Bernstein said crowding across the software sector remains clearly below its long-term historical levels.
In its Oct. 28, 2026 report, the firm used the sharp reversal in crowded trades as a way to read capital flows. In the first half of the year, the cumulative return gap between the most crowded quintile and the least crowded quintile was positive 51.0%. In the second half, that spread turned negative 30.2%. For Bernstein, that swing showed that crowded stocks do not keep winning by default.
Software is still in a relatively low-crowding zone
Bernstein defines crowding mainly as an excessive concentration of active capital in the same stocks. When investors hold similar positions and make similar calls, those names can face shared selling pressure once the market turns. The report uses that measure to track positioning and to identify where expectations diverge across companies.
Compared with July, crowding in software rose in the third quarter. Bernstein linked part of that move to a rebound in application software earnings. Stable or reaccelerating revenue led the market to reassess how urgent the AI disruption risk really is. Even after that repair, software still sits at a relatively low level versus its own history over the past 25 years.
That view does not conflict with the broader technology sector looking crowded. The report said semiconductors and hardware still carry heavier positioning, while software continues to trade under concerns about generative AI replacing applications, reducing seat counts and competing for enterprise budgets. Looking at technology as one block can hide how capital is splitting across different business models.
Bernstein said that backdrop creates a fresh stock-picking setup in software. Lower crowding has taken some enterprise platforms out of overheated positioning, while stable or improving operations have started to pull capital back in. The report combines those two signals to look for gaps between operating performance and market confidence.
Capital is concentrating in usage-growth names
As of Oct. 19, Snowflake had a standardized relative crowding score of 2.13 among the companies covered in the report. Palo Alto Networks stood at 1.86 and Datadog at 1.80. Oracle, by contrast, was at negative 0.37, Adobe at negative 0.70 and HubSpot at negative 1.55. Bernstein said those scores show how investor preferences differ across databases, infrastructure and application software.
Snowflake stood out the most. The report said higher usage on the core platform, together with revenue growth from AI products, helped accelerate the business over the past two quarters. Because the company charges by usage, rising cloud workloads can flow into revenue more directly, and investors have responded by increasing exposure.
Bernstein did not raise its view in step with that hotter positioning. On Snowflake, the report still flagged concerns around a maturing data warehouse market, tougher competition, and the long-term competitiveness of AI tools and data federation approaches. Rising crowding may reflect better growth, but it can also mean optimism about the future has already been pulled forward into the stock price.
Datadog showed a similar split. Bernstein said it is constructive on usage growth from other AI-native customers, while also acknowledging pressure from some large-customer losses and slower non-AI cloud workloads. The report expects spending growth from other AI-native customers to offset those pressures and keep revenue on an accelerating path. Growth in AI workloads remains a key reason institutions favor this kind of model.
Application software still needs operating proof
The rebound in ServiceNow and Atlassian offered Bernstein a clearer set of conditions for capital returning to application software. The firm said revenue reacceleration at those companies, along with their role as enterprise platforms, provides firmer support than an AI narrative on its own. Investor concerns around application software have eased, but the repair has centered on companies that have already produced operating evidence.
The report was cautious on the idea that enterprises will simply replace complex applications by building their own. Using enterprise resource planning, or ERP, as an example, Bernstein said large companies still have to absorb process, data and operating risk. Progress in generative AI tools does not automatically remove those migration and replacement costs. On that basis, the firm remains constructive on the durability of some core enterprise platforms.
Bernstein also said risks need to be broken down by product type and pricing model. In areas such as customer service, job automation could reduce demand for software sold on a per-seat basis. Whether vendors can make up that loss through pricing tied to AI agents remains unproven. Point tools that lack core data and important business context may also face greater substitution risk.
On individual names, Bernstein said it is positive on Oracle’s growth in cloud infrastructure, enterprise applications and databases, though funding constraints and data center buildout execution are still weighing on market confidence. On HubSpot, the report stayed on the sidelines because of weaker growth expectations, lower management confidence and a lack of catalysts. Both companies screen as uncrowded, but the institutional view is clearly not the same.
Microsoft added another example to that framework. Bernstein said concerns over returns on capital spending and Copilot adoption have kept positioning subdued, while recent operating performance has opened the door to a reassessment. Whether capital comes back will depend on checking growth, margins and returns on AI investment on the same operating ledger.
Crowding needs to be read alongside valuation
In Bernstein’s risk model, crowding is treated as a risk created by shared ownership. When investors de-risk at the same time, stocks tied to the same narrative can face concentrated selling pressure. Positions that originally belong to different companies may start to move with strong correlation during the exit phase.
In the worst 10% of months for crowded trades cited in the report, the most crowded quintile underperformed the market average by 3.0%, while the least crowded quintile outperformed by 3.3% on average. Bernstein said that mixing stocks with different crowding profiles in a portfolio can help diversify shared-trade risk. When popular positions are sold in size, less crowded assets have historically offered some protection.
CrowdStrike illustrated the importance of existing ownership. Bernstein said institutional ownership is already high, which limits room for further buying, and actual crowding may be higher than the model score suggests. The report was positive on the company’s competitive position, but said valuation is rich and the growth acceleration expected by the market may be difficult to deliver.
Following the stock-picking logic in the report, the opportunity in software comes from a narrowing of disagreement. Some enterprise platforms are easing substitution concerns through stable growth. Some usage-based companies are turning AI demand into revenue. Other low-crowding names still need to fix growth and execution first. The strongest capital return, Bernstein said, is still going to companies where better operations can reduce market concerns.
The report said capital has started to come back into software, but the repair is not broad yet. Bernstein’s framework is a way to observe positioning. The final answer still comes from the business itself: whether growth can last, how much new revenue AI is adding, and how much improvement current valuations have already priced in.
This article is a整理与解读 by Chaoxiang Research of a third-party brokerage report from Bernstein dated Oct. 28, 2026, combined with public market information. The ratings, target prices, earnings forecasts and related judgments cited in the piece are the views of the brokerage analysts and represent only the position of their institution, not Chaoxiang Research, and do not constitute investment advice.
Markets carry risk, and decisions should be made independently. This article should not be used as a basis for buying or selling any security.
By Rita



