AI Slowdown Fears Hit Data Center Power and Construction Stocks Hard
nashnova research
Three AI leaders made a rare joint call to slow frontier-model development, triggering 6%–10% drops across data-center power and construction stocks Monday — the market is repricing how fast demand arrives, not whether it exists.
Who called for a slowdown, and what did they actually say?
Anthropic CEO Dario Amodei posted over the weekend urging the industry to slow the pace of frontier AI capability gains; OpenAI CEO Sam Altman and Elon Musk publicly backed the call.
This means → the heads of three leading AI firms simultaneously signaled safety concerns, and the market read it as "the industry may throttle itself."
Altman later clarified: frontier development should proceed at a measured pace, not stop — the message was "slow down," not "shut down."
Why did power-equipment and construction stocks fall harder than chips?
Power-equipment names led losses: nVent Electric fell ~10%, GE Vernova 8.5%, Vertiv 7.9%, Eaton 7%, Modine Manufacturing and Powell Industries each down over 6%.
Data-center construction stocks fell in lockstep: Sterling Infrastructure −8.1%, Dycom Industries −7%, Comfort Systems USA −6.2%, Quanta Services −5.1%, Emcor Group −4.2%, Caterpillar ~−4%.
In plain terms = these companies sell transformers, cooling systems, generators, and construction services — if the pace of new data-center builds slows, their entire order timeline shifts back, not just chipmakers.
Why did high valuations amplify the sell-off?
These stocks had sharply outperformed on the AI build-out thesis; valuations were already stretched: Quanta Services traded above 70× earnings, Modine ~66×, Vertiv over 50×.
This means → the market had priced in not just strong demand, but demand arriving fast and being sustained — any hint of a slower pace hits the premium first.
GE Vernova faced an additional company-specific headwind: GLJ Research initiated coverage that day with a sell rating and a $470 price target, questioning valuation and earnings expectations.
Does a slowdown call equal cancelled orders?
No. As of now, no company has announced cancelling data-center builds, cutting capex budgets, or pulling equipment orders.
Inference — the process of running a trained model to answer questions and handle tasks — may see continued growth in compute demand even if frontier-model iteration slows; safety testing itself consumes significant compute.
This reflects a broader constraint: U.S.–China tech competition means American firms face pressure to maintain their lead, and governments increasingly treat AI infrastructure as a strategic asset — leaving limited room for the industry to slam the brakes.
How hard were Asian and chip stocks hit?
Asian markets: SoftBank fell 10.7%, SK Hynix −6.4%, Samsung Electronics −4.1%, Kioxia −6.4%.
U.S. chip names: Nvidia dropped over 2.47% in pre-market; AMD, Intel, and Micron fell roughly 5% at one point.
Bernstein analyst Stacy Rasgon issued a report shortly after: AI semiconductor demand remains intact — Amodei was talking about slowing from "extremely fast" to "still quite fast," not stopping training; AI demand is increasingly inference-driven, and current compute capacity already falls far short of inference needs for existing models.
What is the real signal to watch next?
This sell-off repriced expectations, not orders — the market is recalibrating "how fast demand materializes," not "whether demand exists."
Put simply = the stocks fell on "it's coming a bit slower," not "it's not coming."
The hard indicator: whether AI companies' capex guidance in the next earnings season sees a material downward revision — that would be the real order-level signal.
市场有风险，内容仅供研究参考，不构成投资建议。