A Contrarian View on the Biggest Software Selloff Since 2022
The IGV software ETF is down ~17% year-to-date. Roughly $1 trillion in market cap erased since the start of the year. Atlassian down 48%. Asana down 42%. GitLab down 34%. Intuit and DocuSign — both down 28%. CNBC journalists building apps on live television. The Jefferies-coined “SaaSpocalypse” has convinced the market that AI will kill software companies.
I think the market has it backwards.
Not completely wrong — some software companies will die, and I’ll tell you which kind. But the category as a whole? I believe it’s entering what could be its golden decade. The amount of software about to be produced, consumed, and managed is going to be staggering. And someone has to build, deploy, host, and orchestrate all of it.
Let me walk you through my reasoning.
What I’ll Concede
I believe in steel-manning the other side before arguing against it, so let me start with what the bears get right.
Software is genuinely cheaper to build. A team that needed two years and ten engineers can now ship automations as a side project with one or two people. I’ve seen it firsthand.
Simple, single-purpose SaaS — the kind that charges you $20/month for doing one small thing — is genuinely at risk. Think of it like having a factory that produces phone cases from a mold, and 3D printers suddenly becoming so cheap and practical that anyone prints their own. Your factory is dead. That’s what vibe coding does to shallow SaaS. Ready-made software that used to require a company to build? You can now make it from scratch, to your specifications, in an afternoon. Even software that works for you today gets discarded — AI rewrites it, improved, the next day.
And part of the selloff is a correction that was overdue. Many of these companies were trading at stretched valuations. When forward multiples collapse, some of that is AI fear, but some of it is gravity. So the SaaSpocalypse gets a few things right. But it gets the big picture catastrophically wrong.
The Refinery Doesn’t Die When Oil Drops to $5
Here’s the analogy that keeps running through my head.
You’re a refinery. Oil drops to $5 a barrel. Your margins per barrel compress — that’s painful. But at $5 oil, everyone uses more oil. Far more. The volume flowing through your facility explodes. Your total throughput, and likely your total revenue, goes up even as margins per unit shrink.
Software companies are refineries. Code is the oil. AI has made the raw material dramatically cheaper to produce. But the demand for processed, deployed, managed, and secured software? That’s about to go hyperbolic. I can’t imagine how much software is going to be produced. Think of the difference between a donkey carrying bags and a megaton cargo ship. That’s the scale difference between pre-AI and post-AI software production — and most people haven’t wrapped their heads around it yet.
I have quite strong conviction in this view, arguing that enterprises rarely buy technology directly. They buy solutions to business problems. Those solutions come through software. The more AI you deploy, the more software you need to manage it.
The System, Not the Model
What’s starting to happen is the rise of the digital employee. They appear expensive at thousands of dollars in tokens per year, but they’re roughly 10x cheaper than humans for certain tasks, and in some areas easier to manage and more reliable. What has been cracked — not by a model, but by a system that uses a model — is coding. You can now pay for cheap tokens and get real knowledge work done at a value that’s multiples bigger than the cost.
Right now, small organizations that have figured this out see enormous benefits. But soon, competition normalizes things. You won’t be competing against organizations that don’t use digital employees. You’ll be competing against other organizations that also do. Easy profits won’t last. But decent profits come from doing things better — and the total volume of work flowing through software systems will be orders of magnitude larger.
This is the refinery dynamic again: margins compress, but volume explodes.
Who Orchestrates the Robots?
So if everyone will have digital employees, who puts together the workflows? Who manages the systems that make them useful?
Software companies and IT departments.
I’m not saying all work will be done by software engineers. You’ll have people who are softer engineers — call them vibe coders. But this type of work will ultimately belong to the IT department. A chatbot or assistant tool helps everyone, sure. But that’s not where the massive value creation happens. The massive work is in systems and workflows, orchestrated either by software companies providing solutions to non-tech companies, or by internal IT departments.
What about non-tech companies building their own? Valid concern. But to outcompete software companies with years of experience managing complex systems, they need extraordinary domain knowledge, market position, and execution. And the catch: the few non-tech companies that actually succeed will effectively become software companies. You can’t escape it.
The size of this orchestration work is about to explode.
The Dividing Line
This is the critical distinction the market is completely ignoring. Not all software is the same.
What’s at risk: Solutions for the general public solving one isolated problem. Companies that built billion-dollar businesses on what amounts to a feature. Low switching costs. No deep data. No enterprise processes. These get rebuilt in minutes.
What will thrive: Deep processes. Massive databases. Lock-in. Compliance. Heavy infrastructure. Think of it like the wholesaler in commerce. When e-commerce made it easy for anyone to sell online, people predicted the death of wholesalers. Instead, wholesalers thrived — more sellers needed more supply chains, more logistics, more infrastructure. Enterprise SaaS is the wholesaler of the software economy.
| Characteristic | Shallow SaaS (At Risk) | Enterprise SaaS (Will Benefit) |
|---|---|---|
| Customer type | Consumer / SMB | Enterprise |
| Lock-in | Low | High (months/years to switch) |
| Data depth | Minimal | Massive databases, systems of record |
| Can be rebuilt with AI? | Yes, in minutes | No — requires data, process, trust, compliance |
| Moat | Pure functionality (none) | Integrations, data, compliance, network effects |
Even with vibe coding, the entity that builds it once and resells it is cheap AND has a moat. Horizontal point-solution SaaS with low switching costs will struggle; vertical, domain-specific SaaS with proprietary data and deep enterprise integration will survive and thrive.
The Robot Uses the Fridge
AI agents need APIs, not dashboards. A robot uses a fridge — it won’t become a fridge.
This distinction is what almost everyone misses. The GUI, the pretty interface, the dashboard — that’s for humans. Agents interact with software through APIs, through programmatic access, through structured protocols. They don’t need the user-friendly interface. They need the machine-friendly one.
Think about what this means. Every agent that gets built needs to call APIs. Every workflow that gets automated touches 5, 10, 20 different software services through their APIs. More agents means more API calls means more software usage. Not less. More.
MCP — the Model Context Protocol — is the clearest expression of this trend. The industry is building standardized protocols for AI-to-software communication. Every software company is racing to make their tools accessible to AI agents. This is the opposite of disruption. This is integration.
Jensen Huang has been saying the same thing from the hardware side: more AI means more compute, more software, more infrastructure. The infrastructure layer grows, the software layer grows, the agent layer grows on top. It’s additive, not substitutive. When the CEO of the company supplying the picks and shovels tells you that the demand for what sits ON TOP of his infrastructure is also expanding — pay attention.
Cloudflare reported that weekly requests from AI agents is increasing. This isn’t a theoretical argument. The robot is already using the fridge, and the fridge is getting busier.
The Profitability Explosion
Here’s where I think the market is most wrong — not just about the direction, but about the magnitude.
Two forces are converging for software companies, and almost nobody is talking about both at once.
Force 1: The cost side is dropping. As AI tools make engineers more productive, effective engineering output increases dramatically. More people can contribute meaningful code — vibe coders, citizen developers, AI-assisted specialists. Companies need fewer engineers for the same output, or the same engineers produce 5-10x more. Engineering costs compress.
Force 2: The revenue side is expanding. And not just the current $400 billion SaaS market growing at 13-19% CAGR. Software plus agents are starting to eat into the global knowledge work economy. The total addressable market for knowledge work globally is in the dozens of trillions of dollars. Even capturing a small additional slice of that is an order of magnitude larger than the current software market.
Combined effect? Software company margins should expand. Lower cost base plus growing revenue equals a profitability explosion. The market is pricing in revenue destruction, when the actual dynamic is margin expansion plus TAM growth. A double tailwind.
I want to be honest here. The knowledge work TAM numbers — the $30 to $50 trillion globally — these are my intuition based on heuristics, not hard predictions. I’m not claiming software will capture all of it. But the idea that software can only ever address a $400 billion market when AI allows it to automate and orchestrate trillions in knowledge work? That strikes me as far too conservative. Even the not-particularly-bullish analysts project the SaaS market reaching $1.2-1.5 trillion by 2032-2035. I think the ceiling is much higher.
The refinery analogy extends to its logical conclusion: margins per barrel compress, but the ocean of oil flowing through it goes from a river to a sea. That’s what a golden decade looks like for software companies with deep enterprise integrations, large data moats, strong API ecosystems, and existing customer relationships.
What I See on the Ground
Let me share something from my own experience working with enterprise software infrastructure.
Many have argued that what a team had built over many years, a single person could now build in a month with AI. And in some ways, true — the core logic, the processing, the clever parts. Someone working a few weekends could now build a decent version.
But to make it work at production scale? You need serious infrastructure. Massive data pipelines. Storage. Security. Compliance. Customer support. Even if someone builds the core cheaply, the infrastructure to run it reliably still costs serious money.
So what actually happened was the opposite of what was feared.
Because software is easier to build, small teams build automations as side projects — and those automations need data, need APIs, need infrastructure. The companies I’ve seen up close get bombarded with new API access requests from small teams building AI-powered tools. Demand for structured data and platform services has surged.
And here’s the key: all the magic that happens with AI is the mix of LLM with code in various shapes. LLMs create code. Code calls LLMs. Code calls APIs. APIs access data. Data feeds LLMs. It’s a compounding ecosystem where human-like behavior and computer-like behavior bring their best strengths. And all of it needs structured data, software infrastructure, and API access to function.
The more software that gets built, the more infrastructure companies benefit. That’s the refinery in practice.
The Data Confirms It
I’ve laid out my reasoning. Now let me show you that the data backs it up.
| Company | Quarter | Revenue | YoY Growth | Stock Drop (YTD) |
|---|---|---|---|---|
| Cloudflare | Q4 2025 | $614.5M | +34% | -1% |
| GitLab | Q3 FY2026 | $244.4M | +25% | -34% |
| Atlassian | Q2 FY2026 | $1,586.3M | +23% | -48% |
| Wix | Q3 2025 | $505.2M | +14% | -10% |
| SimilarWeb | Q4 2025 | $72.8M | +11% | -63% |
| Box | Q3 FY2026 | $301M | +9% | -14% |
Not one shows revenue decline. Every single one is growing. The market is pricing in destruction that isn’t showing up in the data.
The market’s own logic doesn’t hold. As BofA’s Vivek Arya noted, investors simultaneously believe AI is too weak to justify capex AND so powerful it destroys software. Pick one. He also pointed out something remarkable: the Russell 1000 Software subsector trades at 32.4x forward earnings versus 43.6x for semiconductors — recurring-revenue businesses with 90%+ gross margins at lower multiples than cyclical chipmakers with 40-60% margins.
J.P. Morgan called the selloff “broken logic”, noting software has lost $2 trillion from its peak. And Bill Gurley’s advice puts it simply: “Channel your inner Warren Buffett. You should be quiet and picking them up off the floor.”
My Bets
These aren’t stock recommendations. I own positions in all of them, so take my views accordingly.
Figma — Deeply integrated in the design-to-code workflow. Vibe coding doesn’t replace design; it uses design tools. For production you need Figma as a tool that external AI or native AI controls. More software built means more teams using it.
Wix — Acquired a coding agent company (Base44) and is integrating AI directly into their platform for existing customers. Playing offense, not defense.
Box — I chose Box betting on the CEO, Aaron Levie. He is one of the few leaders who genuinely understands current AI without fancy fake statements. I trust the leader.
GitLab — Code gets produced super fast now. The bottleneck shifts to deployment, CI/CD, and security scanning. GitLab handles exactly that bottleneck. More code means more deployments.
Atlassian — Nothing great on their platform that can’t be built by others. But the integrations, the large companies already embedded — they won’t change. Like Excel: even if you hate it, you keep using it.
SimilarWeb — Their web analytics data is exactly what AI agents consume. AI-driven revenue is already 11% of total revenue and rising. As more agents need web intelligence data, their dataset becomes more valuable.
My portfolio logic: buy a basket, hold 5 years. Unlike semiconductors, I don’t expect all to work. But I believe 1-2 could be multi-baggers. I buy and forget.
About a year ago, I wrote about why AI is not a bubble and argued that the investment wave had much further to run. That call aged well. I hope I can be right again.
The Honest Assessment
I want to be clear about what I’m saying: the SaaSpocalypse narrative is too simple. It treats all software companies the same, ignoring the difference between shallow SaaS that gets rebuilt in minutes and enterprise infrastructure that becomes more valuable as software production explodes. The robot uses the fridge. Software is the fridge.
But I also want to be honest about uncertainty. Predicting how a specific company fares in this environment is very difficult — each case is different. The macro view — software as a sector benefits enormously from AI — I hold with strong conviction. The micro — which exact companies win — that’s a black box. This is one of the most intriguing things happening in the economy as a side effect of AI.
Here’s my practical test: watch the quarterly revenue.
If after two quarters of AI tools being widespread, companies where common sense says AI should help show no revenue decline, you can start feeling confident the market has misread the situation. If I see a true 10% decline where I expected growth, that’s a very strong signal — and I’m ready to change my mind quickly.
Want to know what actual disruption looks like? Look at Stack Overflow. Its monthly visits fell 50%, questions dropped 76%. You could see the decline within a quarter or two — real disruption shows up fast. Yet even Stack Overflow’s revenue grew from $89M to $125M by pivoting to AI data licensing. Even the most disrupted company found new AI-driven revenue.
So far? Every company I track is growing. Cloudflare’s AI agent traffic doubled. Atlassian hit its first $1 billion cloud quarter. Wall Street keeps panicking over developments that practitioners have known about for months — just like DeepSeek in January 2025, which turned out to be the buying opportunity of the year.
Could I be wrong? Of course. But if you’ve made up your mind that AI kills software companies, I hope this gives you a few reasons to think again.
The SaaSpocalypse is a good story. It’s just not the right one.
Disclaimer: This article represents my personal analysis and opinions. It is not financial advice. I own positions in Wix, Box, Figma, GitLab, Atlassian, and SimilarWeb. Always do your own research before making investment decisions.
Discover more from FinAI
Subscribe to get the latest posts sent to your email.
Leave a Reply