT MISO NY-ISO NE-ISO SPP Non-ISO Regions Total Active projects 1.2 2.1 30.0 2.3 0.3 2.0 0.9 3.9 42.0 Forecast additions 32.0 0.2 0.1 1.3 0.9 35.0 Retirements(2,9)(7.4)(2.3)(14.0)(0.0)(0.1)(

244、3.5)(13.0)(44.0)Net cumulative 1.7 5.3 59.0 12 0.3 3.3 1.7 9.4 33.0 NOTE:Initial inputs for active projects and forecast additions are interconnection requests as reported in October 2024 from CAISO,homepage;ERCOT,homepage;ISO New England,homepage;MISO,homepage;NY-ISO,homepage;SPP,homepage.Available

245、 capacity for generators requesting interconnection in non-ISO regions is as reported in EIA,“august_generator2024.xlsx.”Retirements estimated from resources with planned retirement dates in 20252030 are as reported in EIA,“august_generator2024.xlsx.”Estimating the Potential Impacts of Different Lev

246、ers to Unlock Additional Capacity As illustrated in Figures 2.1 and 2.7,we estimate the capacity lost at each step of our calculation.The most significant loss occurs in Step 2,during which more than 860 GW of projects fail to reach completion.Approximately 50 percent of the remaining capacity is lo

247、st in subsequent steps.As detailed in the“Background and Motivation”section of Chapter 1 and in 23 Appendix A,there are numerous reasons for a project to be withdrawn from the queue or fail to achieve completion.One major reason is the combination of increasing interconnection requests and limited t

248、ransmission capacity,which results in severe delays and nationwide backlogs.61 Another reason is that interconnection costs,driven primarily by the need for broader transmission system upgrades,lead to project withdrawals.62 While FERC and DOE are aiming to transform bulk transmission interconnectio

249、ns by 2035,63 the speed and impact of such a reform might not resolve short-term resource shortage issues.One short-term(one-to three-year)solution proposed by DOE involves developing one-off interventions to mitigate queue backlogs.These interventions include hiring additional temporary staff,outso

250、urcing,implementing temporary fast-tracking,and temporarily rationing queue space.64 For medium-term(three-to five-year)solutions,DOE recommends enhancing and creating fast-track interconnection options that include surplus interconnection service,generation replacement service,and energy-only inter

251、connection service.We use“what if”scenario analysis to quantitatively evaluate the potential of five proposed solutions,or levers:1.prioritize high-CAF projects(mainly storage and gas)2.delay retirements 3.add battery storage to stand-alone VRE projects 4.prioritize highcompletion milestone projects

252、 5.make general improvements to the interconnection process.Additionally,we examine the potential for general improvements to the interconnection and T&D system,a process that requires many longer-term improvements.In the following sections,we provide a brief description of each scenario,outline the

253、 methods and assumptions used for the quantitative analyses,and offer a qualitative assessment of feasibility and timelines for each scenario.The qualitative assessment leverages ongoing efforts from each region,the DOEs evaluation,and our expertise.For the quantitative analysis,we consider only ave

254、rage CAF and completion rates for each resource type,though CAF and completion rates also vary by region;therefore,levers are considered for the contiguous United States.For example,we assume that all gas project completion rates will increase by the same percentage,regardless of location number.The

255、 average CAF for each resource type in our study is shown in Figure 2.4,while completion rates are shown in Figure 2.8.Figure 2.8 reproduces the average values from Table 2.3,weighted by the capacities of planned projects.The average expected completion rates(Figure 2.8)differ 61 See FERC,“E-1:Commi

256、ssioner Clements Concurrence on Order No.2023,”for summaries of comments that were submitted in Docket No.RM22-14-000.62 Seel et al.,“Generator Interconnection Costs to the Transmission System.”63 Gorman et al.,Transmission Interconnection Roadmap;Office of Public Participation,Federal Energy Regula

257、tory Commission,“Explainer on the Interconnection Final Rule.”64 Gorman et al.,Transmission Interconnection Roadmap.24 from the historical completion rates(Table 2.3)because of variations in resource mix and project sizes within the active pipeline.For comparison of completion rates and CAFs by reso

258、urce type,refer to Figure A.3 in Appendix A.Storage,biomass,gas,coal,and hydro have relatively high CAFs compared with other resource types,indicating that these resources are more capable of contributing to firm load.Furthermore,hybrid projects have more available capacity than renewables but less

259、available capacity than the higher-CAF resources.Although storage and gas resources have similar CAFs,the completion rates for projects of these resource types differ;see Figure 2.8.The effective completion rate for storage projects is approximately 14 percent,while that for gas projects is 24 perce

260、nt.The difference in completion rates may be attributed to storage resources being a relatively new technology trying to interconnect alongside a significant increase in interconnection volume and worsening backlogs.65 Figure 2.8.Average Expected Completion Rates Across the United States by Resource

261、 Type SOURCES:Authors analysis of historical interconnection queue data for completed projects reported by CAISO,PJM,NY-ISO,MISO,and SPP and self-reported data from ERCOT;see details in Appendix A.65 Rand et al.,“Queued Up.”0%10%20%30%40%50%CoalGasOtherBiomassSolarWindGeothermalHydroStorageHybrid 25

262、 NOTE:The bars are average values weighted by planned capacity(active projects currently in the queue along with anticipated future additions)across ISOs.These average values differ from those in Table 2.3,which are weighted by historical project capacity across ISOs.Scenario 1:Prioritize High-CAF P

263、rojects(Mainly Storage and Gas)Flexible and firm generation are more valuable than intermittent resources for reliably meeting demand.Storage,gas,biomass,coal,and hydro projects have higher CAFs than the VRE alternatives,indicating that they can make the largest contributions to available capacity w

264、hen completed.Prioritizing completion of a 1-MW gas project will provide a larger increase to available capacity than completing a 1-MW solar project.Table 2.6 summarizes our estimates for the potential increase in available capacity from prioritizing the interconnection of high-CAF projects.66 Tabl

265、e 2.6.Impact of a 1-Percent Increase in Completion Rates by Resource Type,Including Targeting High-CAF Resources Resource Type Average CAF Available Capacity Additions Retirements Net Capacity 1-Percent Increase in Completion Rates Gas 86%14.4 GW(12.3 GW)2.1 GW+0.5 GW Storage 76%49.7 GW 49.7 GW+3.5

266、GW Coal 85%(31.4 GW)(31.4 GW)Biomass 79%(0.2 GW)(0.2 GW)Hydro 75%0.5 GW 0.5 GW Total high CAF 76%64.0 GW(44.0 GW)20.0 GW+4.0 GW All other resources 21%13.0 GW 13.0 GW+1.0 GW Total 46%77.0 GW(44.0 GW)33.0 GW+5.0 GW The combined high-CAF resources contribute approximately 20 GW of the 33-GW net change

267、 in available capacity by 2030 from a 64-GW available capacity addition and a 44-GW reduction from retirement.These additions consist primarily of gas and storage,while retirements are almost exclusively from gas and coal.Beyond the marginal improvement scenario,we also examine the maximum potential

268、 achievable if conditions were improved to the full extent.The upper bound on available capacity,66 A high CAF is a CAF of 70 percent or greater,indicating a higher contribution to resource adequacy and reliability for each MW of nameplate capacity.26 if 100-percent completion rates were achieved fo

269、r all proposed high-CAF resource types,is an additional 330 GW of available capacity.This additional capacity comprises an additional 295 GW of storage and 35 GW of gas capacity that we currently estimate will not be successfully completed by 2030.The marginal impact of a 1-percent increase in all h

270、igh-CAF resources is an additional 4 GW(approximately 0.5 GW gas and 3.5 GW storage).Doubling current completion rates for gas projects would provide an additional 14 GW of available capacity,while doubling completion of storage provides an additional 48 GW of capacity.There is no perceptible increa

271、se in available capacity from proposed coal and biomass projects because so little capacity is planned.Battery storage technology is not only firm and dispatchable;it also has a small footprint and can be rapidly constructed.67 Conversely,the construction of new pipelines and gas power projects face

272、s infrastructure issues and challenges related to EPA permitting requirements.The United States has significant capacity limitations in natural gas pipelines that might not be resolved in time to enhance the project completion despite changes in interconnection priority.68 We assess this solution to

273、 be partially feasible,given that PJM has initiated one-off measures to fast-track requests for projects with higher-ELCC resources to achieve expedited operation by 2029.69 However,fast-tracking can improve completion rates to the extent that the existing grid at the interconnection point can accom

274、modate,including the availability of power for charging the storage asset.According to the DOE,this solution could be implemented within the next one to three years.70 Scenario 2:Delay Retirements Extending or retrofitting planned retirements can address short-term grid reliability and resource shor

275、tages.Currently,44 GW are scheduled for retirement.Each percentage point of planned retirements that are delayed beyond 2030 represents an additional 0.44 GW of available capacity.Delaying half of the proposed retirements would increase available capacity by 22 GW.We assess this solution to be parti

276、ally feasible given that extensions of planned retirements have been authorized in different regions.71 However,each plant may have unique reasons for its retirement plans and will need to meet applicable permitting requirements or coordinate with relevant agencies and organizations.We believe that

277、delaying existing power plant retirements can be done in the short term.67 New York State Energy Research and Development Authority,“Types of Energy Storage.”68 Avalos,Fitzgerald,and Rucker,“Measuring the Effects of Natural Gas Pipeline Constraints on Regional Pricing and Market Integration”;Carter,

278、“Infrastructure Constraints Limit Energy Scalability but Present Opportunity.”69 Projects that have an ELCC of at least 45 percent are eligible.These include thermal resources,energy storage,offshore wind,and landfill gas,according to Bielak,“Reliability Resource Initiative Straw Proposal.”70 Gorman

279、 et al.,Transmission Interconnection Roadmap.71 Sweeney,Kuykendall,and Dlin,“US Power Generators Pump the Brakes on Coal Plant Retirements.”27 Scenario 3:Add Battery Storage to Stand-Alone VRE Projects Across the United States,the average CAF between regions for solar resources is 19 percent,while t

280、hat for wind is approximately 20 percent.In comparison,hybrid projects have a higher CAF of approximately 53 percent because of additional storage capacity that can be used to smooth fluctuating generation from intermittent resources.72 Because of the intermittency inherent in renewable capacity,sol

281、ar and wind resources account for only 9 GW of available additional capacity,despite the 63 GW of additional nameplate capacity.However,battery storage can be connected to solar and wind projects to reduce intermittency and increase availability.Efforts to hybridize all renewable additions between 2

282、025 and 2030 could unlock up to 30 GW of additional available capacity,equivalent to 0.3 GW for every percentage of intermittent projects that are hybridized.If all solar resources were made hybrid,the additional four-hour battery storage could make an additional 22 GW available,while the addition o

283、f storage to all wind projects could add 8 GW.The increase is larger for solar projects because there are more solar projects in interconnection queues,and the benefit from adding storage capacity is greater.We assess this solution to be partially feasible,as FERC Orders 2003,845,and 2023 have estab

284、lished a framework that allows interconnection customers to expedite requests by utilizing existing transmission capacity.However,colocating battery storage can increase available capacity only to the extent that surplus interconnection rights can accommodate it without triggering new studies or inc

285、reasing upgrade costs.According to the DOE,this solution could be implemented within the next three to five years.73 Scenario 4:Prioritize HigherCompletion Milestone Projects This scenario emphasizes prioritizing projects that have completed significant stages of the capacity planning process.There

286、are 7.9 GW of available capacity from projects that have finalized a construction services agreement(CSA)or a similar interconnection milestone,such as an interconnection services agreement or wholesale market participant agreement.These resources have passed most regulatory hurdles and have an incr

287、eased probability of being successfully completed.74 All but 0.2 GW of this capacity was proposed more than 12 months ago,indicating that this scenario focuses mainly on resources already in interconnection queues and not forecast additions.Policies that prioritize resources that already have CSAs a

288、nd similar agreements over ones that do not will make 63 GW available.Each percentage point increase in completion rates for these resources is equal to increasing available capacity by 0.8 GW.Note 72 Wang,Ciobotaru,and Agelidis,“Power Smoothing of Large Solar PV Plant Using Hybrid Energy Storage.”W

289、hen we weight by regional projections of nameplate capacity,solar drops to approximately 12 percent,wind marginally increases to 21 percent,and hybrid projects have an average CAF of approximately 64 percent.73 Gorman et al.,Transmission Interconnection Roadmap.74 Monitoring Analytics,LLC,2024 Quart

290、erly State of the Market Report for PJM.28 that completion rates based on resource types and region are low,with an average of approximately 11 percent.Because completion rate estimates do not take project status into account,the available capacity of these resources is potentially underestimated.75

291、 We assess this solution to be partially feasible.For example,CAISO proposed prioritizing interconnection studies based on project readiness and competition to manage the backlog from an exceptional volume of requests.76 However,fast-tracking can improve completion rates only to the extent that the

292、existing grid at the interconnection point can accommodate.According to the DOE,this solution could be implemented within the next one to three years.77 Scenario 5:Make General Improvements to the Interconnection Process The United States backlogged interconnection process is the single largest barr

293、ier to increasing the power capacity available for AI.Large-scale improvements will require long-term efforts because of subsidiary barriers related to established legal and regulatory precedent for interconnection queue processes.Continued efforts to strengthen and streamline the interconnection st

294、udy process can improve the grid access timeline.Our analysis of interconnection queue data indicates that a 1percentage point improvement in all completion rates is equivalent to a 5-GW increase in available capacity.If completion rates increase by up to 5 percentage points,an additional 26 GW may

295、become available,increasing net available capacity to approximately 58 GW.Note that completion rates are one of the largest sources of FTM uncertainty.Small changes in completion rates can lead to large changes in available capacity.Although this analysis has focused on the potential levers to incre

296、ase available capacity,if completion rates were to fall,the capacity additions to the U.S.electricity grid could be substantially less than those forecasted here.We assess this solution to be marginally feasible,as improving the interconnection process requires further development and relies on cont

297、inued technological progress,which might not be achieved if required short-term actions are unsuccessful.We also anticipate regional variations in how and when the interconnection queue process is streamlined and made more efficient.According to the DOE,this solution would require more than five yea

298、rs to implement,extending beyond 2030.78 75 Monitoring Analytics,LLC,2024 Quarterly State of the Market Report for PJM,estimates completion rates by interconnection milestone,with rates indicating the increasing Bayesian probability of a project going into service as each step of the planning proces

299、s is completed and regulatory documents are submitted.Completion rates are based on historical rates,regardless of project status.76 Weaver et al.,“Re:California Independent System Operator Corporation Docket No.ER24-_-000.”77 Gorman et al.,Transmission Interconnection Roadmap.78 Gorman et al.,Trans

300、mission Interconnection Roadmap.29 Summary Table 2.7 shows the estimated impacts of each lever.Levers are not mutually exclusive,and,with sufficient resources,multiple levers may be used to unlock additional available capacity.However,implementing these solutions can have secondary impacts;for insta

301、nce,prioritizing and fast-tracking certain projects might lower the completion rate for other projects because of such constraints as limited interconnection hosting capacity,supply chain issues,and restricted staffing and resources.Although we did not include these secondary impacts and risks in th

302、is report,we recognize their significance.We will explore these issues further in our subsequent report.Furthermore,we qualitatively assess the feasibility and timeline for each scenario,drawing on ongoing efforts from each region,the DOEs assessment,and our expertise to the fullest extent possible.

303、We discuss policy recommendations that can unlock this capacity in Chapter 3.Table 2.7.Summary of Potential Levers to Unlock Additional Capacity Scenario Qualitative Assessment of Feasibility and Timeline Additional Available Capacity from Reference Case Incremental Potential Maximum Potential 1.Pri

304、oritize high-CAF projects(mainly storage and gas)Partially feasible and short term 4 GW from a 1-percent increase in completion rate 330 GW total assuming a 100-percent completion rate:Storage:295 GW Gas:35 GW 2.Delay retirements Partially feasible and short term 0.44 GW from a 1-percent decrease in

305、 retired capacity 44 GW total from delaying all planned retirement:Gas:12 GW Coal:32 GW 3.Add battery storage to stand-alone VRE projects Partially feasible and medium term 30 GW total from adding storage for all stand-alone VRE projects:Solar:22 GW Wind:8 GW 4.Prioritize highcompletion milestone pr

306、ojects Partially feasible and short term 0.8 GW from a 1-percent increase in completion rate 63 GW total assuming a 100-percent completion rate 5.Make general improvements to the interconnection process Marginally feasible and long term 5 GW for a 1-percent increase in completion rate across all reg

307、ions and all resources Not applicable NOTE:The estimated maximum potential for delaying retirements is a one-hundred-fold multiple of the 1-percent marginal change,while the maximum potential from prioritizing high CAF and projects that have achieved significant interconnection milestones is less th

308、an 100 times the 1-percent change,because completion rates for different projects in the base case are greater than 0 percent.For example,guaranteeing completion of a project that was previously expected to be completed 20 percent of the time will deliver five times as much expected available energy

309、,not 100 times as much.30 Projecting BTM Available Capacities by 2030 We also assessed potential BTM capacity,which is defined as power-generation capacity that is expected to be available on the side of the consuming entity.This definition has three constraints.The first is that the entity must be

310、connected to the grid,as off-grid capacity falls outside the scope of this report.79 The second is that we focus only on projects that are in regular operation and can provide a substantial contribution to the overall power capacity.This means that capacity that is available only in exceptional circ

311、umstances,such as for backup generation,is also excluded from this analysis.80 Finally,we include only resources that increase the overall supply of power-generating capacity of the grid.Therefore,we exclude measures that reduce demand,such as demand response measures and energy efficiency measures.

312、We do,however,include various forms of storage.Ultimately,we are interested in power capacity that can provide a useful contribution to powering data centers.While common BTM configurations designed around solar PV and battery storage can be used to power data centers,fully powering data centers wit

313、h these resources requires a significant overbuilding of power capacity to account for intermittency,peaks in demand,and grid stability requirements.81 BTM resources by themselves are therefore unlikely to provide the reliable power at the geographically localized scale required by AI data centers.N

314、evertheless,the BTM deployment of solar PV with battery storage means that the corresponding load is drawing less power from the grid.This reduces congestion in the grid and frees up capacity for loads that benefit from grid access.BTM resources can be employed to reduce peak demand and reduce the n

315、eed for peaking plants that see use for only a few hours each year.82 To assess the real-world reliability contribution that BTM resources can offer,we need to assess the percentage of the overall nameplate capacity that would be available in the case of looming capacity shortfalls.83 For this reaso

316、n,we scale the projected nameplate capacity of BTM resources by their ELCCs to determine the available future BTM capacity that can reliably be deployed to cover a load.More information on the methodology is included in Appendix B.79 The potential of fully off-grid capacity,such as bridge power solu

317、tions,will be explored in later reports.80 The potential of backup generation will be assessed in later reports.81 Tozzi,“Why You Cant Power Your Data Center Only with RenewablesBut Should Try Anyway.”82 For example,see Figure 2 in DOE,“Clean Energy Resources to Meet Data Center Electricity Demand”;

318、and McNamara,“Energy Storage to Replace Peaker Plants.”83 Aagaard and Kleit,“Marginal vs.Average Effective Load Carrying Capability.”31 BTM Results Step 1:Estimate Projected BTM Reliable Capacity Additions First,we estimate the projected additions of BTM capacity by 2030.Here,we rely on projections

319、made by NREL,which developed a scenario-based approach that outlines the possible U.S.power-sector futures that incorporate all power-generating technologies.84 Our BTM assessment focuses on solar generation and battery storage technologies.Table 2.8 shows NRELs BTM projections for 2024 by resource

320、type and region.Across the ISOs,over 51 GW is deployed,of which almost 39 GW is solar PV capacity and 12 GW is battery storage capacity.85 Table 2.8.NREL Projections of BTM Capacity by Resource Type and Region in 2024(GW)CAISO ERCOT NE-ISO MISO NY-ISO PJM SPP Non-ISO Regions Total BTM solar 12.8 1.7

321、 4.3 3.1 1.9 7.1 0.7 7.3 39.0 Storage 8.3 2.3 0.0 0.0 0.1 0.1 0.2 1.3 12.2 Total 21.1 4.0 4.4 3.1 2.0 7.2 0.9 8.6 51.2 Step 2:Account for Current BTM Installed Capacity The 2024 NREL projections are based on algorithms using 2022 data.We compared the 2024 NREL estimates for total installed capacity

322、with the ISO data and found that the NREL projections underestimate the installed capacity(Table 2.9).To address this discrepancy,we used the 2024 ISO data,where available,as the initial condition for our 2030 projections and used the NREL estimates where no ISO data were available.This revised base

323、line is shown in Table 2.9,and the actual ISO data are shown in red.Table 2.9.NREL Projections and ISO Data on BTM Growth by Resource Type and Region in 2024(GW)CAISO ERCOT NE-ISO MISO NY-ISO PJM SPP Non-ISO Regions Total BTM solar 17.0a 2.5b 4.0c 5.7d 5.2e 8.4f 0.7 7.3 50.8 Storage 2.0g 2.3 0.1h 0.

324、0 0.1 0.1 0.2 1.3 5.9 Total 18.9 4.7 4.1 5.7 5.3 8.4 0.9 8.6 56.8 84 NREL,“NREL Releases the 2023 Standard Scenarios.”85 PV-battery hybrid technology was excluded from the NREL modeling(Gagnon et al.,2023 Standard Scenarios Report).32 CAISO ERCOT NE-ISO MISO NY-ISO PJM SPP Non-ISO Regions Total a Wi

325、lson,“Historical BTM PV and Storage Adoption Trends in California”;red text denotes actual data.b ERCOT,2024 ERCOT System Planning;red text denotes actual data.c ISO New England,“Explainer”;red text denotes actual data.d MISO Distributed Energy Resources Task Force,“2024 OMS DER Survey Results”;red

326、text denotes actual data.e McPherson,“NYCA Renewables 2023”;red text denotes actual data.f PJM,“Distributed Solar Generation Forecast by Zone Cumulative Nameplate Capacity Includes Historical Degraded Values and HIS Forecast”;red text denotes actual data.g Nyberg,“California Energy Storage System Su

327、rvey”;red text denotes actual data.h Vu,“DER Forecast Improvements”;red text denotes actual data.The NREL projections for BTM solar and battery storage capacity additions use a two-year interval.Therefore,we used the average of the capacity additions for the two-year interval to determine the yearly

328、 growth rate from 2024 through 2030.The resulting growth rates for these technologies indicate that BTM solar PV capacity may increase by 56 GW(Table 2.10)and BTM storage capacity may increase by 36 GW(Table 2.11),respectively.Overall,this results in a projected 106-GW solar capacity and 41-GW stora

329、ge capacity.The projections show significant growth in BTM solar PV across all regions of the United States,with a doubling of nameplate capacity between 2024 and 2030.For instance,ERCOT is projected to experience an increase in solar PV capacity by 2030 of 3.5 times,growing from 2.5 GW to 8.8 GW.No

330、n-ISO regions will see an increase of 3.3 times,rising from 7.3 GW to 24 GW.Meanwhile,regions that already have a significant amount of solar PV capacity,such as CAISO,will see more-modest growth,of 1.5 times,with a 7-GW increase.For storage,the projections show a nationwide eightfold increase in in

331、stalled nameplate capacity,but the deployment is not regionally uniform.For example,while the projection for ERCOT is a fivefold increase in storageapproximately 10 GWno increase in storage capacity is projected for CAISO.We have less confidence in these projections than we do for the solar PV proje

332、ctions.33 Table 2.10.BTM Solar PV Nameplate Capacity Growth by Region Out to 2030(GW)Region Actual Forecast 2024 2025 2026 2027 2028 2029 2030 ISO regions CAISO 17.0 18.1 19.1 20.3 21.6 22.8 24.1 ERCOT 2.5 3.2 3.8 4.9 5.9 7.4 8.8 NE-ISO 4.0 4.3 4.6 4.9 5.2 5.6 6.0 MISO 5.7 6.5 7.4 8.6 9.8 11.9 13.8

333、NY-ISO 5.2 5.4 5.5 5.7 6.0 6.3 6.6 PJM 8.4 9.8 11.2 13.0 14.9 17.5 20.1 SPP 0.7 1.0 1.2 1.6 1.9 2.5 3.0 Non-ISO regions 7.3 9.2 11.1 13.8 16.4 20.2 24.1 Annual incremental 6.5 6.5 8.9 8.9 12.4 12.4 Total 50.8 57.4 63.9 72.8 81.7 94.1 106.5 Table 2.11.BTM Storage Nameplate Capacity Growth by Region Out to 2030(GW)Region Actual Forecast 2024 2025 2026 2027 2028 2029 2030 ISO regions CAISO 1.9 1.9 1.