Power Scarcity or Compute Oversupply in the AI Selloff
Goldman Sachs' Rich Privorotsky says investors are using two competing bearish cases to sell AI infrastructure: insufficient power and data-center capacity to support planned spending, or excess compute that undermines leasing economics. Demand remains strong, with Azure, AWS and Google Cloud reporting quarterly growth of 48%, 39% and 103%, respectively. But Goldman colleague Lee Coppersmith estimates hyperscaler capital spending will reach $1.2 trillion in 2027, while aggregate Q3 free cash flow has turned negative. Intel's 29% single-day plunge followed unexpectedly severe PC-demand guidance, which Privorotsky said was more related to consumer PCs, margins and manufacturing than to AI demand itself.
The report highlights a tension between robust AI infrastructure demand and uncertain returns on exceptionally large capital spending. It also says fund managers have reduced exposure while total leverage remained high; Morgan Stanley separately found hedge funds underweight AI-related holdings for the first time since early 2023.
高盛点破AI市场两难:电力短缺与算力过剩两套逻辑虽然互斥,却都成为了抛售AI交易的理由。云厂商需求仍强,资本开支却剑指1.2万亿美元,现金流转负叠加高杠杆;年末或出现轧空反弹,但反弹亦或是卖点。
美股超大规模云厂商刚刚交出连续第六个季度加速增长的成绩单,英特尔却经历了半个世纪来最惨烈的单日暴跌(-29%)。夹在中间的对冲基金,整个10月都在抛售一切带“AI”字样的标的。
市场究竟在怕什么?高盛EMEA Delta One交易主管Rich Privorotsky在周一早间的周报中表示,没人真正知道该担心什么。他写道:
关于AI的大辩论正在变得有些循环。
一半市场认为,我们根本不可能建造足够多的电力和数据中心来支撑2027年的资本开支计划,因此'铲子和镐头'类股的增长必然见顶,估值倍数将被压缩。
另一半市场则认为,我们即将大规模过度生产算力,这将摧毁租赁经济学,同样要求更低的估值倍数。
#Goldman Sachs#Intel#Amazon#Alphabet#Microsoft#Azure#AWS#Google Cloud#Morgan Stanley#UBS#Kioxia