封面-假如3万亿的AI投资热潮失败了怎么办？

Whatifthe$3trn AIinvestmentboom goeswrong? 如果3万亿美元的人工智能投资热潮出问题怎么办？ Evenifthetechnology achieves itspotential,plentyofpeople willlosetheir shirts 即便技术实现了潜力，也会有大量人血本无归 Share分享 Save保存 插图：danielstolle Sep11th2025|5minread Listentothisstory收听此故事 ITALREADY RANKS among thebiggest investment booms inmodern history.ThisyearAmerica’s largetech firmswillspend nearly$400bn on theinfrastructureneededtorunartificial-intelligence(ai)models.OpenAIandAnthropic,theworld’sleadingmodel-makers,areraisingbillionsevery few months;theircombined valuationisapproachinghalf atrillion dollars.Analysts reckon thatbytheend of2028 thesumsspentworldwide ondatacentres willexceed$3trn. 它已经跻身现代史上最大的投资热潮之列。今年，美国大型科技公司将在运行人工智能（AI）模型所需的基础设施上投入近4,000亿美元。OpenAI和Anthropic这两大全球领先的模型制造商每隔几个月就筹资数十亿美元；它们的合并估值正接近5,000亿美元。分析师估计，到2028年底，全球在数据中心上的支出将超过3万亿美元。 Thescaleofthese bets issovast thatitisworthasking what willhappen atpaybacktime.Even ifthetechnology succeeds,plenty ofpeople willlosetheir shirts.Andifitdoesn’t,theeconomic and financial painwillbeswift and severe. 这些押注的规模如此庞大，以至于值得问问在回报到来时会发生什么。即使技术成功，也会有大量人血本无归。如果技术失败，经济和金融上的痛苦将来得既迅速又严重。 Investorsalways flocktopromising technologies,buttheAIrush ismore extreme thanmanypast booms.Boosters saythatartificial general intelligence(AGI)—models thatarebetterthantheaverage human atmost cognitive tasks—could beonly afewyearsaway.The firstfirmtoachieve itcould reapunimaginable returns.Investorsandinnovatorsknow they may notbe backing therightmodel.But iftheyspend slowly andcautiously,they may aswell notbother tospendatall. 投资者总是蜂拥于前景可观的技术，但这次的AI热潮比许多以往的繁荣更为极端。支持者称，通用人工智能（AGI）——在大多数认知任务上优于普通人的模型——可能仅在几年之内就会出现。第一个实现它的公司可能收获难以想象的回报。投资者和创新者知道他们可能并未押注在正确的模型上。但如果他们花费得太慢、太谨慎，那还不如根本不去投入。 Consequently,arelentlessinvestment race isunder way,with bigtechfirms splurgingonthecomputing power needed tobuild thebiggest models.Agrowing castofextraplayers,from property developers toelectricity generators,has been drawn in.Oracleisthelatesttojointheparty.Itsvaluesurged on September 10thafter itpublished anambitiousforecastfor itsAI-relatedcloud business,brieflyturning itsboss,LarryEllison,into theworld’s richest man. 因此，一场不懈的投资竞赛正在进行，大型科技公司纷纷在构建最大模型所需的计算能力上挥金如土。越来越多的参与者被吸引进来，从房地产开发商到发电企业。Oracle是最新加入的公司。9月10日其公布了对AI相关云业务的雄心勃勃的预测后，市值飙升，短暂地使其负责人Larry Ellison成为世界首富。 Nomatter what happens,many investors willlosemoney.Intherosiestscenario agiwillarriveand usherinanewworld ofeconomic growth ofperhaps 20%ayear,aswewroteinJuly.Some shareholders would enjoy astronomical returns;many others wouldfacebiglosses. 无论发生什么，许多投资者都会亏损。正如我们在七月所写，最乐观的情形是AGI到来并带来每年或许20%的新经济增长。一些股东将获得惊人的回报；许多其他人则会面临巨大的损失。 Moremundane scenarios should alsobeconsidered,however.The technology mayevolveinways thatinvestors do notexpect.When alternating current eventuallyprevailedinAmerica inthe19thcentury,forinstance,direct-current electricity firmswereovershadowed and forced toconsolidate.Today investors reckon thattheprobableaiwinners arethose thatcanrunthebiggest models.But,aswe reportthisweek,earlyadopters areturning tosmaller language models,whichcould suggest thatlesscomputing capacity may beneeded afterall. 然而，也应考虑更为平凡的情形。技术的发展可能与投资者的预期不同。例如，19世纪交流电最终在美国占据主导时，直流电企业被边缘化并被迫整合。如今投资者普遍认为，可能的AI胜出者将是能够运行最大模型的企业。但正如我们本周报道的那样，早期采用者正在转向更小的语言模型，这可能表明最终所需的计算能力会更少。 Ortheroadtowidespread adoption could beslower and bumpier thaninvestorsexpect,givingtoday’s ailaggards afighting chance.Niggles inthetechnology,thedifficultyofquickly supplying electricpower,ormanagerial inertiacould mean thattake-upismore gradual than firsthoped.Asthey revisedown theirexpectations forairevenues,many investors and creditors could become lesswillingtocountenance hugeinvestments.The flowofcapitalcould slow;some startups,strugglingunder the weightoflosses,could fold altogether. 或者，普及之路可能比投资者预期的更慢、更坎坷，从而给当下的AI落后者留下一线生机。技术上的小问题、迅速供应电力的困难或管理惯性，可能导致采用速度比最初预期更为缓慢。当他们下调对AI收入的预期时，许多投资者和债权人可能会变得不那么愿意支持巨额投资。资本流动可能放缓；一些在亏损重压下苦苦支撑的初创企业，可能会完全倒闭。 Whatwould such an aichillbe like?For astart,alotoftoday’s spending could proveworthless.After its19th-century railway mania,Britainwas leftwith track,tunnels andbridges;much ofthisservespassengers today.Bitsand bytes stillwhizz through thefibre-opticnetworks built inthedotcom years.The aiboom may leave alessenduringlegacy.Although theshellsofdatacentres and new power capacity could findotheruses,more thanhalfthecapex splurge has been onservers and specialised chips thatbecomeobsoleteinafewyears. 这种人工智能寒冬会是什么样子？首先，今天的许多支出可能被证明是毫无价值的。19世纪的英国铁路狂热过后，留下了轨道、隧道和桥梁；其中大部分今天仍为乘客服务。比特与字节依然在互联网泡沫时期建设的光纤网络中穿梭。人工智能繁荣可能留下的遗产则不那么持久。尽管数据中心的外壳和新增电力产能可以找到其他用途，但超过一半的资本支出浪潮投入在几年内就会变得过时的服务器和专用芯片上。 Thegood news isthattoday’s financial system could probably absorb theblow.Sometechnologicalbusts have been brutal;afterBritain’srailway bubble burst inthe1860sbankswere leftwithbiglosses,leading to acreditcrunch.So far,however,much oftheinvestmentindatacentres hasbeen bankrolled from thedeep well ofbig-tech profits. 好消息是，现今的金融体系大概能够吸收这一冲击。一些技术泡沫的破灭曾极为残酷；在19世纪60年代英国铁路泡沫破裂后，