Sponsored by edgeful

Software stocks are getting absolutely destroyed.

Salesforce is down roughly 26% year to date. Adobe is off more than 30%. ServiceNow, which was supposed to be the invincible one, has shed close to 28%. Intuit dropped nearly 11% in a single session and is now down over 34% on the year. The IGV software ETF has entered a technical bear market, falling more than 20% from its late 2025 peaks. In the first week of February alone, roughly $300 billion in software market value evaporated in a 48-hour window.

And this isn't just a US story. Japanese IT firms like TIS plunged 16% in a single session. India's Nifty IT index dropped nearly 6%. Chinese software names from Kingdee to Baidu sold off hard. This is a global repricing of what software is worth.

But here's what makes this so unsettling. If you look at the actual financial results these companies are reporting, they're... fine? Median revenue growth across public software companies has held in the 12-18% range heading into 2026. Net retention has been consistent. A growing number of companies have been guiding for acceleration. These are not the numbers of an industry in freefall.

So the stocks are crashing while the businesses are holding up. That disconnect is exactly what we need to talk about.

Here's the thing about software investors. For over a decade, they've been trained to pay premium multiples for one specific quality: visibility. The entire reason software stocks commanded 30x, 40x, sometimes 60x forward earnings was the predictability of the revenue stream. You sign a customer to a multi-year subscription. They pay every month. They almost never leave. You can model out the next three to five years of revenue with surprising precision, and investors rewarded that clarity with historically generous valuations.

That compact is breaking.

It's not that software revenue is falling off a cliff tomorrow. It's that the forward visibility that justified those premium multiples is now clouded by a level of uncertainty the sector has never dealt with before. And when investors can't see the road ahead, they don't pay up for it. It's that simple.

Think about what the market is actually pricing in. The industry's average forward P/E ratio has dropped to roughly 21x, down from 39x just eight months ago. That's the largest four-month valuation compression since the dot-com bust in 2002. Software price-to-sales ratios have compressed from 9x to about 6x, levels we haven't seen since the mid-2010s. Adobe's forward P/E has been cut from 30x to 12x. ServiceNow from 67x to 28x.

The market is no longer willing to give these companies credit for the future. It's only willing to pay for what it can see right now.

This is why the sell-off feels so bizarre when the quarterly numbers look decent. The market isn't reacting to Q4 2025 results. It's reacting to the growing possibility that the next five years look nothing like the last ten. And when you're an institutional investor at a multi-strategy fund, you cannot afford to be the person holding the bag while that question gets answered.

I want to hammer this point because I think a lot of people are missing it.

You don't need revenue to fall for a software stock to get cut in half. You just need the multiple to compress. A financial data company that traded at 15x revenue when it had unquestioned pricing power and 95% retention might trade at 6x revenue when the market believes both of those things are eroding. Revenue stays flat. The stock drops 60%.

That's exactly what's happening right now. The market isn't pricing in a revenue collapse. It's pricing in the end of the premium multiple, because the competitive advantages that justified that multiple are dissolving in real time.

Goldman Sachs strategist Ben Snider drew a parallel worth paying attention to here. He compared the software sell-off to what happened to newspapers and tobacco, both industries that faced structural disruption from technology or regulation and experienced prolonged, multi-year declines in stock prices. In those cases, the share prices only stabilized after earnings had already fallen for years. By the time profits bottomed, most of the equity value was already gone.

That doesn't mean software follows the same path. But it illustrates a critical point: when there's a genuine structural question over a sector's future earnings power, the market de-rates first and asks questions later. The stocks only recover when companies prove the threat was overblown or that they've successfully adapted. Until that proof shows up in actual numbers, the multiple stays compressed.

To understand why the multiples are compressing, you have to understand what's actually happening to the competitive advantages that supported them. Not everything in software is equally vulnerable. But the specific things that kept competitors out and pricing power high are exactly the things AI is dismantling.

The learned interface is collapsing. This is probably the most underappreciated moat in software. A Bloomberg Terminal user has spent years learning keyboard shortcuts, function codes, and navigation patterns. "We're a FactSet shop" or "We're a Bloomberg house" aren't statements about data quality. They're statements about muscle memory. People invested thousands of hours learning these tools, and that investment isn't transferable. When the interface is a natural language conversation, all of that muscle memory becomes worthless. The switching cost that justified $25,000 per seat per year just dissolves.

Business logic is migrating from code to plain language. This is the most devastating long-term shift. Traditional software encodes how an industry works in thousands of lines of code, if/then branches, validation rules, compliance checks, approval workflows, all built by engineers who also understand the domain (which is incredibly rare). Modifying any of it requires development cycles, QA, deployment. Now a portfolio manager who's done 500 DCF valuations can describe their entire methodology in a document that an AI agent executes. Years of specialized engineering, replaced by a week of writing by a domain expert. The accumulated business logic that took software companies a decade to build can be replicated in weeks.

The data accessibility layer is being commoditized. A massive portion of vertical software's value was making hard-to-access data easy to query. FactSet makes SEC filings searchable. LexisNexis makes case law searchable. These are genuine services. But frontier AI models already know how to parse a 10-K filing. They understand the difference between GAAP and non-GAAP revenue. They can navigate nested tables of segment disclosures without being taught the schema. The parsing and structuring infrastructure that vertical software spent decades building is now a commodity capability baked into foundation models. The data isn't worthless. But the "making it searchable" layer, which is where a lot of the pricing power lived, is evaporating.

The talent barrier to entry is inverting. Building vertical software used to require people who understand both the domain and the technology. Finding an engineer who can write production code AND understands how credit derivatives work, or how litigation workflows function, is extremely rare. That scarcity naturally limited competition in each vertical to two or three serious players. Now domain experts can encode their knowledge directly into AI-readable formats without writing a single line of code. The engineering bottleneck that kept competitors out is gone.

The bundling moat is weakening. Software companies expand by bundling adjacent capabilities. Bloomberg started with market data, then added messaging, news, analytics, trading, compliance. Each module increases switching costs because customers depend on the entire ecosystem. But AI agents break this because the agent IS the bundle. It can pull market data from one source, news from another, run analytics through a third, and compile the results in a single workflow. The user never knows or cares which underlying providers were queried. When the integration layer moves from the software vendor to the AI agent, the incentive to pay a premium for the bundled suite weakens considerably.

Not everything is breaking. And understanding which competitive advantages hold up is just as important as knowing which ones don't.

Proprietary data becomes more valuable, not less. If your data genuinely cannot be replicated, scraped, or licensed from a third party, AI actually strengthens your position. Bloomberg's real-time pricing data from trading desks, S&P Global's credit ratings backed by decades of default data, Dun & Bradstreet's business credit files on 500 million entities. In an AI world, truly unique data becomes the scarce input that every agent needs. The pricing power on proprietary data may actually increase. But here's the critical distinction: if your data can be obtained, licensed, or synthesized by someone else, you're not safe. You become a commodity supplier to the AI agent, competing on price while the agent captures the customer relationship and the margin.

Regulatory and compliance lock-in is structural. HIPAA doesn't care about AI. FDA certification doesn't get easier because a new model drops. SOX compliance requirements don't change because Anthropic released a new plugin. In healthcare, Epic's dominance is fundamentally a regulatory fortress. The 18-month implementation cycles, the compliance certifications, the integration with hospital billing systems. None of that is affected by AI. In fact, regulatory requirements may actually slow AI adoption in exactly the verticals where compliance lock-in is strongest.

Network effects remain sticky. Some software becomes more valuable as more industry participants use it. Bloomberg's messaging function is the de facto communication layer for Wall Street. If every counterparty uses Bloomberg, you have to use Bloomberg. Not because of the data. Because of the network. AI doesn't break network effects. If anything, the information flowing through those networks becomes more valuable as context and signal.

Transaction embedding is durable. When your software sits directly in the money flow, processing payments, originating loans, settling trades, switching means interrupting revenue. Nobody does that voluntarily. An AI might sit on top of these rails as a better interface, but the rails themselves remain essential. Stripe isn't threatened by AI. Neither is FIS or Fiserv.

System of record status holds for now. When your software is the canonical source of truth for critical business data, switching isn't just inconvenient, it's existentially risky. Though this is worth watching long-term, because AI agents are quietly building their own contextual memory across every system they touch. Over time, the agent that sees your CRM, your email, your documents, and your Slack may accumulate a richer picture of your business than any single system of record.

Here's what makes this moment truly dangerous for incumbent software companies. They're not just getting attacked from one direction. They're caught in a pincer movement.

From below, hundreds of AI-native startups are entering every vertical. When building a credible financial data product required 200 engineers and $50 million in data licensing, markets naturally consolidated around three or four players. When it requires a small team with frontier model APIs and domain expertise, the market fragments violently. You don't go from three competitors to four. You go from three to three hundred. And that's what craters pricing power. Fifty AI-native startups offering 80% of the capability at 20% of the price is a competitive dynamic that no amount of brand loyalty or switching costs can fully withstand.

From above, horizontal platforms are going deep into vertical territory for the first time. Microsoft Copilot inside Excel now does AI-powered DCF modeling and financial statement parsing. Copilot inside Word does contract review and case law research. Anthropic is doing the same from the other direction, with Claude's Cowork plugins targeting specific industries like legal and finance. The horizontal tool becomes vertical through AI, not through engineering. And this is arguably the most existential threat, because these horizontal players have distribution advantages that no vertical startup can match.

The technology that enables vertical depth (AI models plus pluggable data access plus domain-specific instructions) is the same technology that lets horizontal platforms compete in territory they could never reach before. Vertical software used to be safe specifically because horizontal players couldn't go deep enough. That barrier just fell.

On top of the competitive moat erosion, there's the direct revenue model threat. The mainstream narrative frames this as "AI replaces software." That's not quite right. The more precise version, and the more dangerous one, is this: AI replaces the humans who use software.

The software itself might stick around. But the per-seat revenue model that powered these businesses for two decades is under direct structural pressure. If AI agents can do the prospecting, the data entry, the pipeline management, the follow-ups, and the reporting that 100 sales reps used to do, maybe you still need 20 humans supervising the process. The work output is the same. The software revenue is down 80%. No customer churned. Retention looks fine. But the revenue base just structurally shrank.

This is already forcing companies to rethink their pricing. Salesforce has started experimenting with consumption-based credits. Adobe is moving toward generative credit systems. Zendesk, which went private specifically to navigate this transition away from quarterly scrutiny, has begun charging per resolved issue rather than per seat. The industry is moving from "per seat" to "per outcome." But that transition introduces exactly the kind of revenue volatility and unpredictability that investors refuse to pay premium multiples for.

Here's the part nobody wants to hear.

Even if these businesses survive, and many will, the valuation multiples investors got used to from 2015 to 2024 are very likely not returning. The reasons stack on top of each other.

Growth is decelerating structurally, not cyclically. Public SaaS growth rates have declined every single quarter since the 2021 peak. Every quarter. By Q4 2025, median revenue growth had fallen to around 12%, down from 40% at the pandemic peak. Forecasts point to continued deceleration through at least mid-2026. When you're a sector that got valued on growth, and growth keeps shrinking, the multiple comes down regardless of anything else.

Revenue predictability is being undermined. The whole thesis behind paying 40x for ServiceNow was that its recurring revenue was almost annuity-like. But when AI agents reduce headcount at customer organizations, every enterprise renewal becomes a question mark. Not because the customer is unhappy, but because they need fewer seats. Churn stays low while revenue retention quietly deteriorates. That guts the visibility argument.

Outcome-based pricing introduces volatility. Per-seat pricing was beautiful for financial modeling. Smooth, predictable, expandable. Outcome-based pricing is inherently variable. It depends on usage, volume, and AI performance. That's better for customers. It's worse for the revenue predictability that Wall Street rewards with premium multiples.

Stock-based compensation prevents valuation floors. The median public software company runs about 5% GAAP net income margin once you factor in stock comp. That's not enough to fund buybacks, dividends, or the real free cash flow that attracts value investors when stocks get beaten down. There's no natural buyer stepping in at these prices to stop the bleeding.

Capital is rotating away structurally. A January 2026 CIO survey showed IT budget growth decelerating to 3.4%, but the internal reallocation is far more dramatic. Money is being pulled from application software and redirected toward AI infrastructure. Hyperscalers plan to spend roughly $660 to $690 billion on AI infrastructure in 2026, nearly double 2025 levels, and a meaningful chunk of that is coming directly from enterprise software budgets.

The barrier to entry has collapsed. This might be the single most important factor. When your vertical had two or three serious competitors, you could maintain pricing power. When it has 300, you can't. The explosion of AI-native competitors in every vertical software category is a permanent structural change, not a temporary disruption.

Put all of this together and what you get is a sector where growth is declining, visibility is deteriorating, the pricing model is shifting toward something less predictable, the financials don't support traditional valuation floors, the capital flows are moving away from you, and competition is exploding from both directions. That's not an environment where 35x forward earnings comes back. That's an environment where 15-20x becomes the new ceiling.

Not all software is equally exposed. If you're trying to figure out which companies make it through this and which don't, there are really three questions that matter.

Is the data proprietary? If the company owns or generates data that genuinely cannot be obtained, licensed, or synthesized by anyone else, the moat holds. If the data is public or licensable, the accessibility layer that justified premium pricing is collapsing. AI actually accelerates this bifurcation. Companies with truly proprietary data win bigger. Companies without it lose everything.

Is there regulatory or compliance lock-in? If the switching cost is driven by regulatory certification, compliance infrastructure, and deep integration with mission-critical workflows, AI doesn't change the equation. HIPAA, FDA validation, SOX compliance. These are structural moats that have nothing to do with whether a new model can write code.

Is the software embedded in the transaction? If the software processes payments, originates loans, or settles trades, AI sits on top of it as a better interface. It doesn't replace the rails. The transaction layer is infrastructure, and infrastructure endures.

If a company scores zero out of three, it's in serious trouble. One out of three, it's a mixed bag where the stock probably hasn't finished declining. Two or three, it's likely going to be fine, even at compressed multiples.

I want to be really clear about what I'm saying here. I don't think software as an industry goes to zero. There's too much complexity in enterprise systems, too much embedded data, too much regulatory scaffolding for these businesses to just vanish. Jensen Huang also recently said: the idea that AI kills the software industry is illogical.

But, and this is the critical part, the multiple that the market is willing to assign to these businesses has structurally changed. The forward P/E dropping from 39x to 21x in four months isn't a panic that snaps back in a few quarters. It's a repricing that reflects a genuine, permanent increase in uncertainty about long-term earnings power. The specific moats that justified premium valuations, the learned interfaces, the business logic, the data accessibility layers, the talent barriers, the bundling strategies, are exactly the ones that AI is dismantling. The moats that hold, proprietary data, regulatory lock-in, network effects, transaction embedding, are real but not universal. Plenty of software companies don't have them.

Software is becoming what some people are calling "headless." The interface disappears. Everything flows through the AI agent. What matters isn't the software anymore. It's who owns the customer relationship. And increasingly, the answer to that question is the AI agent, not the legacy software vendor.

So don't buy the index. Don't buy the dip blindly, because this might not be a dip. This might be a new normal. Look company by company. Apply the three-question test. Ask yourself, does this business own data that can't be replicated? Is it protected by regulation? Is it embedded in the transaction? If the answer is no to all three, the stock is probably still expensive even after a 30% decline.

The companies that pass those tests will be solid investments even at compressed multiples. The ones that don't will continue to bleed, slowly, for years, while the market waits for proof that never comes.

The businesses might not die. But the era of paying 40x for the privilege of owning them? That's over.