《OpenAI:2024基础设施决定命运:美国在民主AI领域的投资经济回报报告(英文版)(15页).pdf》由会员分享,可在线阅读,更多相关《OpenAI:2024基础设施决定命运:美国在民主AI领域的投资经济回报报告(英文版)(15页).pdf(15页珍藏版)》请在三个皮匠报告上搜索。
1、SEPTEMBER 2024Infrastructure is Destiny Economic Returnson US Investmentin Democratic AIOverviewRevolutionary technology propels advances in infrastructure.Capital flows determine where and how theinfrastructure is built.Those decisions determine whether a society leads or lags on technologicalinnov
2、ation,often with far-reaching consequences.This is why infrastructure is destiny.The United States leads the world today in development of artificial intelligence because of decisionsmade decades ago to install fiber-optic cables,coaxial lines and other broadband infrastructure that putthe country a
3、t the forefront of the early digital revolution.The 1996 Telecommunications Act,withbipartisan support from forward-thinking lawmakers,reinforced the infrastructure as a national strategy.As revolutionary as electricity,and promising similarly distributed access and benefits,AI can power areindustri
4、alization across the US,extend its global competitiveness,and boost national,state andhousehold finances for the long term.Investment to extend the US lead in AI can yield tens of thousandsof jobs;significant growth in GDP;a modernized,cleaner energy grid and energy policy,featuring nuclearpower;and
5、 a state-of-the art network of semiconductor manufacturing facilities invigorating localeconomies across the country.These are New Deal-sized stakes and scale.Capital spending on AI already rivals the mainframe era of the late 1960s and the fiber optic deploymentof the late 1990s1 with an estimated$
6、175 billion in global infrastructure funds waiting to becommitted.2The question is not whether that funding will flow,but where.If it doesnt flow intoUS-backed global infrastructure projects that advance a global AI that spreads the technologys benefitsto the most people possible,then it will flow t
7、o China-backed projects that leverage AI to cement andexpand autocratic power.There is no third option.Analysts expect the build to require unprecedented scaling of compute.2023 saw massive growth indemand for AI data centers around 167%year-over-year3 that shows no sign of slowing.The totalmarket f
8、or GPU and AI ASIC chips alone is expected to exceed$150 billion by 2025 and exceed$230billion in 2029.4Goldman Sachs now estimates that around 47GW of incremental capacity is needed toserve data center-driven load growth in the US through 2030.Earlier this year,OpenAI engaged outside experts to wor
9、k with us on an analysis of the potential jobs andGDP impacts of building 5GW data centers in various locations across the US to help informconversations between industry and government about the economic benefits of this next stage ofgrowth for Americans and their communities.4Generative AI:Impacts
10、 on Processor,Memory,Advanced Packaging and Substrates,Yole Intelligence 20243OpenAI internal analysis,20242Global Infra Leaders 2024 Global Report,Houlihan Lokey Digital Infrastructure Industry Update Q2 2024,Preqin.Sum based on the followingscreening criteria:Asset Class:Infrastructure;Strategy:Co
11、re,Core-Plus,Debt,Opportunistic,Value Added;Sector:Energy,RenewableEnergy,Telecommunications,Utilities1A severe case of COVIDIA:prognosis for an AI-driven US equity market,JP Morgan,Sept.3,20242Based on computational demand trends for frontier AI models,we estimated at the time we undertookthe analy
12、sis that a 5GW data center campus would meet the projected requirements,and that the profileof a single 5GW data center would include:30 millionsquare feet14,000constructionworkers2 millionGPUs$100 billionin investment$40 billionin annualrevenueBased on our resulting analysis,employment and GDP grow
13、th in the following example states wouldresult from the construction of one 5GW data center:StateNew jobs created/supported*Growth in GDPArizona42,954$6,908,848,125California38,327$6,955,707,185Georgia43,861$7,042,536,567Michigan41,900$6,605,018,124Nevada36,321$6,214,337,553New York33,204$6,361,394,
14、870North Carolina41,625$6,629,619,203Ohio42,769$6,783,610,663Pennsylvania40,996$6,646,750,168Texas44,340$7,128,132,155West Virginia34,324$5,575,422,922Wisconsin40,253$6,406,250,583*14,000 construction jobs+jobs supported through business-to-business transactions+jobs supported by worker spending.Tot
15、als differ due to jurisdictional differences in labor costs,etc.See Methodology for more details.3Each data center is a hub of durable economic growth.In addition to new construction jobs,thousands ofadditional jobs will be supported in adjacent and related sectors,and more will be supported through
16、 anoverall increase in local economic activity.All of this stems from data center investments.We estimatethe annual economic impacts of operations for a single 5GW data center to be:80 employees per100MW of power,equaling4,000 employees total$27.97 billionin cost of revenue andoperating expensesEsti
17、mating$40 billion in annual revenue5and using 2028 dollars to allow for construction,OpenAIcalculates the following annual impacts:StateNet new jobs created/supported*Growth in GDPArizona46,595$19,667,170,624California36,371$17,991,414,075Georgia46,404$19,376,372,902Michigan48,720$19,334,237,823Neva
18、da35,730$17,706,285,785New York37,314$19,926,313,370North Carolina43,391$19,434,623,329Ohio41,050$18,384,464,952Pennsylvania42,130$19,148,832,827Texas48,279$20,396,059,524West Virginia31,671$16,186,312,913Wisconsin38,512$17,725,148,272*4,000 jobs+jobs supported through business-to-business transacti
19、ons such as power,utilities and building operations+jobssupported by worker spending.Totals differ due to jurisdictional differences in labor costs,etc.See Methodology for more details.5OpenAI internal analysis4The CHIPS Act has been a strong first step toward onshoring leading-edge semiconductorman
20、ufacturing.Construction of new manufacturing facilities to boost computing power already is drivinggrowth in US GDP,6but this is just one front in the required forward movement on infrastructure thenext front is unprecedented investment in data centers and power,propelled by policies that support ne
21、wenergy and data center capacity.Who will control the future of AI is the urgent question of our time.The rapid progress being made meansthat we face a strategic choice about what kind of world we are going to live in one in which the US andallied nations advance a global AI that spreads the technol
22、ogys benefits and opens access to it,or one inwhich China and other nations that dont share our values use AI to cement and expand their power?6Bloomberg,Aug.22,20245Chinas Infrastructure Surge At-a-GlanceThe US currently has a lead in AI development,but continued leadership is far from guaranteed.A
23、uthoritarian governments the world over are willing to spend enormous amounts of money to catch upand ultimately overtake us.The Peoples Republic of China aims to be the global leader in AI by 2030.Chinas AI ecosystemChina already counts more than 230 LLMs,more than 180 of which have been approved f
24、orpublic use by the government.Notable GPT-4o level AI models include:Alibabas Qwen2,01AI Yi-34b,SenseTime 5.5,WuDao,and Baidu Ernie 4.0.DeepSeek-Coder-V2 outperforms GPT-4 Turbo on evals.These models excel at Chinese language tasks.Hallmarks of Chinas infrastructure buildRecently approved 11 nuclea
25、r reactors across five sites an investment of at least 220 billionyuan($31 billion),with construction expected to take about five years.More nuclear reactorsunder construction than any other nation in the world:10 new reactors approved per year for eachof the last two years.717 AI Pilot Zones to hel
26、p grow its AI industry via financial support and favorable local regulation.For its“Eastern Data and Western Computing”project,a fully integrated national computingnetwork,eight national computing power hubs planned to enable“supercomputing-as-a-service”,providing more efficient access to advanced c
27、ompute.Its Digital Silk Road initiative supported the development of 155 AI-related projects in 64 countriesbetween 2000 and 2017.8The initiative also supports“Luban Workshops”vocational training program to develop IT talentacross the Global South,including 30 workshops in 25 countries between 2016-
28、2023.Through Safe City/Smart City,Huawei,ZTE Corporation,Hikvision,Dahua,Alibaba and othersdeploy AI to make city services more efficient,focusing on improvements to traffic flow,logistics,law enforcement and more.Chinas compute capacity goalsChina is#2 in global compute capacity behind the US.On ta
29、rget to hit 300 EFLOPS by 2025.8RAND7Bloomberg,Aug.19,20246Data Center Impacts on Jobs and GDPUnprecedented investment is required for the US to secure its economic and national security.Thisinvestment will in turn yield economic benefits at a scale and speed not seen since the New Deal nearly acent
30、ury ago.The jobs and GDP calculations below only begin to reveal the magnitude of thistransformation since they dont include construction of new energy sources or actual semiconductormanufacturing.Each data center is a hub of durable economic growth.In addition to new construction jobs,thousands ofa
31、dditional jobs will be supported in adjacent and related sectors,and more will be supported through anoverall increase in local economic activity.All of this stems from data center investments.Based on ourinternal analysis,the sample states below would see the following impacts from construction of
32、one5GW data center:For the economic impacts of construction of one 5GW data center:30 millionsquare feet14,000constructionworkers2 millionGPUs$100 billionin investment$40 billionin annualrevenueFor the annual economic impacts of operations for a single 5GW data center:80 employees per100MW of power,
33、equaling 4,000employees total$27.97 billion incost of revenueand operatingexpenses9An estimated$40 billion inannual revenue102028 dollars toallow for timefor construction10OpenAI internal analysis9OpenAI internal analysis7Arizona14,000 NEWCONSTRUCTIONJOBSSupports 7,127 jobs in the community21,828 jo
34、bs through spendingTotaling 42,955 jobs created and supportedAltogether driving$5,299,407,046 in labor incomeAltogether driving$6,908,848,125 in GDP4,000 NEWDATA CENTERJOBSSupports 26,016 jobs in the community16,580 jobs through spendingTotaling 46,596 jobs created and supportedAltogether driving$4,
35、105,372,491 in labor incomeAltogether driving$19,667,170,624 in GDPCalifornia14,000 NEWCONSTRUCTIONJOBSSupports 6,469 jobs in the community17,859 jobs through spendingTotaling 38,328 jobs created and supportedAltogether driving$5,360,948,132 in labor incomeAltogether driving$6,955,707,185 in GDP4,00
36、0 NEWDATA CENTERJOBSSupports 19,883 jobs in the community12,487 jobs through spendingTotaling 36,3710jobs created and supportedAltogether driving$3,798,627,358 in labor incomeAltogether driving$17,991,414,075 in GDP8Georgia14,000 NEWCONSTRUCTIONJOBSSupports 8,291 jobs in the community21,570 jobs thr
37、ough spendingTotaling 43,861 jobs created and supportedAltogether driving$5,343,820,222 in labor incomeAltogether driving$7,042,536,567 in GDP4,000 NEWDATA CENTERJOBSSupports 26,794 jobs in the community15,610 jobs through spendingTotaling 46,404 jobs created and supportedAltogether driving$3,953,81
38、1,509 in labor incomeAltogether driving$19,376,372,902 in GDPMichigan14,000 NEWCONSTRUCTIONJOBSSupports 6,852 jobs in the community21,048 jobs through spendingTotaling 41,900 jobs created and supportedAltogether driving$5,243,794,293 in labor incomeAltogether driving$6,605,018,124 in GDP4,000 NEWDAT
39、A CENTERJOBSSupports 28,137 jobs in the community16,583 jobs through spendingTotaling 48,720 jobs created and supportedAltogether driving$4,240,891,539 in labor incomeAltogether driving$19,334,237,823 in GDP9Nevada14,000 NEWCONSTRUCTIONJOBSSupports 5,654 jobs in the community16,667 jobs through spen
40、dingTotaling 36,321 jobs created and supportedAltogether driving$4,848,326,114 in labor incomeAltogether driving$6,214,337,553 in GDP4,000 NEWDATA CENTERJOBSSupports 21,704 jobs in the community10,026 jobs through spendingTotaling 35,730 jobs created and supportedAltogether driving$2,982,962,556 in
41、labor incomeAltogether driving$17,706,285,785 in GDPNew York14,000 NEWCONSTRUCTIONJOBSSupports 4,798 jobs in the community14,406 jobs through spendingTotaling 33,204 jobs created and supportedAltogether driving$5,038,954,668 in labor incomeAltogether driving$6,335,283,722 in GDP4,000 NEWDATA CENTERJ
42、OBSSupports 20,018 jobs in the community13,296 jobs through spendingTotaling 37,314 jobs created and supportedAltogether driving$4,732,634,953 in labor incomeAltogether driving$19,926,313,370 in GDP10North Carolina14,000 NEWCONSTRUCTIONJOBSSupports 8,067 jobs in the community19,558 jobs through spen
43、dingTotaling 41,625 jobs created and supportedAltogether driving$5,174,063,070 in labor incomeAltogether driving$6,599,133,242 in GDP4,000 NEWDATA CENTERJOBSSupports 24,793 jobs in the community14,598 jobs through spendingTotaling 43,391 jobs created and supportedAltogether driving$3,959,860,878 in
44、labor incomeAltogether driving$19,434,623,329 in GDPOhio14,000 NEWCONSTRUCTIONJOBSSupports 7,237 jobs in the community21,531 jobs through spendingTotaling 42,769 jobs created and supportedAltogether driving$5,219,380,379 in labor incomeAltogether driving$6,783,610,663 in GDP4,000 NEWDATA CENTERJOBSS
45、upports 23,572 jobs in the community13,477 jobs through spendingTotaling 41,050 jobs created and supportedAltogether driving$3,355,222,269 in labor incomeAltogether driving$18,384,464,952 in GDP11Pennsylvania14,000 NEWCONSTRUCTIONJOBSSupports 6,463 jobs in the community20,533 jobs through spendingTo
46、taling 40,996 jobs created and supportedAltogether driving$5,328,712,024 in labor incomeAltogether driving$6,646,750,168 in GDP4,000 NEWDATA CENTERJOBSSupports 22,732 jobs in the community15,398 jobs through spendingTotaling 42,130 jobs created and supportedAltogether driving$4,087,935,836 in labor
47、incomeAltogether driving$19,148,832,827 in GDPTexas14,000 NEWCONSTRUCTIONJOBSSupports 8,378 jobs in the community21,962 jobs through spendingTotaling 44,340 jobs created and supportedAltogether driving$5,454,255,837 in labor incomeAltogether driving$7,128,132,155 in GDP4,000 NEWDATA CENTERJOBSSuppor
48、ts 27,933 jobs in the community16,346 jobs through spendingTotaling 48,279 jobs created and supportedAltogether driving$4,154,308,157 in labor incomeAltogether driving$20,396,059,524 in GDP12West Virginia14,000 NEWCONSTRUCTIONJOBSSupports 5,180 jobs in the community15,144 jobs through spendingTotali
49、ng 34,324 jobs created and supportedAltogether driving$4,595,664,845 in labor incomeAltogether driving$5,575,422,922 in GDP4,000 NEWDATA CENTERJOBSSupports 19,706 jobs in the community7,965 jobs through spendingTotaling 31,671 jobs created and supportedAltogether driving$2,519,169,092 in labor incom
50、eAltogether driving$16,186,312,913 in GDPWisconsin14,000 NEWCONSTRUCTIONJOBSSupports 6,730 jobs in the community19,523 jobs through spendingTotaling 40,253 jobs created and supportedAltogether driving$5,107,814,872 in labor incomeAltogether driving$6,406,250,583 in GDP4,000 NEWDATA CENTERJOBSSupport
51、s 23,051 jobs in the community11,461 jobs through spendingTotaling 38,512 jobs created and supportedAltogether driving$3,076,909,505 in labor incomeAltogether driving$17,725,148,272 in GDP13ConclusionThe infrastructure build needed to sustain the US edge on AI development is the kind of visionaryund
52、ertaking for which the country is uniquely known and equipped to execute a massive effort shapedby democratic values and designed to broadly distribute economic benefits.In contrast,Chinastop-down,centralized AI infrastructure strategy presents a real and competitive alternative shaped byautocratic
53、values that would deploy the technology and dole out the benefits in ways that cement its owninfluence.The US can only meet this challenge through the kind of partnership between government and industrythat creates conditions at home attracting investment and spurring rapid construction to enable us
54、 tobuild tomorrows infrastructure today.14MethodologyThe economic impact analysis included in this report does not account for the building of any newlyrequired energy infrastructure or semiconductor manufacturing that would occur d to support new datacenters meaning the economic impact estimates in
55、 this report are intentionally low.To calculate economic impacts,this report uses an input-output model developed by IMPLAN.11IMPLANs model is designed to capture all monetary market transactions between industries in a giventime period.The resulting mathematical formulae allow for examinations of t
56、he effects of a change inone or several economic activities on an entire economy(impact analysis).For more information onIMPLAN,and their assumptions made as part of their input-output analyses,refer to the articles onInput-Output Analysis and Assumptions and Detailed Key Assumptions of IMPLAN&Input
57、-OutputAnalysis.Based on IMPLANs input-output tables,a set of multipliers that reflects the capital investments andoperating expenditures from potential data centers were created to derive GDP,employment and laborincome estimates.All operating expense categories were included in GDP,Labor Income and
58、Employment calculations.For capital investments,only building construction categories were included inGDP,Labor Income and Employment calculations,using regional and industry average income data.Networking,Computer&IT Equipment,and Software were excluded from the capital investmentestimates.At OpenA
59、I,were building artificial intelligence that helps people solve hard problems.By helping withthe hard problems,AI can benefit the most people possible through better healthcare andeducation,more scientific discoveries,improved productivity,and new tools for creativity.11IMPLAN model,2022 Data,using inputs provided by the user and IMPLAN Group LLC,IMPLAN System(data and software),16905Northcross Dr.,Suite 120,Huntersville,NC 28078.www.IMPLAN.com15