谷歌称必须每六个月将AI算力翻倍以满足需求，AI基础设施负责人向员工表示

Google must double AI compute every 6 months tomeet demand, AI infrastructure boss tells employees ⾕歌称必须每六个⽉将AI算⼒翻倍以满⾜需求，AI基础设施负责⼈向员⼯表示 P U B L I S H E D F R I , N O V 2 1 2 0 2 5 7 : 3 0 A M E S T U P D AT E D 5 3 M I N A G O WATCH LIVES H A R E Jennifer Elias@ J E N N _ E L I A S KEY POINTS核⼼要点 At a recent all-hands meeting, Google’s head of AI infrastructure, Amin Vahdat, said thecompany has to race to build out compute capacity in order to meet demand. 在最近的⼀次全员会议上，⾕歌⼈⼯智能基础设施负责⼈阿明·瓦赫达特表示，为满⾜需求，公司必须加快提升计算能⼒的建设步伐。 Vahdat gave a presentation and included a slide that said, “Now we must double every6 months.... the next 1000x in 4-5 years” Vahdat在演讲中展示了⼀张幻灯⽚，上⾯写道：”现我们必须每6个⽉翻⼀番......未来4-5年内要实现1000倍增⻓”。 Google CEO Sundar Pichai spoke at the meeting and addressed employee questionsabout the potential of an AI bubble bursting and capital expenditures. ⾕歌⾸席执⾏官桑达尔·⽪查伊在会上发⾔，就AI泡沫破裂可能性和资本⽀出等问题回答了员⼯提问。 GOOGL⾕歌A(GOOGL)+7.02(+2.43%)+7.02（+2.43%） Amin Vahdat, VP of Machine Learning, Systems and Cloud AI at Google, holds up TPU Version 4 at Google headquarters inMountain View, California, on July 23, 2024. 2024年7⽉23⽇，⾕歌机器学习、系统与云AI副总裁Amin Vahdat在加州⼭景城总部展示TPU第四代芯⽚。 Marc Ganley Google ’sAI infrastructure boss told employees that the company has to double itscompute capacity every six months in order to meet demand for artificial intelligenceservices. ⾕歌⼈⼯智能基础设施负责⼈向员⼯透露，为满⾜⼈⼯智能服务需求，公司必须每六个⽉将计算能⼒提升⼀倍。 At an all-hands meeting on Nov. 6, Amin Vahdat, a vice president at Google Cloud, gave apresentation, viewed by CNBC, titled “AI Infrastructure,” which included a slide on “AIcompute demand.” The slide said, “Now we must double every 6 months.... the next 1000xin 4-5 years.” 11⽉6⽇的全员会议上，⾕歌云副总裁Amin Vahdat进⾏了题为《AI基础设施》的演讲。CNBC获得的演示⽂稿中包含“AI计算需求”幻灯⽚，其中明确指出：”现在我们必须每六个⽉翻倍......未来4-5年实现1000倍增⻓。”微信公众号 404K微信公众号 404K “The competition in AI infrastructure is the most critical and also the most expensive part ofthe AI race,” Vahdat said at the meeting, where Alphabet CEOSundar Pichaiand CFO LIVESTREAM Anat Ashkenazi also took questions from employees. “AI基础设施的竞争是⼈⼯智能竞赛中最关键，同时也是成本最⾼的环节，”Vahdat在内部会议上表示。Alphabet⾸席执⾏官桑达尔·⽪查伊与⾸席财务官阿纳特·阿什肯纳⻬也出席了该会议并回答了员⼯提问。 The presentation was delivered a week after Alphabetreportedbetter-than-expected third-quarter results and raised its capital expenditures forecast for the second time this year, toa range of $91 billion to $93 billion, followed by a“significant increase”in 2026.Hyperscaler peersMicrosoft,AmazonandMetaalso boosted their capex guidance,and the four companies now expect tocollectively spendmore than $380 billion this year. 这次汇报会召开前⼀周，Alphabet刚公布了超预期的第三季度财报，并第⼆次上调了今年资本⽀出预期⾄910亿⾄930亿美元区间，且2026年将迎来“显著增⻓”。同属超⼤规模云计算⼚商的微软、亚⻢逊和Meta也同步提⾼了资本⽀出指引。⽬前这四家企业预计今年合计投⼊将超过3800亿美元。微信公众号 404K微信公众号 404K Google’s “job is of course to build this infrastructure but it’s not to outspend the competition,necessarily,” Vahdat said. “We’re going to spend a lot,” he said, adding that the real goal isto provide infrastructure that is far “more reliable, more performant and more scalable thanwhat’s available anywhere else.” “我们的任务当然是建设基础设施，但未必⾮要靠压倒性投⼊取胜，”Vahdat强调，“虽然⽀出规模会很⼤”，但真正的⽬标是打造“⽐任何竞争对⼿都更可靠、性能更强且更具扩展性”的基础设施。 In addition to infrastructure build-outs, Vahdat said Google bolsters capacity with moreefficient models and through its custom silicon. Last week, Google announced the publiclaunch of its seventh generation Tensor Processing Unit calledIronwood, which thecompany says is nearly 30 times more power efficient than itsfirst Cloud TPU from 2018. 瓦赫达特表示，除了基础设施建设外，⾕歌还通过开发更⾼效的模型及采⽤⾃研定制芯⽚来提升算⼒。上周⾕歌宣布第七代张量处理单元“Ironwood”正式公开亮相，官⽅宣称其能效⽐2018年⾸代云端TPU提升近30倍。微信公众号 404K微信公众号 404K Vahdat said the company has a big advantage with DeepMind, which has research on whatAI models can look like in future years. 瓦赫达特指出，公司在DeepMind⽅⾯具有显著优势，该团队已开展针对未来数年AI模型形态的前瞻性研究。 LIVESTREAM networking for essentially the same cost and increasingly, the same power, the sameenergy level,” Vahdat said. “It won’t be easy but through collaboration and co-design, we’regoing to get there.” 我们需要实现以基本相同的成本和功耗⽔平——未来还要维持相同能耗⽔平，提供1000倍以上的性能、算⼒和存储⽹络能⼒，”瓦赫达特表示，”这绝⾮易事，但通过协作与协同设计，我们终将达成⽬标。 Sundar Pichai, chief executive officer of Alphabet Inc., during the Bloomberg Tech conference in San Francisco, California, US, onWednesday, June 4, 2025. 2025年6⽉4⽇周三，在美国加利福尼亚州旧⾦⼭举⾏的彭博科技⼤会上，Alphabet公司⾸席执⾏官桑达尔·⽪查伊发表讲话。 David Paul Morris | Bloomberg | Getty Images⼤卫·保罗·莫⾥斯|彭博社|美国财经电视台 Pichai told employees at the meeting that 2026 will be “intense,” citing AI competition andthe pressure to meet cloud and compute demand.微信公众号 404K微信公众号 404K ⽪查伊在会上向员⼯表示，由于⼈⼯智能领域的竞争以及满⾜云计算与算⼒需求的双重压⼒，2026年将是”紧张激烈”的⼀年。 He also answered a question about a potential AI bubble, a topic that’s gained resonanceacross Silicon Valley and Wall Street of late as investors have grown skeptical about LIVESTREAM whether the trillions of dollars in anticipated spend in the coming years is justified. 他还回应了关于⼈⼯智能潜在泡沫的问题。随着投资者对未来数年预期投⼊的数万亿美元是否合理产⽣疑虑，这个话题近期在硅⾕和华尔街引发⼴泛共鸣。 The employee