Software has taken a brutal beating in recent weeks, with many names down ~20% year-to-date. The question for investors is whether the current software sell-off signals a structural shift or a market overreaction.
The consensus is that software has become a commodity. My view is more nuanced. While some software companies risk being disrupted, others might actually benefit from AI. The key distinction is between companies that have “artificial limiters” — non-code moats like network effects, regulation, and physical infrastructure that restrict competitive entry and sustain pricing power — and those who don’t. The current indiscriminate sell-off is creating compelling public-market opportunities for both value and growth investors. Paradoxically, the more fragile arena is in the private markets.
This post covers the following:
What’s driving the SaaS / software crash (seat risk, build-vs-buy, and multiple compression)
Why mission-critical SaaS isn’t going away and the opportunity for value investors
The “artificial limiters” framework and how to screen for multi-baggers
Why venture math looks harder in an era of lower public multiples
Let’s dive in.
Understanding the collapsing multiples in software-land
To level-set the conversation, let’s first highlight the current market dynamics. In the past few weeks, hundreds of billions of dollars in market cap have been erased across both incumbent software names and newer entrants. Some examples (as of time of writing):
Workday: down 25.5% YTD (year-to-date)
Salesforce: down 23.7% YTD
Adobe: down 20.6% YTD
Applovin: down 23.5% YTD
Snowflake: down 15.8% YTD
Shopify: down 19.1% YTD
This has been especially painful for high-growth companies, where EV / NTM revenue multiples (enterprise value / next twelve months revenue) have collapsed to just 5x. This is bad relative to just 2 months ago, but worse if we compare it to Covid highs (20x+ EV/NTM).
There are multiple reasons for the current sentiment (some may not be true):
Software companies (broadly) monetize via seat-based pricing. If AI agents can do the work and there are more layoffs, then software vendors won’t be able to sell as many seats
AI-native entrants can move faster and win market share before incumbents ship comparable offerings
Enterprises now believe that they can vibe-code at least a portion of their software in-house vs. paying large contracts for “generic” horizontal SaaS
System of records (Salesforce, Workday, SAP) might be priced as utilities if agents end up taking actions vs. humans
Software companies were overvalued to begin with because premise of SaaS throwing off cash predictably is wrong. The innovation cycle is happening too fast for predictable cash flows and high SBC (stock-based compensation) serve as a further drag on returns
The death of (mission-critical) SaaS is overblown
Of the reasons listed above, the one that seems most worrying is that companies can build much of their software internally. This worry has been amplified after the launch of recent models (Opus 4.6 and GPT-5.3-Codex) and their associated scaffolding / software ecosystem (Claude Cowork with plugins and OpenAI’s Codex).
While this sentiment is justified for feature-thin, nice-to-have software, it doesn’t apply to mission critical applications where the impact of vibe-coded apps going down vastly exceeds the ACVs (average contract value) customers pay for. Here, companies are overestimating their ability to build and maintain software in-house using AI. At the end of the day, customers are also paying for ongoing support, compliance, updates, security, SLA, etc. (which requires full-time headcount). Once the “fully loaded” costs of development are factored in, building software in-house might be higher.
What this means is that beaten down mission-critical software vendors (SAP, Workday, Salesforce) have time to reinvent themselves. Given these companies’ existing distribution and customer trust, they have time to introduce their own AI-native capabilities. Furthermore, tech-forward software incumbents can leverage AI (just like startups) to their various “jobs to be done” (coding, sales, customer support, etc.). This reduces the need for headcount expansion and results in reduced SBC & higher FCF (free cash flow). Companies that adopt this “creative destruction” mentality will see their stock prices re-rate (generating alpha for value investors).
Companies with “artificial limiters” represent multi-bagger opportunities
While the software re-rating story is interesting, I’m more fascinated by beneficiaries of AI. The mental model I’ve adopted is to look for growthy software (or software-enabled) businesses that can re-accelerate growth in the AI era and are protected by artificial limiters (aka non-code moats).
Some (non-exhaustive) categories that I’ve been thinking about:
Social networks: Meta, ByteDance, and to a lesser extent Reddit sit on entrenched user graphs and compounding data advantages. In practice, that means more precise AI ad targeting, measurement, and model training. It’s difficult for a challenger to usurp an incumbent even if they have access to the incumbents’ codebase
Physical infrastructure networks: Cloudflare is a good example. Traditionally known for its CDN & security products, Cloudflare operates a global footprint of points of presence (PoPs) with a material portion of internet traffic flowing through its network. Its footprint, ISP relationships, and the operational know-how creates “limiters” that a startups cannot easily replicate. In Cloudflare’s case, Workers AI and pay-per-crawl become additional growth levers
Enabling infrastructure / primitives: companies like Snowflake, MongoDB, and Datadog offer products that sit directly on top of (or close to) bare metal infrastructure — with high demands on availability, performance, and correctness. Even with AI-assisted development, most enterprises likely do not want to build and maintain core databases/observability primitives. Here, “buy” continues to dominate “build”
Fintech: Coinbase, Robinhood, and SoFi operate inside regulatory and partner constraints: broker-dealer frameworks, bank charters, custody, and monetization relationships (e.g., payment-for-order-flow counterparties) so code velocity isn’t necessarily the bottleneck
One important caveat here is that a “good company” doesn’t mean an “undervalued stock.” Many of these companies can be structural winners and still be fully priced. Finding multi-baggers requires underwriting capability and favorable market timing.
The perils of venture investing in the age of AI
While I’m optimistic on the opportunity set in the public markets, I’m more uneasy in venture-land. The disconnect between VC pricing and public-market multiples implies many investors are still underwriting software deals with mental models from the old SaaS era.
Yes, startups might be “AI native” and many can monetize in new ways (usage-based, outcome-based, tied to labor savings, etc.). But in many cases, it’s still SaaS by another name. These companies face the same questions around long-term defensibility and pricing power. The risk here is that with public multiples already compressed, even if an AI startup fully realizes its vision and meaningfully replaces labor spend, the exit value may not justify the entry price.
This is compounded by structural issues inside VC itself. Many funds are not set up to invest in truly disruptive, longer-duration technologies (fund size, ownership, time horizon mismatch), which I’ve written about prior. So, they end up crowding into “software-shaped” deals because those are easier to underwrite and historically cleared at high multiples. That doesn’t mean VC is devoid of alpha (e.g. my conviction in frontier AI labs). Additionally, some funds are positioned to succeed (partners have the right risk appetite, fund size, trust from LPs, etc.).
The opportunities and perils of the commoditization of code
So, the conclusion here is that AI is a true paradigm shift and tech investing won’t be business as usual going forward. However, the market’s response has been too blunt. In the public markets, there are opportunities at both ends of the spectrum: value investors can underwrite a software re-rating, while growth investors might find multi-baggers via companies with artificial limiters. My take is that mispricings are easiest to find in public equities. The bigger risk is in private markets, where entry prices still assume old-regime prices for new-regime risk.
This post was originally posted my Substack. I can be reached on LinkedIn and X.
Software has taken a brutal beating in recent weeks, with many names down ~20% year-to-date. The question for investors is whether the current software sell-off signals a structural shift or a market overreaction.
The consensus is that software has become a commodity. My view is more nuanced. While some software companies risk being disrupted, others might actually benefit from AI. The key distinction is between companies that have “artificial limiters” — non-code moats like network effects, regulation, and physical infrastructure that restrict competitive entry and sustain pricing power — and those who don’t. The current indiscriminate sell-off is creating compelling public-market opportunities for both value and growth investors. Paradoxically, the more fragile arena is in the private markets.
This post covers the following:
Let’s dive in.
Understanding the collapsing multiples in software-land
To level-set the conversation, let’s first highlight the current market dynamics. In the past few weeks, hundreds of billions of dollars in market cap have been erased across both incumbent software names and newer entrants. Some examples (as of time of writing):
This has been especially painful for high-growth companies, where EV / NTM revenue multiples (enterprise value / next twelve months revenue) have collapsed to just 5x. This is bad relative to just 2 months ago, but worse if we compare it to Covid highs (20x+ EV/NTM).
Source: Needham: SaaS Monthly Drive-Thru (February 2026)
There are multiple reasons for the current sentiment (some may not be true):
The death of (mission-critical) SaaS is overblown
Of the reasons listed above, the one that seems most worrying is that companies can build much of their software internally. This worry has been amplified after the launch of recent models (Opus 4.6 and GPT-5.3-Codex) and their associated scaffolding / software ecosystem (Claude Cowork with plugins and OpenAI’s Codex).
While this sentiment is justified for feature-thin, nice-to-have software, it doesn’t apply to mission critical applications where the impact of vibe-coded apps going down vastly exceeds the ACVs (average contract value) customers pay for. Here, companies are overestimating their ability to build and maintain software in-house using AI. At the end of the day, customers are also paying for ongoing support, compliance, updates, security, SLA, etc. (which requires full-time headcount). Once the “fully loaded” costs of development are factored in, building software in-house might be higher.
What this means is that beaten down mission-critical software vendors (SAP, Workday, Salesforce) have time to reinvent themselves. Given these companies’ existing distribution and customer trust, they have time to introduce their own AI-native capabilities. Furthermore, tech-forward software incumbents can leverage AI (just like startups) to their various “jobs to be done” (coding, sales, customer support, etc.). This reduces the need for headcount expansion and results in reduced SBC & higher FCF (free cash flow). Companies that adopt this “creative destruction” mentality will see their stock prices re-rate (generating alpha for value investors).
Companies with “artificial limiters” represent multi-bagger opportunities
While the software re-rating story is interesting, I’m more fascinated by beneficiaries of AI. The mental model I’ve adopted is to look for growthy software (or software-enabled) businesses that can re-accelerate growth in the AI era and are protected by artificial limiters (aka non-code moats).
Some (non-exhaustive) categories that I’ve been thinking about:
One important caveat here is that a “good company” doesn’t mean an “undervalued stock.” Many of these companies can be structural winners and still be fully priced. Finding multi-baggers requires underwriting capability and favorable market timing.
The perils of venture investing in the age of AI
While I’m optimistic on the opportunity set in the public markets, I’m more uneasy in venture-land. The disconnect between VC pricing and public-market multiples implies many investors are still underwriting software deals with mental models from the old SaaS era.
Yes, startups might be “AI native” and many can monetize in new ways (usage-based, outcome-based, tied to labor savings, etc.). But in many cases, it’s still SaaS by another name. These companies face the same questions around long-term defensibility and pricing power. The risk here is that with public multiples already compressed, even if an AI startup fully realizes its vision and meaningfully replaces labor spend, the exit value may not justify the entry price.
This is compounded by structural issues inside VC itself. Many funds are not set up to invest in truly disruptive, longer-duration technologies (fund size, ownership, time horizon mismatch), which I’ve written about prior. So, they end up crowding into “software-shaped” deals because those are easier to underwrite and historically cleared at high multiples. That doesn’t mean VC is devoid of alpha (e.g. my conviction in frontier AI labs). Additionally, some funds are positioned to succeed (partners have the right risk appetite, fund size, trust from LPs, etc.).
The opportunities and perils of the commoditization of code
So, the conclusion here is that AI is a true paradigm shift and tech investing won’t be business as usual going forward. However, the market’s response has been too blunt. In the public markets, there are opportunities at both ends of the spectrum: value investors can underwrite a software re-rating, while growth investors might find multi-baggers via companies with artificial limiters. My take is that mispricings are easiest to find in public equities. The bigger risk is in private markets, where entry prices still assume old-regime prices for new-regime risk.
Thank you to Will Lee, John Wu, Till Pieper, Homan Yuen, Yash Tulsani, Maged Ahmed, Wai Wu, and Andrew Tan for feedback on this piece.