Bernstein: Indiscriminate AI Stock Sell-Off Creates Oversold Opportunities, Names Six Outperform Picks
nashnova research
Bernstein's quant analysis shows the market's blanket AI sell-off has pushed select stocks well past fair-value discounts, with six companies rated outperform — the core logic being that different data-center exposure deserves different pricing, not one universal haircut.
What is the market afraid of, and why does Bernstein say it overshot?
The market fears AI infrastructure demand is deteriorating across the board, so it has applied a similar discount to every company in the AI supply chain — regardless of how much each actually depends on large-scale data centers.
Bernstein ran a regression across 12 covered names: for every 10-percentage-point increase in data-center contribution to operating-profit growth, forward valuations dropped about 7.2 percentage points relative to three-year averages. The model explains roughly 64% of the variation.
This means → the panic follows a pattern, but many companies got punished by association — their actual declines far exceed what the model says is justified.
In plain terms = Bernstein is saying the market didn't bother to check who really lives on data-center revenue and who only has a toe in.
Which two stocks are the most oversold?
SanDisk (SNDK) trades at a forward P/E of roughly 3.7×, about 65% below its post-spin historical average. Bernstein's model implies a fair discount of only about 36% — the gap is nearly 30 percentage points.
Bernstein argues the market underestimates SanDisk's long-term agreement protection: contracts already lock in roughly 50% of flash-bit supply for FY2027 and about two-thirds for FY2028, with committed purchase volumes and pricing safeguards.
Super Micro (SMCI) is even more extreme: forward P/E of about 4.2×, roughly 72% below its three-year average versus a model-implied fair discount of about 34% — an overshoot of roughly 38 points. Yet Bernstein is more cautious here, assigning only a market-perform rating — customer concentration, order-conversion uncertainty, and potential equity dilution add extra risk.
This means → both are oversold, but SanDisk gets an outperform while Super Micro gets only market-perform — the difference is whose revenue is better protected by contracts.
Who is Bernstein's top pick, and why?
Dell Technologies (DELL) is Bernstein's top pick. Emerging AI-cloud-related business is forecast at about 26.9% of FY2027 revenue, but because margins are thinner, it contributes only about 11.7% of operating profit — the bulk of earnings still comes from traditional servers, enterprise infrastructure, and storage.
In plain terms = Dell is not a "pure AI play." Its profit base sits on traditional business; AI is the upside kicker.
Traditional server orders grew 122% year-over-year. AI server backlog stands at roughly $95 billion. Management says 80%–90% of planned deliveries match customer data-center readiness, and most of the backlog is non-cancellable.
Forward P/E sits at about 12.6×, only about 9.6% below the three-year average. Bernstein sets a $650 target — roughly 75% above the Nov 12 close.
What is the logic for hard drives and data-center operators?
Seagate (STX) and Western Digital (WDC) both rated outperform. Bernstein judges the HDD price-upcycle is still early-stage: strong demand + limited capacity + cautious supplier expansion all support pricing.
Seagate leads in HAMR — heat-assisted magnetic recording, a technology that boosts storage capacity without a proportional increase in disk count. The two trade at forward P/Es roughly 46% and 55% below historical averages; targets are $1,100 and $600, respectively.
Equinix (EQIX) and Digital Realty (DLR) are data-center operators, but Bernstein stresses they are not the same as emerging AI-cloud providers. Equinix's core business is enterprise colocation and interconnection — monthly recurring revenue posted double-digit sequential growth for a fourth straight quarter. Digital Realty's cloud-computing leases run 15 to 20 years, with direct exposure to emerging AI-cloud providers near zero.
Who does Bernstein flag as still at risk?
CoreWeave (CRWV). Even though its forward EV/EBIT has fallen about 61% from the peak, Bernstein sees room for further compression.
This reflects CoreWeave's core risk: the business depends almost entirely on AI data centers. It must convert signed compute contracts into live capacity quickly — and turn a profit after depreciation and financing costs. If any link breaks, the valuation thesis loosens.
In plain terms = the six outperform names share one trait — none is fully tied to the AI-data-center line alone. CoreWeave is the opposite: its fate is lashed almost entirely to that single line.
What will ultimately validate this framework?
Bernstein's answer is structural divergence, not systemic deterioration — companies with different data-center exposure should not bear the same discount.
Whether the framework holds, though, hinges on upcoming quarterly order conversion and earnings delivery at Dell and peers.
This means → if Dell's backlog converts smoothly into revenue and SanDisk's long-term agreements do stabilize earnings, the "divergence" thesis stands; if not, the market's "one-size-fits-all" haircut may turn out to have been right.
市场有风险,内容仅供研究参考,不构成投资建议。
