SaaSpocalypse and the "Shoot First" Trade

Programming note: our next issue lands 6 March, where we will examine the current market narratives about key AI players and platforms.

AI Eats Software

Over the past few years, the market's operating assumption was that Artificial Intelligence was a magical productivity booster that would lift all boats.

Since the beginning of this year, however, that sentiment has curdled into what Wall Street is calling the AI Scare Trade. The logic has flipped: if AI is a magical productivity booster, it might boost us out of needing to buy software at all.

This panic culminated in a brutal, rolling sell-off across the software sector—dubbed the "SaaSpocalypse"—triggered almost entirely by Anthropic releasing a few new plug-ins for its Claude AI model. The damage is widespread: the benchmark IGV ETF is down 24% year-to-date. Intuit has been gutted (-38%), while category leaders like Salesforce (-26%) and Adobe (-25%) have been decimated. Even "safe" bets with resurgent narratives like IBM (-19%) and ServiceNow (-30%) have been caught in the downdraft, and Asana (-48%) has surrendered nearly half its value in just ten weeks.

It is a fascinating moment of market psychology. But as software investor Jared Sleeper noted on Bloomberg podcast, the panic reveals a structural truth about Wall Street: very few people who invest in software actually understand what software does. They know the financial spreadsheets, but they don't actually know what the customer is buying.

The market is currently operating under the delusion that "Software" is just a synonym for "Code," and that if an AI can write code, the value of software goes to zero. But software is rarely just code. It is workflow integration, third-party interoperability, and uptime guarantees.

Two of the steepest casualties of the recent sell-off perfectly illustrate how the market has confused the language of a product with the value of a business.

Thomson Reuters and Decision Insurance

In early February, Anthropic released a legal plug-in for Claude that can automate contract reviewing and legal briefings. The market instantly wiped 16% off the market cap of Thomson Reuters (TRI), treating the legal information giant as if it had just been rendered obsolete.

The assumption here is that lawyers buy Thomson Reuters to save time typing. But that fundamentally misunderstands corporate legal practice.

When an in-house legal team or a government department is making a compliance decision, the priority is not efficiency or cost. The priority is not being wrong. A lawyer's job is not to produce legal text; it is to make a defensible decision based on established market practice.

Thomson Reuters does not sell a typing assistant; it sells proprietary legal research and a definitive database of precedent. It sells decision insurance. If an in-house counsel makes a bad call but can point to the definitive TRI database to show it was standard practice, his professional backside is institutionally covered. If he makes a bad call and says, "I asked the chatbot to review the contract," he will be fired, immediately.

As Morningstar pointed out in a research note, Claude's plug-in has nothing to do with legal research, which is the core value proposition of Thomson's legal businesses. The market panicked because AI can generate legal text, forgetting that lawyers are paid for defensible legal decisions—a product LLMs are notoriously terrible at providing.

IBM and the "Syntax vs. Platform" Fallacy

If the Thomson Reuters sell-off was a misunderstanding of liability, the IBM crash was a misunderstanding of physics.

Last week, IBM suffered its steepest daily drop since the year 2000—a 13% collapse—after Anthropic published a blog post claiming its AI could modernize COBOL, the ancient programming language that still runs on IBM mainframes.

The market logic: IBM makes money maintaining old COBOL systems. AI can translate COBOL instantly. IBM is dead.

The market looked at the headline and concluded that IBM's moat was gone because anyone could now translate COBOL into modern code. But as IBM's Chief Commercial Officer Rob Thomas pointed out in a rebuttal, the market is confusing syntax with system architecture:

The modernization challenge is not a COBOL language problem. It is everything the application runs on and integrates with. Enterprise COBOL on IBM Z sits inside a vertically integrated stack: z/OS, CICS, IMS, Db2, RACF, MQ, Parallel Sysplex, and Cybervault with DS8K Storage. That stack is what enables 25 billion encrypted transactions per day on a single system, 450 billion AI inferences per day at 1 ms response time, up to eight nines of availability, quantum-safe encryption, and sustained 100 percent utilization without impacting SLAs. Translating COBOL does not move any of that.

...Given the depth of on-premises dependencies, it is difficult to see how a SaaS-only solution can replace the COBOL applications on the mainframe, meeting the demands of the enterprise. And given everything happening around digital sovereignty and data residency, would an organization make its most critical transactions dependent on a provider operating in a jurisdiction it does not control?

In other words, the real work is data architecture, transaction integrity, and baking non-functional requirements into the platform itself. That is system-level engineering, not language conversion.

COBOL handles an estimated 95% of US ATM transactions. These are the definition of mission-critical operations, where the cost of being wrong is systemic. The reason banks use IBM mainframes for this is not because they have a sentimental attachment to the COBOL language. They use it because the IBM Z platform is a vertically integrated stack designed to never, ever crash. No sane CIO is going to trade time-tested, hardware-verified reliability for an experimental "vibe coding" session.

The market panic proved that investors think migrating a global banking system is as easy as running Google Translate. They saw an AI translate syntax and assumed it had replicated decades of physical hardware optimization.

The Pod-Shop Panic

So why is the market selling off so violently if the fundamental businesses remain intact?

Part of the answer lies in the structure of modern trading. As Jared Sleeper pointed out, the market is dominated by multi-manager hedge funds (like Citadel and Millennium) where dozens of small, independent trading teams—or "pods"—operate under one roof. Each pod runs on tight risk limits and simply cannot afford a drawdown, meaning the mandate is always to "shoot first and ask questions later."

Because the threat of AI feels existential, and because the actual financial damage to these software companies won't be measurable for years, there is no fundamental "floor" to stop the selling.

The AI Scare Trade is pricing in a world where enterprises happily hand over their legal liability and their ATM networks to a somewhat unpredictable language model just to save a few bucks on IT. It is a world where the code is everything, and the context is nothing.

It is a very exciting narrative for the AI labs. But it's probably not how a bank is going to run its business.

P.S. The SaaSpocalypse highlights a critical problem: How do you distinguish a structural fundamental shift from a temporary narrative panic? For the past few months, we've been building a quantitative diagnostic model to answer exactly that. It "X-rays" the tech sector to reveal the specific valuation lens investors are prioritizing (Revenue, EBITDA, FCF), whether a tech company is historically cheap under that lens, and if price momentum confirms the story. We are opening a private beta for the dashboard next week. If you want early access, apply here.

More on SaaSpocalypse:

• Why SaaS Stocks Have Dropped—and What It Signals for Software’s Next Chapter (Bain)
• Software selloff shows AI acceleration (Blackrock)

You received this message because you are subscribed to ARPU newsletter. If a friend forwarded you this message, sign up here to get it in your inbox.