SaaS pocalypse: why AI disruption makes it nearly impossible to get excited about software broadly

Summary

The core tension in SaaS today isn't earnings misses or revenue declines—it's that investors can no longer see far enough into the future to justify the multiples the sector once commanded.

Brad Gerstner, founder of Altimeter Capital, frames the selloff as fundamentally rational. Three years ago, Salesforce traded at 35 times free cash flow because that cash flow was predictable. Now, with AI rapidly improving every quarter, investors face genuine uncertainty about terminal value. Stocks have been cut in half not because companies are missing numbers today, but because the market has put future cash flows into what Gerstner calls "the too hard bucket." The only way that reverses is if SaaS companies can show they're accelerating revenue growth and actually benefiting from AI—not being disrupted by it.

A few names clear that bar. Databricks grew revenue over 60% last quarter. Snowflake and Clickhouse are accelerating because AI models depend on them. But application software—the broad middle of the SaaS market—remains stuck. The problem isn't cyclical. It's structural: AI models will keep getting better, agents will keep improving, and that relentless improvement is a permanent threat to software margins and defensibility.

The practical erosion is already visible. Monday.com, facing guidance concerns tied to AI disruption, represents a larger category problem—seat-based pricing breaks down when a single agent can do the work of many seats. KPMG accidentally announced its own business model was under attack when it tried to force its auditors to accept lower fees because "accounting work can be significantly automated by AI." If an auditing firm believes that, why should anyone else pay for auditing?

The question investors face is not whether SaaS broadly will recover, but whether it can while the underlying threat keeps accelerating. SaaS would benefit from a model plateau—a point where capabilities plateau and companies can deploy them confidently. Instead, the conversation is about whether there's a floor at all, or whether multiples just compress toward minimal free cash flow returns.

Within that broader sell, some patterns emerge. Companies with regulatory moats, customer lock-in, or distribution advantages may prove defensible. Payment rails, for example, are harder to disrupt than workflow software because customers are locked in by bank charters, regulatory requirements, and customer relationships that AI alone cannot replicate. But fragmentation is rising too. With AI commoditizing the cost of building competitive software, small competitors can now raise $10 million and spend a year building a credible alternative to a $1 billion public company that took a decade and ten times the capital to reach scale. Their margins may be 30 percent instead of 60 percent, but that math works for a startup. It doesn't for an incumbent trying to justify public multiples.

Gerstner's view is that the only way to generate excitement about software broadly—not individual companies, but the category itself—is to see evidence that the narrative can durably improve even as models keep improving. That evidence hasn't arrived, and the rate of AI improvement suggests it may not.