Written by: Rita
Crowded stocks don’t always win. The cumulative return differential between the most crowded quintile and the least crowded quintile was +51.0% in the first half of the year but turned negative at -30.2% in the second half. Bernstein, in its report on October 28, 2026, used this sharp reversal as a lens to observe capital flows.
Internal changes within software companies are more nuanced. Bernstein believes that overall industry congestion remains significantly below long-term historical levels. Capital is beginning to flow back to companies with stable performance or renewed growth momentum. Low congestion creates an opportunity to reassess firms, evaluating evidence of growth and AI monetization to determine which companies are best positioned to attract capital back.
The software as a whole remains in a low-congestion zone.
Bernstein's concept of crowding primarily refers to the excessive concentration of active capital in the same stocks. Similar positions and judgments can cause these stocks to face synchronized selling pressure when market conditions shift. The report uses this metric to analyze capital structure and identify divergent expectations among companies.
Compared to July, the congestion level in the software sector rebounded during the third quarter. The report attributes part of this change to a rebound in performance by application software, with stable or accelerating revenue prompting the market to reassess the urgency of AI’s impact. Even with this rebound, software remains at relatively low levels compared to its own historical averages over the past 25 years.
This assessment does not contradict the overall overcrowding in the technology sector. The report notes that semiconductors and hardware continue to attract high levels of interest, while software faces concerns over generative AI replacing applications, compressing market share, and competing for corporate budgets. Viewing technology stocks as a whole obscures the divergence in capital flows between different business models.
Bernstein thus saw an opportunity to reconsider software companies. Low crowding allowed some enterprise platforms to shed overheated positions, while stable or improving business fundamentals began attracting capital back. The report ties these two threads together, focusing on identifying gaps between operational performance and market sentiment.
Funds are concentrated on the growth side of usage.
As of October 19, Snowflake had a standardized relative congestion score of 2.13 among the reported companies. Palo Alto Networks scored 1.86, and Datadog scored 1.80. In contrast, Oracle scored -0.37, Adobe scored -0.70, and HubSpot scored -1.55. These scores visually illustrate the differing preferences of capital between databases, infrastructure, and application software.
Snowflake’s changes have been particularly notable. The report highlights that increased usage of its core platform, combined with revenue growth from AI products, has accelerated the company’s business over the past two quarters. Its usage-based pricing model makes it easier for increased cloud workloads to translate into revenue, boosting investor enthusiasm for allocation.
However, Bernstein did not raise its rating in line with the growing popularity of the position. Regarding Snowflake, the report still expresses concerns about the maturity of the data warehousing market, intensifying competition, and the long-term competitiveness of AI tools and data federation solutions. Rising popularity may reflect improved growth, but it could also mean that optimistic expectations for the future have already been priced in.
Datadog presented similar divergence. Bernstein is optimistic about usage growth from other AI-native customers, while acknowledging pressures such as customer churn among large clients and slowing non-AI cloud workloads. The report expects growth in consumption from other AI-native customers to offset these pressures and drive continued revenue acceleration. Growth in AI workloads remains a key rationale for institutions supporting this business model.
The app's recovery must be proven by its performance.
The rebound of ServiceNow and Atlassian has outlined specific conditions for capital reallocation into application software. Bernstein noted that revenue growth has accelerated again for these companies, and their position as enterprise platforms provides more solid support than AI narratives alone. Market concerns about application software have eased, but the recovery has primarily focused on companies that have delivered tangible operational evidence.
The report expresses skepticism toward the argument that enterprises can replace their custom-built complex applications. Using enterprise resource planning (ERP) systems as an example, large enterprises bear significant risks related to processes, data, and operations. The advancement of generative AI tools does not automatically eliminate these migration and replacement costs. Bernstein therefore sees enduring value in certain core enterprise platforms.
Risks still need to be segmented by product and pricing model. The report notes that automation in areas such as customer service may reduce demand for software charged on a per-seat basis. It remains to be seen whether vendors can offset these losses with intelligent agent-based pricing. Single-point tools lacking core data and key business semantics are also more vulnerable to substitution pressure.
Bernstein is optimistic about Oracle’s growth in cloud infrastructure, enterprise applications, and databases. However, funding constraints and execution challenges in data center construction continue to weigh on market confidence. Regarding HubSpot, the report remains cautious due to weaker growth expectations, declining management confidence, and a lack of catalysts. Both companies are not overvalued, but institutional assessments differ significantly.
Microsoft’s case adds to this framework. The report suggests that concerns over capital expenditure returns and Copilot adoption rates have dampened investor sentiment. Recent business performance provides a basis for reassessment. Whether capital flows back depends on aligning growth, profitability, and returns on AI investments on the same financial ledger.
Crowd indicators should be combined with valuation.
In the risk model, Bernstein views crowding as risk arising from shared positions. When investors reduce risk simultaneously, stocks under the same narrative face concentrated selling pressure. Positions originally spread across different companies may exhibit strong correlation during the exit phase.
In the 10% of months with the worst performance for crowded trades, the most crowded quintile of stocks underperformed the market average by 3.0%, while the least crowded quintile outperformed by an average of 3.3%. Bernstein concludes that including stocks with varying levels of crowding in a portfolio helps diversify common trading risk. Historically, less popular assets have provided protection when popular positions faced concentrated selling.
CrowdStrike highlights the importance of existing holdings. Bernstein notes that institutional ownership is already high, limiting further buying potential, and the actual level of crowding may be higher than model scores suggest. The report offers a positive assessment of the company’s business competitiveness but views its valuation as elevated, making it difficult to meet market expectations for accelerated growth.
Following the stock selection logic in the report, opportunities in the software sector arise from the narrowing of divergences. Some platform companies are alleviating substitution concerns through steady growth, others are realizing revenue through AI demand, while some low-conviction companies still need to first address growth and execution challenges. The areas receiving the strongest capital inflows remain companies whose operational improvements can ease market concerns.
Software fund flows have already emerged, but repairs have not yet been widely implemented. Bernstein’s report provides an entry point for observing portfolio structure, but ultimately, business performance must answer: Can growth be sustained? How much new revenue is AI generating? And how much improvement has already been priced into the current valuation?
Disclaimer
This article is a compilation and interpretation by Chaoxiang Research of a third-party brokerage research report (Bernstein, October 28, 2026), combined with publicly available market information. The ratings, price targets, earnings forecasts, and related judgments cited herein reflect the views of the brokerage’s analysts and represent the position of their respective institution, not the views of Chaoxiang Research, nor do they constitute any investment advice.
The market carries risks; make decisions independently. This article should not be used as a basis for buying or selling any securities.