In February 2026, software stocks lost roughly $1 trillion in market value in under two weeks [1]. Earnings didn't fall, as forward estimates for most of these companies were still rising [2]. The loss came as the market was repricing a mistaken belief that recurring software revenue was close to risk free.
That belief is gone. The question every investor now must answer before underwriting a software deal is, “Does this company own a system of record, or is it a system of reference?” [3]
Systems of record hold the data a business runs on, including the CRM entry that triggers a commission, the compliance record a regulator will ask for, and the ledger that reconciles at close. They carry high switching costs, because switching is genuinely risky [4]. Foundation models can sit on top of these systems, but they struggle to replace them.
Systems of reference surface or reformat information that lives somewhere else, often through dashboards or lightweight retrieval tools. An AI agent doesn't need to disrupt these directly. It can route around them. That's roughly what happened when a single foundation model update triggered an 11%+ single day drop in two well-known horizontal software names [5]. Without a decline in overall product quality, the market now understood that systems of reference could be replicated.
The valuation data backs this up. Vertical workflow-embedded software companies, like Datadog, are trading in the 13x revenue range [6]. Horizontal tools with seat based pricing and thin workflow integration have compressed into the low single digits on revenue [7]. One prominent sell side desk found that horizontal software fell nearly 35% over a recent 5-month window, against roughly 24% for vertical software [8].
The smartest capital is betting that software’s value is shifting. One recent estimate puts a fresh $100 billion opportunity on the table for software that helps orchestrate work across systems rather than inside just one single system [9]. Diligence must go deeper than a product demo.
Start with the seat count. What share of ARR is tied to per-seat pricing, and has anyone modeled what happens if AI agents take over 30% of that headcount? [10]
Then, run the substitution test. Hand the product's core prompt to a frontier model directly. If a technical user gets most of the way to the same output, the interface and the brand are real. The interface and the brand can provide a head start, but they do not constitute a moat.
Look at the data next: not whether it exists, but whether it compounds. A static dataset is an asset, but a dataset that improves with every interaction is a moat.
Last, press on switching costs until you know what they're made of. Regulatory exposure and integration depth hold up under pressure. Habit and inertia don't, and AI is good at dissolving both.
Underwriting still comes down to judgment, but that judgment is only as good as the questions behind it. Today, most diligence checklists haven't caught up to what the market already knows.
Sources:
Note: this list is illustrative rather than comprehensive. It draws on a mix of financial newswire reporting (Reuters, Mizuho), sell-side equity research (Seaport, William Blair, Needham, ICICI Securities), management consulting research (Bain), and independent M&A/valuation analysis (Houlihan Lokey, Development Corporate). It represents a sample of the source types underlying the article's claims rather than a full citation trail.
Reuters, "US software stocks slammed on mounting fears over AI disruption, lose $1 trillion in week," February 2026
Seaport Research Partners, "Our 100 Favorite Charts: July 2026"
Bain & Company, "Will Agentic AI Disrupt SaaS?" (2025 Technology Report)
ICICI Securities Limited, "INTERNET: Decoding NEW-AGE TECH in India," May 2026
Mizuho Securities USA, "On the Frontier: OpenAI & Anthropic, and the Consumer and Enterprise Implications," March 2026
William Blair Research, "Datadog: Unified Observability Platform With Expanding Opportunities for the AI Era," May 2026
Needham & Co., "MNDY Bull/Bear Debate; At the Core of the AI Eating Software Narrative," June 2026
Houlihan Lokey, "Artificial Intelligence and Software Valuations: A Framework for Assessing Enterprise Values," March 2026
Bain & Company, "SaaS' next $100 billion opportunity could come from agentic AI," May 2026
Development Corporate, "AI SaaS Investment Trends Are Flashing Red," citing TechCrunch VC investor survey, March 2026