As concerns grow that AI model development could slow, stocks tied to U.S. data centre expansion expectations are wobbling.

On Sept. 15, CNBC reported that Wall Street has begun rechecking whether the U.S. data centre investment boom, which has grown around Anthropic and OpenAI, can continue.

The unease was fully priced in on Sept. 15 after Dario Amodei, Anthropic's chief executive, proposed over the weekend slowing the pace of frontier-model development. AI infrastructure shares fell broadly and GE Vernova slid nearly 9 percent. Caterpillar dropped more than 4 percent and Vertiv fell nearly 8 percent. Oracle also declined nearly 4 percent.

The market reacted sharply because data centre expansion is directly tied to earnings assumptions across many industries. Over the past 18 months, Oracle has been reshaping itself into an AI beneficiary by investing billions of dollars in computing equipment and data centres while cutting staff in slower-growing units. GE Vernova, Caterpillar and Vertiv have focused on equipment investment needed to supply power to AI servers. New cloud companies such as Nebius and CoreWeave have also emerged as key suppliers providing hyperscalers with computing resources and access to AI infrastructure. Dell and Hewlett Packard Enterprise are tied to the same trend.

An anonymous technology investor said companies linked to AI buildouts could take a direct hit if expansion slows. The investor said that significant delays could become a cost for these companies.

Specific worries were also raised about Oracle. Rishi Jaluria, an equity analyst at RBC Capital Markets, pointed to a high likelihood that a slowdown in model development and training would be a headwind for Oracle's cloud infrastructure business. He also noted that this business has been the driver of Oracle's share price.

Amodei drew a line, saying the slowdown argument does not mean stopping AI development. He said adjusting the pace of development does not mean halting model training or technological progress. The market did not immediately take comfort from the remarks and focused on whether the speed and scale of data centre expansion could fall below prior expectations.

Financing conditions are also emerging as a burden. As pressure around data centres grows, the industry expects AI-related borrowing demand to surge over the next 6 weeks. The rise in nationwide backlash against data centres was also cited as a factor affecting financing conditions.

Hyperscalers are in fact heading back to bond markets. Amazon raised about 4.25 billion pounds, about $6 billion, last week. Alphabet previously secured about $10 billion by issuing euro-denominated bonds in May. An industry official said new borrowing deals coming this autumn could carry much higher interest rates than in prior issuance. Another official said fixed-income investors are demanding greater compensation.

That leaves two main points for the market to watch. One is whether demand for AI model development, centered on Anthropic and OpenAI, actually slows. The other is at what cost companies raise the funding needed for data centre expansion. Even if enthusiasm for AI infrastructure investment holds, higher borrowing costs could force another adjustment in the pace of data centre expansion and expectations for related companies' earnings.