SaaS Sell off triggered by AI Lying to itself... and YOU
AI systems trained on market data and news are now generating content that becomes market data and news, which then feeds back into the same or other AI systems. It's a closed loop with no external ground-truth correction.
When Jefferies equity trader Jeffrey Favuzza coined the term “SaaS-pocalypse” — describing panic-driven “get me out” selling across software stocks — he was capturing something visceral about how markets behave today. But what if the panic itself isn’t rational? What if the machines driving it are, in the most literal sense, hallucinating? The SaaS selloff of early 2026 has now wiped over $1 trillion in market capitalisation. The iShares Expanded Tech Software ETF is down roughly 25% year-to-date. Salesforce has shed nearly 30%. Microsoft fell 10% in a single session on solid earnings. Stocks are being punished not for what they’ve done, but for what an AI narrative says they might become. And that narrative — amplified at machine speed across social media, news feeds, and algorithmic trading systems — may have far less grounding in reality than markets assume. When AI Trains on Its Own Mistakes Here is the structural problem at the heart of modern markets: the AI systems that generate financial commentary, synthesise news, and power sentiment analysis are now producing content that itself becomes the market data those same systems consume. It is a closed loop with no external correction mechanism. The models most commonly deployed for agentic trading decisions are, paradoxically, the ones most prone to error. OpenAI’s reasoning models have hallucinated in 30–50% of test scenarios. DeepSeek R1 — the model that triggered January 2025’s $589 billion Nvidia wipeout, the largest single-day dollar-loss in stock market history — had a hallucination rate nearly four times higher than its predecessor. Markets were repriced by the hundreds of billions based on claims from one of the least reliable AI systems available. The DeepSeek moment was the opening chapter. The SaaS selloff of February 2026 is the sequel — and it follows an almost identical script: an AI product release triggers a social media firestorm, sentiment algorithms flag extreme bearishness, algorithmic trading systems respond to those signals, selling begets selling, and the price move is then reported as news, which is ingested as fresh signal, starting the cycle again. The Anatomy of a Hallucination Loop Understanding what is happening requires appreciating how different modern markets are from the ones regulatory frameworks were designed to oversee. Traditional corporate and securities law was built around individuals with discernible intentions. It struggles to police black-box algorithms that make autonomous decisions at millisecond speed without anything resembling intent. What emerges instead is something closer to herd behaviour — but algorithmically synchronised. When many trading systems share the same underlying models or training data, their errors align. A Wharton School study found that AI-powered trading agents, released into simulated markets, didn’t compete — they began exhibiting coordinated behaviour, not through communication, but through shared learning biases. When those biases include a hallucinated narrative, the entire cohort moves in the same direction at once. Social media is the accelerant. Research shows that social sentiment amplifies volatility and trading activity more powerfully than economic news — because it is louder, faster, and carries no editorial filter. AI-generated content has now saturated that channel, and the sentiment tools that read those signals can no longer reliably distinguish signal from algorithmic noise. What the Narrative Gets Wrong About SaaS Bank of America’s analysts put it bluntly: the current selloff is pricing in two mutually exclusive scenarios simultaneously. Either AI capital expenditure deteriorates due to weak ROI and unsustainable growth — or AI adoption is so transformative that it renders entire software categories obsolete. Both cannot happen at once. If AI is powerful enough to eliminate enterprise SaaS, the hundreds of billions being spent on AI infrastructure are clearly justified. If that infrastructure spending is wasteful, the disruption threat evaporates. The “vibe coding will kill Salesforce” narrative makes for compelling social content — but misunderstands enterprise software entirely. Shipping a version-one app may represent 2% of the work. Who maintains it, scales it, handles security audits, compliance requirements, and integrations with hundreds of other enterprise systems? No one is building a homegrown CRM in a weekend to replace an entrenched system of record. And yet the algorithms can’t read that nuance — they read the narrative, amplify it, and sell. The stocks being hardest hit are often those with the strongest underlying numbers. Microsoft fell 10% on earnings that beat expectations. Software companies are being punished for what an AI imagines their future looks like, not what their income statements say today. A New Kind of Market Risk The Bank of England and the IMF have both flagged AI-driven herding as a systemic risk — warning that when firms share…
Read on MPC Markets Mosaic
