电力优先：数据中心发展的新剧本

The new playbook to energize AI data center development Authors Neeraj Vadhan Senior Managing Director,Supply Chain & Engineering,Data Center Services Lead Chief Executive OfficerDLB Associates David is a visionary entrepreneur with over 25years of experience. He is a licensed ProfessionalEngineer in 49 states, Certified Energy Manager,LEED Accredited Professional and has chairedASHRAE TC9.9. As co-owner of several companies, Neeraj leads the Data Center Services practiceof Accenture, drawing on years of experienceleading complex global technology and operationsengagements for major clients in Silicon Valley and James Hebblethwaite Bethany Patton Managing DirectorSoftware & Platforms and Managing Director,Industry & Enterprise, Utilities Along with leading Accenture’s AI Infrastructureoffering, James serves as UK & Ireland lead forIndustry & Enterprise Reinvention, helping clientstackle the biggest challenges in this hypergrowthsector. He has deep expertise in B2B, digital Bethany has nearly 20 years of experience in powergeneration and utilities. Before Accenture, sheworked as a mechanical design engineer for a largeutility company, and in operations at a nuclear power Caroline LaroseSenior Manager,Supply Chain Paul Johnson Senior ManagerSoftware & PlatformsResearch Lead Paul specializes in research and analysis focusing ondigital infrastructure, cloud and AI demand. He worksclosely with industry leaders and cross functional teams Caroline leads client work at the intersection ofsustainability, energy transition and infrastructuretransformation. In the data center sector, shehelps operators solve power and infrastructurechallenges, assessing stranded capacity, flexible 123465 Introduction: The data centersthat get power (and the onesthat don’t)Page 4 The cost of falling behindPage 6 Community approval can makeor break data center growth Page 8 Today’s data center demandis outpacing the grid Page 11 Building against a movingpower targetPage 13 Four no-regret moves to start today 1. Site for power readiness2. Earn community permission3. Engineer flexibility into design 78 Putting strategy into actionPage 22 The data center developers whoget it rightPage 24 Introduction:The data Picture this: Two data centerdevelopers submit separate powerrequests to their utilities on the sameday. The first files a 500-megawatt(MW) request, a rough site mapand a target go-live date. Thesecond brings a documented loadprofile, a phased ramp schedule,a curtailment commitment and a centers thatget power(and the ones Both want the same outcome.Only one secures power on schedule. For years, the data center industry has treated power as a transactionalproblem to solve after selecting a site. That approach no longer works.The projects moving forward today are operating in partnerships across The cost of taking either power availability or local support for grantedis significant. For an AI model provider, a single year of delay on a 100MW facility defers more than US$1 billion in revenue. At the currentmarket speed, deferred revenue quickly becomes lost revenue and lost $1B+ estimatedrevenuedeferred from Unlike chips, where capital and relationships can move buyers upthe queue, power cannot be secured simply by spending more. Now, with AI pushing demand to unprecedented levels, the race tosecure as much capacity as possible is compounding the problem.Utilities and infrastructure providers simply cannot meet large capacity Our analysis and client experience point to four moves developers canmake before breaking ground to improve their path to energization.Winning developers will select potential sites based on power With interconnection requests outpacing what utilities can realisticallydeliver in the next three to five years, and behind-the-meter (BTM)infrastructure also in short supply, the first data centers to come onlinewill be those that are easiest to serve and deliver clear benefits to their The costof fallingbehind Global CapEx investments for AIinfrastructure are expected to reachUS$1 trillion in 2027 and grow tonearly US$1.6 trillion by 2031.2At current build costs, a 100 MW “Speed equals revenue.Whoever wins the speed Sophie Smith,Vice President of theAmericas at Soben The new playbook to energize AI data center development Speed requires both planning a path to power and getting all the keystakeholders aligned at the start. For an AI model provider, long delays cede room for rivals to train andlaunch the next generation of models and accumulate the usage data For colocation operators, a one-year delay on a 100 MWfacility defers approximately US$235 million in annualizedlease revenue, and lease rates continue to climb at Delay risk is not limited to monthly rent revenue. Prolonged delays cantrigger missed financing milestones, broader financing strain, loss ofproject control through the exercise of step-in rights and, ultimately, theloss of committed tenants. The colocatio