Bernstein: Software stock congestion is still at an all-time low. Crowded trading has now reversed the signal, and Microsoft (MSFT.US) has become a “must-have”
Entering the fourth quarter, how “crowded” are software stocks? Bernstein gave the answer: The software sector's congestion level rebounded slightly in the third quarter from July, but it was still significantly lower than the historical average for the past 25 years. Based on this, chief analyst Modler's team judged that software stocks as a whole are still “unpopular,” but structural opportunities are hidden in the unpopular.
The so-called “degree of congestion” measures whether actively managed funds have concentrated their holdings in a certain stock too much. Stocks that are more crowded are more likely to experience collective sell-offs and liquidity stampedes once the market is under pressure.
The data shows that in months where the VIX Volatility Index is above average, the most crowded combination outperformed the least popular group by about 40 basis points per month. There was a rare exception in the first half of this year: overcrowded stocks outperformed unpopular stocks by 13.3%, and the TMT sector was even higher than 51%; however, the trend reversed in the second half of the year, and the most crowded names in TMT outperformed the most unpopular group by 30.2%. Bernstein cautioned that crowding is critical to downside risk control, diversification, and subsequent returns.
Looking at each track, there is a sharp division within the software. Databases, consumer infrastructure, and cybersecurity are still the three “most crowded” directions — Snowflake (SNOW.US) has gone from being the “most unpopular” to the “most crowded” individual in coverage, followed by Datadog (DDOG.US), Palo Alto Networks (PANW.US), Twilio (TWLO.US), and Okta (OKTA.US).
On the other hand, traditional application software that is priced by “number of seats”, such as Adobe (ADBE.US), HubSpot (HUBS.US), GitLab (GTLB.US), and Zoom (ZM.US), continues to be snubbed due to poor performance or strategic adjustments. In SaaS applications, ServiceNow (NOW.US), Atlassian (TEAM.US), Salesforce (CRM.US), etc. have recovered from congestion due to steady revenue growth.
At the giant level, although Microsoft (MSFT.US) congestion has declined, it is still positive. It is the “must-have standard” in the report — its GAAP operating profit margin has stabilized at around 45%, and Azure and Office 365 Copilot payments have accelerated, but the growth rate of capital expenditure is slowing down, and the valuation is lower than the historical center.
Oracle (ORCL.US), on the other hand, has slipped from “most crowded” to the bottom due to capital constraints and data center execution concerns; Bernstein believes that its performance continues to be solid and that the dimensional value is expected to increase several times over 5-10 years, but unpopularity is an attractive buying point. SAP (SAP.US) ranks in the middle and is regarded as a “compound interest target”.
Bernstein's core position is that the short-term “AI disrupts SaaS” (SaaS Catastrophe) narrative is overrendered. Modler believes that the idea of companies using generative AI to develop their own software is overrated both in terms of the technology maturity cycle and willingness to take risks, and complex large enterprise ERPs are even less likely to be self-developed in-house. The real risks focus on seat-based segments such as help desks, and single-point software that lacks a systematic record of data and semantic knowledge.
Regarding the software sector as a whole, Bernstein still maintains the “worth investing in” conclusion, but emphasizes the need for careful selection — looking for targets with solid fundamentals and capable of crossing the AI cycle in the unpopular, rather than chasing crowded deals that have already been overheld.