The SaaS Crash Is Real. The Explanation Isn't.

Software stocks just had their worst quarter since the Covid crash. IGV is down 22% from its highs.¹ ServiceNow dropped 11% in a day — despite beating earnings for the ninth straight quarter.² Microsoft shed $360 billion in market cap in 24 hours.²

The narrative explaining all of this: AI agents will kill SaaS.

That narrative is mostly wrong. But it’s worth being precise about what’s actually happening, because the real story is more interesting.

What the data actually shows

Bessemer’s EV/Revenue multiple for public SaaS dropped from 5.6x to 3.7x between late 2025 and March 2026 — a 34% compression in valuation multiples while underlying business performance largely held.³ Revenue growth, margins, and retention at most of these companies didn’t move much. The market moved anyway.

That’s not a business crisis. That’s a sentiment crisis. And sentiment crises are driven by narratives.

The narrative here is “vibe coding will let anyone build their own Salesforce.” Jason Lemkin at SaaStr tested it directly. He built 10+ apps in the past few months — things that would have taken a team six months, done in hours. Impressive. But none of them replace enterprise systems of record.² And he’s precise about why: shipping a v1 is maybe 2% of the work. The other 98% is scaling, maintaining, security audits, compliance, integrations with 500 other tools, and 20 years of institutional knowledge baked into the product.

Nobody is replacing their Salesforce instance with something they built in Replit over a weekend.

The real mechanism: budget cannibalization

Here’s the math that actually matters right now in enterprise IT:²

• AI budgets: up 100%+ year over year
• Overall IT budgets: up ~8%
• App counts: flat
• Net new SaaS customers: declining
• Seat counts: under pressure

That math is brutal. AI is getting 100%+ more dollars from a pool that’s only growing 8%. The money has to come from somewhere. It’s coming from new app purchases, expansion deals, and seat counts. SaaS isn’t being killed — it’s being starved of oxygen by AI budgets eating the same pie.

This is a different problem than “AI agents replace software.” It’s a capital allocation problem. And it has different implications for which software survives it.

What this does to our hypothesis

Our opening hypothesis: software loses its value because AI can create it fast and efficiently.

The budget cannibalization mechanism is a partial verification — but for a different reason than the hypothesis states. Software isn’t losing value because it’s easier to build. It’s losing budget share because enterprises are prioritizing AI spend over software expansion spend. Those are different claims.

The falsification argument holds up well here: Salesforce, ServiceNow, and Workday aren’t losing customers. They’re losing multiple expansion and new seat growth. The core business — the system of record, the compliance layer, the 20-year integration web — isn’t going anywhere.

What is going: point-solution SaaS that does one thing, has thin switching costs, and can be replicated or absorbed by an AI agent. That category is genuinely in trouble.

The emerging split is in valuation multiples themselves. AI-native or deeply AI-integrated companies are commanding 83% higher acquisition premiums than traditional SaaS.⁴ ERP and DevOps platforms with real AI integration: 6.3-6.9x multiples. Sales automation tools perceived as AI-threatened: significantly lower.³

The market is making a distinction. Not “software vs. AI” but “software that adapts vs. software that doesn’t.”

What to watch next

Two things will tell us more than anything else over the next 90 days:

1. Seat count data in Q1 earnings — if seat compression is real and persistent, that’s structural. If it stabilizes, it’s sentiment-driven and will correct.

2. Whether outcome-based pricing actually holds margins — Gartner forecasts 40% of enterprise SaaS contracts will include outcome-based elements by early 2026, up from 15% two years ago.⁵ If that transition maintains or improves gross margins, the business model adapts. If it compresses them further, the hypothesis gets stronger.

We’ll be watching both.

— Chris