The "SaaS-pocalypse" is good clickbait, but it's lazy analysis
The sell-off is real. But for SaaS companies with genuine moats, this is exactly the kind of entry point you've been waiting for
The "SaaS-pocalypse" is good clickbait, but it's lazy analysis The sell-off is real. But for SaaS companies with genuine moats, this is exactly the kind of entry point you've been waiting for The SaaS sector has now shed over $1 trillion in market capitalisation since the start of 2026. The iShares Expanded Tech Software ETF ( NYSE: IGV ) is down roughly 25% year-to-date. Salesforce ( NYSE: CRM ) has lost nearly 30%. Microsoft ( NASDAQ: MSFT ) fell 10% in a single session on what were, by any measure, solid earnings. The market’s verdict is swift, simple, and almost unanimous: AI code agents are about to commoditise every software product on the planet, and no valuation multiple is safe. However, this sell-off is analytically lazy. And it’s being driven, at least in part, by the very technology it fears hallucinating on its own research 1-year chart of SaaS companies and ETFs: IGV ETF ( NYSE: IGV ) -25% Aussie Tech ETF ( ASX: ATEC ) -34% Servicenow ( NYSE: NOW ) -46% Wisetech ( ASX: WTC ) -61% When the market feeds on its own mistakes Here’s the issue. The AI systems generating financial commentary, synthesising news flow, and powering sentiment analysis are now producing content that itself becomes the data those same systems consume. It’s a closed feedback loop with no external correction mechanism. The numbers are striking. OpenAI’s frontier reasoning models have hallucinated in 30–50% of test scenarios. DeepSeek R1 — the model that triggered January 2025’s $589 billion Nvidia single-day wipeout, the largest dollar-loss in stock market history, carried a hallucination rate nearly four times higher than its predecessor. Markets repriced by hundreds of billions on the back of outputs from one of the least reliable AI systems available. The SaaS sell-off of early 2026 follows an almost identical script. An AI product release triggers a social media firestorm. Sentiment algorithms flag extreme bearishness. Algorithmic trading systems act on those signals. Selling begets selling. The price move gets reported as news. That news is ingested as fresh signal. The cycle restarts. What took days in 2016 now takes minutes — and when the inputs are hallucinated, the repricing can be catastrophic and completely untethered from fundamentals ( more details here ). We believe the difference this time is that investors have the opportunity to look through the noise and identify the SaaS businesses where the structural moats are not just intact, they’re actually widening. AI is lying to itself....and you. The moat was never in the code Let’s be clear: AI has genuinely destroyed certain competitive advantages in software. Tools like Cursor and enterprise-grade agent frameworks can replicate the majority of standard SaaS functionality within days. Traditional defences built on elegant code, polished UX, or rapid iteration cycles are under real pressure. We don’t dispute that. But the companies being indiscriminately sold are often those whose actual protection was never in the codebase to begin with. The durable moats live outside the software entirely, in proprietary data rights, regulatory licences, institutional relationships, deep workflow embedding, and sustained frontier research. None of these can be prompt-engineered into existence. Proprietary private data is the most powerful flywheel. Longitudinal transaction histories, sensor streams, clinical outcomes, and behavioural graphs cannot be scraped or synthesised by frontier models. Bloomberg’s terminal data, Palantir’s ( NASDAQ: PLTR ) government datasets, Veeva’s ( NASDAQ: VEEV ) pharmaceutical records — these aren’t features, they’re the product. Every new user enriches the dataset; every richer dataset sharpens the AI inside the platform. General-purpose models simply lack the consent, the legal right, and the accumulated context to replicate this. Regulated access and compliance are even harder to replicate. FedRAMP, ITAR, HIPAA, ASIC, CHESS, SWIFT — none of these licenses can be generated by a language model. AI can write compliant code, but it cannot sit in a three-year audit cycle with a government agency, secure institutional relationships, or accept legal liability for a breach. Switching costs in these environments routinely run into the tens of millions and span multi-year implementation timelines. That’s not a feature moat — it’s an institutional one. Add deep workflow embedding to the mix and the picture becomes clearer still. When a SaaS platform is the system of record inside core banking, hospital EHRs, or government case management, replacement isn’t a technical decision, it’s an organisational trauma. Staff retraining, data migration, permission re-architecture, and regulatory re-certification make a rip-and-replace approach impractical, even when a cheaper AI-built alternative exists on paper. The SaaS revaluation was already happening and current prices are no where near 2021's frothy prices Who has a Moat? In this wire, we’ll highlight the…
Read on MPC Markets Mosaic
