This article first appeared on GuruFocus.
CoreWeave (NASDAQ:CRWV) could be among the biggest casualties if artificial-intelligence companies slow the pace of model training, according to Bernstein, putting fresh pressure on a stock already highly exposed to the AI infrastructure buildout. The firm argues CoreWeave's concentration in rural and lower-tier data-center markets makes it more vulnerable than metro-focused operators if demand shifts away from massive training workloads. Shares fell 8% Monday.
Warning! GuruFocus has detected 4 Warning Signs with CRWV.
Is CRWV fairly valued? Test your thesis with our free DCF calculator.
CoreWeave is part of the fast-growing neocloud segment supplying computing infrastructure for artificial-intelligence workloads. Its investment case is closely tied to continued spending on large-scale AI training and the expansion of data-center capacity.
Bernstein's concern follows calls from Anthropic CEO Dario Amodei to slow model development, while OpenAI CEO Sam Altman has said frontier AI should be paced rather than stopped.
From the infrastructure and neocloud perspective: a decline or slowing in training pulls demand away from rural data center locations, many of which have been purpose-built for this latency-insensitive training demand, Bernstein analyst Maidson Rezaei said.
Bernstein estimates 25% of CoreWeave's existing active U.S. power and about 74% of its contracted power sit in Tier 3 and Tier 4 markets. Existing backlog is largely protected by take-or-pay agreements, limiting near-term risk, but demand for contracted capacity that has not yet been sold could weaken if training growth slows.
By contrast, Equinix, Digital Realty and Csquare have more than 90% of capacity in major or minor metro areas, which Bernstein views as better positioned as AI workloads shift toward inference.
Bernstein rates CoreWeave Underperform with a $74 price target, making utilization of future power capacity the key metric to watch.
The risk is not necessarily CoreWeave's existing contracted backlog, but whether future rural capacity can find enough customers if model-training demand cools.
Continued hyperscaler spending and strong training growth would weaken Bernstein's bearish case. A sustained industry shift toward inference, however, could favor metro-based infrastructure and leave CoreWeave carrying more exposed capacity in markets built primarily for training workloads.