AI stocks tumble on safety warnings. How to proceed — including 2 names on our shopping list

AI data-center buildout stocks sold off sharply premarket after Anthropic CEO Dario Amodei called for slower AI-capability development, a view echoed by OpenAI's Sam Altman and xAI's Elon Musk. CoreWeave, Nebius and Vertiv fell more than 8%, while Eaton declined roughly 6%; Intel and Micron were each down about 6%. Altman also ruled out an OpenAI IPO in 2026, citing AI-safety concerns, intensifying investor worries that AI infrastructure demand could decelerate.

Analysis

The important transmission channel is not a near-term reduction in installed AI capacity, but a higher discount rate applied to the most duration-sensitive parts of the buildout. CRWV and NBIS require sustained external financing and aggressive utilization assumptions; a narrative-driven pause can widen funding spreads and impair their ability to convert contracted demand into incremental capacity. VRT and ETN have more defensible backlog and aftermarket exposure, but their multiples remain vulnerable if investors begin valuing orders on a shorter commitment horizon.

INTC and MU are not clean substitutes for data-center infrastructure exposure. MU's upside depends on HBM content, qualification cadence, and pricing discipline; a broad capex de-rating can hurt sentiment before it affects bit demand. INTC may outperform in a reversal because its valuation embeds less AI optionality, but that is a relative-value argument, not evidence that its execution risk has improved. A one-day selloff is insufficient to establish whether hyperscalers are revising 2026 procurement plans or whether leveraged AI infrastructure is simply being de-risked.

Over the next 1-3 months, the key catalyst is any evidence from MSFT, AMZN, GOOGL, META, and ORCL that committed data-center capex, power reservations, or GPU leases are changing. If none emerges, the highest-beta names can rebound sharply from forced selling; if capex guidance falls, the second-order damage will be greatest at power/cooling suppliers with elevated expectations and at compute lessors whose equity value is highly sensitive to debt availability. The contrarian view is that safety rhetoric may slow frontier-model training intensity while increasing demand for inference, governance, and enterprise deployment; that shifts spending composition rather than eliminating it.

Falsification for a tactical long is a confirmed hyperscaler capex reduction or evidence that backlog conversion is slipping, rather than merely weaker share prices. For CRWV/NBIS, monitor financing terms, collateral requirements, and customer concentration; a widening credit spread or incremental equity issuance would signal that the equity drawdown is fundamental. For VRT/ETN, watch orders and book-to-bill rather than headline AI commentary.

Key Decisions for Investors

• Do not buy the opening weakness in CRWV or NBIS. Place them on a 1-3 month watchlist pending evidence of stable financing capacity and unchanged customer commitments; avoid longs if either signals incremental equity issuance or materially higher funding costs.
• Express a relative-value rebound only after a failed breakdown/reclaim of the first-hour low: long INTC versus short SMH in equal beta for a 2-6 week horizon. The thesis is lower AI-expectation duration, not a fundamental turnaround; exit on adverse foundry/client guidance or if INTC underperforms SMH by 5%.
• Accumulate MU in tranches only if HBM pricing and 2026 supply allocation remain intact through the next industry checks. Target a 3-6 month holding period; cut the position if management indicates weaker HBM demand, qualification delays, or a reversal in DRAM pricing discipline.
• Prefer ETN over VRT for eventual re-entry after volatility normalizes, given broader electrical-distribution and utility exposure. Use a 1-3 month horizon and require order/backlog confirmation; avoid adding if data-center-related order growth decelerates materially at the next report.
• Use any hyperscaler capex-guide reduction as a trigger to short a basket of CRWV/NBIS/VRT rather than ETN. The likely downside is concentrated in financing- and multiple-sensitive exposures; cover if management commentary confirms only timing shifts with backlog and utilization intact.