Investors Reassess AI Infrastructure Risk as Calls for Slower Model Progress Emerge

Calls to moderate the pace of progress in the most capable AI models have introduced a new risk factor for investors assessing the scale and durability of AI infrastructure spending.

BofA Global Research estimates that Microsoft, Alphabet, Amazon, Meta and Oracle will make combined capital expenditures of about $795 billion in 2026 and $1.08 trillion in 2027. These are forecasts for aggregate company CAPEX, rather than figures specifically allocated to data centers, AI infrastructure, power systems or equipment.

The market concern is that a slower trajectory for advanced AI models could eventually translate into lower equipment orders or reduced data-center construction activity. However, the source reported no signs of mass cancellations of data-center projects or infrastructure equipment orders at the time of publication.

The immediate exposure is likely to be greater for chipmakers and suppliers of servers and other data-center infrastructure than for hyperscalers. Their valuations and demand outlooks are more directly tied to continued rapid expansion in AI-related capital spending.

This should be viewed as an increase in market uncertainty rather than evidence of a confirmed contraction in physical data-center capacity deployment. Some investors also argue that clearer AI safety rules could reduce long-term investment risk for data-center infrastructure rather than halt development. Until specific changes in capital programs, equipment orders or construction schedules emerge, the effect on delivered capacity remains unconfirmed.