Cybersecurity stocks fell roughly 8% on average after a single Anthropic research preview. CrowdStrike, Cloudflare, and Okta all slipped. Zoom out and SaaS has lost hundreds of billions in weeks, with some names down 60–80% from their peaks.
This isn’t about one tweet. It’s about expectations. Public markets price what companies can charge five or ten years from now, and the control layer of software just shifted.
For two decades, SaaS monetised seats. Humans log in, click buttons, generate value. Revenue scaled with headcount.
Frontier AI challenges that. If models can review code, detect vulnerabilities, write patches, automate workflows, and replace repetitive knowledge tasks, the economic unit migrates from human seats to computation.
In cybersecurity, many vendors monetise vulnerability detection, scanning, threat analysis, and identity verification. If foundation models excel at static analysis, fuzzing, and reasoning, detection gets cheaper and margins compress.
Investors then ask whether a 40× multiple business becomes a 20× multiple business. That compression alone can erase billions even if revenue hasn’t moved.
The deeper shift: software is moving from workflow tools to autonomous agents. One agent replacing five seats decouples revenue growth from hiring growth, breaking the old scaling assumption.
Markets are forward-looking prediction engines. Every new AI capability triggers the question: does this eliminate an entire category? Valuations reprice on belief, not completion.
AI doesn’t need to replace these companies; it only has to convince investors it might. The discount rate on disruption collapsed. Research previews now move markets like product launches.
The value accrues to whoever owns the model, data, execution environment, compliance, and distribution. Everyone else gets squeezed. This is more than volatility—it’s a repricing of control.