Bernstein’s warning on AI training slowdowns triggered a sector-wide sell-off, forcing investors to differentiate between rural-heavy infrastructure providers and metro-focused data center operators.
Bernstein named CoreWeave as the biggest loser from a potential AI training slowdown due to its rural capacity exposure, driving a 6.75% intraday drop.
Nebius shares fell 8% amid a broad sell-off in AI infrastructure names after Bernstein's warning.
Lumentum shares dropped 8% as networking equipment makers sold off on AI capex concerns.
Marvell Technology fell 8% as chipmakers tied to AI infrastructure were hit by the slowdown warning.
Arm Holdings declined 8% in the broad semiconductor sell-off triggered by AI deceleration fears.
Applied Digital dropped 6% as the AI infrastructure sell-off hit neocloud peers.
Ciena fell 7% as investors rotated out of AI infrastructure suppliers.
Coherent dropped 7% alongside other AI-driven networking equities.
AMD shares fell 6% as AI-exposed chipmakers sold off on concerns of a training investment pullback.
Intel dropped 6% as AI infrastructure spending fears weighed on the semiconductor sector.
Equinix was rated Outperform by Bernstein, praised for its 95% metro capacity that is safer for an inference shift.
Digital Realty received an Outperform rating from Bernstein due to its high metro market exposure, insulating it from a training slowdown.
DigitalOcean declined 5% in sympathy with the broader AI infrastructure downturn.
IREN lost 4% as AI-related stocks fell on training slowdown fears.
Arista Networks declined 5% as the AI sell-off spread to networking hardware names.
Broadcom slipped 4% in the AI stock rout, reflecting broad-based profit-taking.
Bernstein rated Csquare Outperform citing its 94% metro capacity as a buffer against declining training demand.
Wall Street’s unified AI trade fractured on Sept. 14 as Bernstein warned that a potential slowdown in AI model training disproportionately threatens companies with heavy rural capacity exposure. While CoreWeave shares slumped 6.75% on concerns over its non-metro footprint, firms like Equinix and Digital Realty gained favor for their proximity to high-value inference markets.
AI training is latency-insensitive and can occur in rural areas, but inference—the process of running AI products—requires proximity to users in metro markets to minimize latency.
Anthropic CEO Dario Amodei’s call to slow the pace of AI model development, combined with Bernstein’s analysis of geographic capacity risks, caused investors to reassess the sector.