GE Vernova fell almost 9 percent on Monday, Vertiv dropped close to 8 percent, Caterpillar sank more than 4 percent and Oracle slipped almost 4 percent, according to CNBC. The trigger was a weekend proposal from Anthropic CEO Dario Amodei urging a slower pace of frontier model development, a call Amodei clarified does not mean “halting model training or technical progress.”
The selloff reveals how narrow the AI infrastructure trade has become. CNBC reported that the entire U.S. data center boom traces back largely to two customers: Anthropic and OpenAI. Every layer of the supply chain built around their demand, from energy equipment makers to server manufacturers to the neoclouds renting out compute, now trades on an assumption that model training keeps accelerating without pause.
Oracle sits at the center of that exposure. The company restructured itself over the past 18 months around the AI opportunity, pouring billions into data centers and compute while laying off people in its slower-growing units, per CNBC. RBC Capital Markets equity analyst Rishi Jaluria told the outlet that slower model training would likely weigh on the cloud infrastructure unit that has powered Oracle’s recent stock gains. That single-narrative dependence is precisely what Monday’s price action punished.
Industrial suppliers face the same concentration risk one step removed. GE Vernova, Caterpillar and Vertiv built out capacity to power AI servers, while Nebius and CoreWeave positioned themselves as compute intermediaries for hyperscalers. Dell and Hewlett Packard Enterprise round out a group whose earnings trajectories now hinge on chip and AI system demand staying high, according to CNBC’s reporting. One tech investor, granted anonymity by CNBC to speak candidly, put it bluntly: “Any significant delay could cost them.”
The financing side of the buildout adds urgency to the timing question. CNBC reported that two people familiar with the matter, speaking anonymously because the discussions are private, expect a rush to secure AI-related debt over the next six weeks as data centers face mounting local opposition. Amazon went back to borrowing last week, taking on roughly £4.25 billion in new debt (close to $6 billion), and Alphabet tested European investors in May with a $10 billion euro bond sale. A source told CNBC that debt deals priced this fall will carry meaningfully higher rates than earlier issuances, because, in that person’s words, “fixed income investors are demanding more reward.”
That pricing shift is the tell. Debt investors are not waiting to see whether Amodei’s proposal becomes policy; they are already charging a premium against the possibility that hyperscaler capex growth decelerates before the bonds mature. Equity markets moved on Monday’s headline, but credit markets are pricing a structural risk that predates it: years of infrastructure commitments made on the assumption that two labs’ training budgets only grow.
Amodei’s own framing, that a slowdown does not mean stopping technical progress, has not stopped the market from treating it as a demand shock. The gap between what Amodei proposed and how GE Vernova and Vertiv traded on Monday says more about balance-sheet fragility across the AI supply chain than about the merits of his proposal. Operators evaluating exposure to data center suppliers over the next quarter should watch whether the coming wave of AI debt issuances prices in a genuine slowdown premium or merely reflects one bad trading day.
Reporting by Seema Mody for CNBC, published September 15, 2026.
