-+ 0.00%
-+ 0.00%
-+ 0.00%

AI stocks have experienced a “one-size-fits-all” sell-off! Bernstein named 6 major overrun opportunities: Dell (DELL.US), SanDisk (SNDK.US), etc. received “outperforming the market” ratings

Zhitongcaijing·11/16/2026 15:01:36
Listen to the news

The Zhitong Finance App learned that with the recent sharp sell-off in artificial intelligence (AI) related stocks, Wall Street has begun to re-examine the risks and rewards of investing in AI infrastructure. Bernstein analyst Mark Newman and his team pointed out in a recently released research report that market concerns about delays in data center construction, political resistance, and financing risks for emerging AI cloud service providers have caused the valuation of some AI-related stocks to fall excessively, yet the current evidence is insufficient to show that demand for AI infrastructure is deteriorating across the board.

Bernstein believes that investors are imposing similar valuation discounts on companies involved in different segments of the AI industry, while ignoring the significant differences in business structure, customer composition, and degree of dependence on large-scale data center construction.

The bank's quantitative analysis shows that ultra-micro computers (SMCI.US) and SanDisk (SNDK.US) are the two companies whose valuations have dropped the most during the recent sell-off, and the valuations of Dell Technologies (DELL.US), Seagate (STX.US), and Western Digital (WDC.US) have also surpassed the decline to a certain extent. In contrast, although the stock price of AI cloud service provider Coreweave (CRWV.US) has been drastically adjusted, its valuation may not fully reflect the potential risks.

In terms of investment opportunities, Bernstein is particularly optimistic about Dell Technologies, Seagate, Western Digital, and SanDisk, as well as data center operators Equinix (EQIX.US) and Digital Realty (DLR.US). The six companies mentioned above all received the bank's “outperforming the market” rating.

Bernstein pointed out that an important issue with the current market sell-off is that investors confuse the extent to which companies participate in the AI industry with their degree of dependence on data center construction for large cloud service providers and emerging AI cloud service providers.

In reality, the risks faced by different types of AI requirements are not the same. Infrastructure construction for large cloud service providers and emerging AI cloud service providers still focuses on model training, which is directly affected by power supply, regulatory approval, construction progress, and customer deployment preparations, so it is most vulnerable to delays in data center projects.

In contrast, enterprise-level AI requirements are more focused on inference and AI agent applications. It can be deployed by enterprises building their own servers, private clouds, storage and network equipment, etc., and the dependence on building large-scale data centers is relatively low.

In addition, consumer-grade AI can also be commercialized through terminal devices and ecosystems. For example, Apple (AAPL.US) is expected to benefit from end-side AI, commercialization of existing user groups, and future equipment update requirements without directly bearing the risks faced by large-scale data center construction.

Bernstein believes that the market's concerns that the US midterm election results may hinder data center construction are also exaggerated to a certain extent. The bank's bottom-up project analysis shows that after the midterm elections, the US may face stricter data center construction restrictions. The planned data center power capacity will increase from about 249 gigawatts (GW) to 306 gigawatts. However, the capacity of the project, which has a high degree of reliability, only increased from 78 gigawatts to 105 gigawatts, adding about 27 gigawatts.

Of this 27 gigawatts of additional capacity, about 10 gigawatts have already entered the construction phase and are expected to remain eligible for approval. The remaining 17 gigawatts may be affected by stricter regulatory measures.

Bernstein pointed out that even if all of these 17 gigawatt projects cannot be implemented, it is only equivalent to about 9% of the capacity of data center projects with a high degree of reliability in the US. In a more realistic scenario, the bank expects that the new regulatory restrictions brought about by the midterm elections may ultimately only affect about 2% to 4% of the US's relevant project capacity.

Furthermore, local governments are more likely to require data center developers to bear additional grid infrastructure construction costs and strengthen the protection of local communities rather than completely blocking data center construction. As a result, related political risks may be reflected more in rising project construction costs and delayed delivery times rather than large-scale loss of computing power requirements.

The departure of Nvidia (NVDA.US) CEO Hwang In-hoon has brought another layer of uncertainty to the AI infrastructure industry. Bernstein pointed out that in recent years, Nvidia has become an important financing and credit supporter to support the expansion of emerging AI cloud service providers. According to reports, the total amount of various promises and guarantees provided by the company to support customer data center construction reached 164.5 billion US dollars, of which about 40 billion US dollars was used to support AI cloud computing partners, and these partners are mainly emerging AI cloud service providers.

However, Bernstein believes that the relevant agreement that Nvidia has signed is a contractual obligation and is not expected to be significantly affected by management changes. What is really worth paying attention to is the additional financing and guarantee agreements that have not yet been signed.

If Nvidia adopts a more cautious financing support policy in the future, the capital costs for emerging AI cloud service providers to build a new round of data centers may rise. However, the bank believes that this is mainly a long-term risk, and it is concentrated on enterprises that are most dependent on large-scale AI data center construction.

For hardware vendors, Bernstein believes that the current shortage of parts supply may pose a more direct business limitation than delays in data center construction.

Currently, demand for AI infrastructure still exceeds available data center capacity and hardware supply capacity, and GPUs, CPUs, memory, storage, optical communication, and power-related components all face varying degrees of supply constraints. As a result, customers placed orders in advance to ensure that scarce equipment was available.

While overall demand still exceeds supply, further delays in individual data center projects may affect hardware delivery times or supply distribution among customers rather than directly causing related equipment requirements to disappear. Companies with a diverse customer base, scarce products, and flexible supply chains can be expected to mitigate the impact of project delays by adjusting delivery sequences or supplying to other customers.

Using ultra-microcomputers as an example, Bernstein pointed out that in the latest quarter, the company confirmed about $2.3 billion in revenue that had previously been postponed due to customer data centers not being ready, indicating that part of the demand for delayed delivery can still eventually be converted into actual revenue.

In order to further assess the risks of different companies, Bernstein conducted a quantitative analysis of the 12 companies it covered, focusing on measuring the share of the data center business of large cloud service providers and emerging AI cloud service providers in the growth of each company's operating profit and operating profit, rather than simply comparing the size of AI-related revenue.

The analysis results show that the higher the data center business contributes to the growth of operating profit, the more obvious the valuation of related companies usually falls. According to the bank's regression analysis, for every 10 percentage point increase in the contribution of the data center business to operating profit growth, the decline in enterprise forward valuation compared to the average of the past three years increased by about 7.2 percentage points.

Despite the small size of the study sample, this relationship still explains about 64% of the difference in valuation changes, indicating that data center business risk exposure has become an important influencing factor in the recent valuation performance of related stocks.

Among them, the valuation adjustments for ultra-microcomputers were the most drastic. The company's projected price-earnings ratio for the next year has dropped to about 4.2 times, about 72% lower than the average of 15.2 times over the past three years, while the Bernstein model's reasonable valuation based on its data center business risk exposure has dropped by about 34%.

This means that the actual valuation decline of ultra-microcomputers was about 38 percentage points higher than the model's estimation results, and the bank covered the company with the most obvious degree of overvaluation decline.

However, Bernstein remains relatively cautious about ultra-microcomputers. The bank expects that businesses related to large cloud service providers and emerging AI cloud service providers will contribute about 68.4% of ultra-microcomputer revenue, 47.7% operating profit, and 58.9% operating profit growth in fiscal year 2027.

In addition, factors such as high customer concentration, uncertainty about backlog order conversion, business execution risk, and potential equity dilution also mean that the company should bear higher risk discounts.

Bernstein currently gives the ultra-micro computer a “in sync with the market” rating, and the target price is 42 US dollars. The bank believes that if the company's business execution stabilizes and demand for AI infrastructure remains strong, its undervaluation will bring greater potential for stock price recovery, but the investment risk is still significantly higher than Dell Technology.

SanDisk is another company that Bernstein believes its valuation has been severely overcompressed. The company's projected price-earnings ratio for the next year has dropped to about 3.7 times, about 65% lower than the historical average of 10.5 times that can be obtained since its independent listing, while the valuation drop estimated by the Bernstein model is only about 36%.

In addition to delays in data center construction, the market is also concerned that the introduction of additional NAND flash memory production capacity from 2028 to 2029 may cause product prices and corporate profits to quickly return to normal levels. However, Bernstein believes that the market may be underestimating the profit protection effect of SanDisk's long-term customer agreements and the consumption of NAND production capacity by new AI storage applications.

SanDisk's new business model agreement already covers approximately 50% of the supply of flash memory bits in FY2027 and about two-thirds of FY2028's supply. These agreements include customer purchase volume commitments, pricing mechanisms, and financial guarantee clauses, which are expected to reduce cyclical fluctuations in the company's future profits.

At the same time, the development of enterprise-level AI inference and AI agent applications will drive enterprises to generate, process, and store more data, creating new requirements for edge storage services other than SanDisk's data center business.

Bernstein also pointed out that new applications such as single-layer cell (SLC) NAND and high-bandwidth flash memory (HBF) may consume far more wafer production capacity than traditional NAND products, thereby offsetting the impact of future additional production capacity on the industry's supply and demand pattern.

The bank estimates that if high-bandwidth flash memory accounts for 5% of NAND bit requirements in the future, and its wafer consumption intensity is about 6.5 times that of traditional three-layer unit (TLC) NAND, the industry may need an additional 28% increase in wafer production capacity to meet related demand. This will help prolong the NAND industry's cycle of tight supply and price increases.

Bernstein has now given SanDisk an “outperforming the market” rating, and the target price is as high as $3,000. Based on the stock's closing price of $937.46 on November 12, the potential upside is about 220%.

In the field of AI infrastructure hardware, Dell technology is Bernstein's preferred target. The bank believes that Dell has a relatively limited profit risk exposure for large AI data centers, strong enterprise-level AI business growth, high-quality backlog of orders, and attractive valuations.

Bernstein predicts that the AI server business related to emerging AI cloud service providers will contribute about 26.9% of Dell's revenue in fiscal year 2027, but due to the relatively low profit margin of the related business, its operating profit contribution is only about 11.7%, contributing about 16.1% to the increase in operating profit.

Most of Dell's profits still come from traditional server, enterprise infrastructure, and storage businesses, which are relatively limited by delays in the construction of large data centers. Additionally, Dell's traditional server orders increased 122% year over year in the second quarter of fiscal year 2027, reflecting the modernization of enterprise infrastructure, increased market share, and emerging AI inference requirements that are creating new growth drivers for its traditional business.

Dell currently has a backlog of approximately $95 billion in AI server orders, and related orders have been assessed on customer credit, financing and technical design. Company management said that planned server deliveries match customer data center readiness by about 80% to 90%, and that most backlog orders cannot be cancelled.

Bernstein believes that Dell has more than 6,500 AI customers, covering emerging cloud service providers, sovereign AI and enterprise customers. Its diverse customer base and supply chain management capabilities will help mitigate the impact of delays in individual large-scale projects.

Currently, Dell's projected price-earnings ratio for the next year is about 12.6 times, which is 14 times lower than the average for the past three years by about 9.6%. Bernstein believes that the market pays too much attention to the data center construction risks of emerging AI cloud service providers, yet underestimates Dell's enterprise-level AI growth opportunities, diversified profit sources, and business execution capabilities, and suggests investors take advantage of recent weak stock prices to increase allocation.

The bank gave Dell an “outperforming the market” rating. The target price was $650, which is about 75% higher than the closing price of $371.17 on November 12.

In contrast, although HPE.US is less directly dependent on large-scale AI data center construction, its valuation already fully reflects this defensive advantage. Bernstein believes that compared to HPE Technology, Dell has more attractive fundamentals and room for valuation growth.

In the field of mechanical hard drives, Bernstein continues to be optimistic about Seagate Technology and Western Digital. He believes that the hard drive price increase cycle is still in the early stages, and the price increase is expected to accelerate further in the future.

Strong demand, limited industry production capacity, and careful expansion of production by suppliers will continue to support the rise in mechanical hard disk prices. At the same time, the popularity of enterprise-level AI inference and AI agent applications is expected to drive enterprises to generate and store more data, thereby expanding the demand for mechanical hard drives and gradually reducing the industry's dependence on large cloud service providers.

Bernstein is particularly optimistic about Seagate's leading edge in heat-assisted magnetic recording (HAMR) technology. This technology can increase the storage density of hard drives, so that companies can provide more storage capacity without increasing hard drive production in the same proportion, thereby strengthening their cost advantage per unit of storage.

Currently, Seagate Technology's price-earnings ratio for the next year is about 11 times, which is about 46% lower than the average of the past three years; Western Digital's corresponding valuation is about 9.9 times, which is about 55% lower than the historical average.

Bernstein believes that although the two companies are currently highly dependent on large-scale data center demand, the mechanical hard disk industry is still in an upward price cycle. Coupled with the continuous expansion of demand for enterprise-grade AI storage, both companies have attractive risk rewards. The bank gave Seagate Technology and Western Digital a “outperforming the market” rating. The target prices were 1100 US dollars and 600 US dollars, respectively, which were about 102% and 93% higher than the closing price on November 12, respectively.

Bernstein is also optimistic about Equinix and Digital Realty, and believes that the market should not equate traditional data center operators with emerging AI cloud service providers.

Equinix's business mainly focuses on enterprise-level data center hosting and interconnection services, and is less directly dependent on large-scale AI model training projects. The company's standardized monthly recurring revenue for the third quarter achieved double-digit growth for the fourth consecutive quarter, and raised its full-year results guidance for the third consecutive quarter.

Digital Realty, on the other hand, has more data center businesses for large cloud service providers, but its customer quality is high. The current cloud computing business portfolio has a large number of rental contracts for 15 to 20 years, and its direct exposure to emerging AI cloud service providers is close to zero. The company set a record leasing business in the third quarter and raised its full-year core operating capital (FFO) guidance.

Bernstein gave Equinix and Digital Realty “outperform the market”. The target prices were $1,270 and $226, respectively.

In contrast, Bernstein believes that CoreWeave still faces the risk of further valuation cuts. The company's business is almost entirely dependent on AI data centers, and needs to quickly convert signed computing power contracts into actual operating capacity and achieve profits after deducting depreciation and financing costs. Although the ratio of its forward corporate value to profit before interest and tax has dropped by about 61% from its previous high, Bernstein's model shows that there may still be room for further compression in its valuation. The bank gave CoreWeave a rating of “outperforming the market”, and the target price was $54.

Furthermore, although Apple played a relatively safe-haven role in the recent AI stock sell-off, its expected price-earnings ratio for the next year has reached 37.3 times, which is about 24% higher than the average for the past three years. Considering rising memory costs, pressure on gross margin, and slowing service business growth, Bernstein believes that once risk aversion in the market subsides, Apple's stock price may also face pressure in the short term.