Software Multiple Reset

What happened

Brad Gerstner frames a widely discussed software correction precisely: "the multiple correction just took software from a place where they were way more expensive than the market multiple and brought them into the category of the market multiple. Now they're trading at about 22-23x real, stock-comp-included GAAP earnings, about where the market is trading."¹

The repricing logic: software earnings were previously treated as nearly impenetrable and earned a premium multiple as a result. AI introduced the real possibility that a meaningful share of that revenue, three to five years out, gets replaced rather than defended. Markets raised the implied discount rate and compressed the multiple, but only down to the level of the broad market, not below it, and Gerstner's warning is that a distribution of outcomes still sits ahead, with software that stays outside the flow of AI-driven demand likely to trade below the market multiple rather than at it.

The Nvidia comparison

The sharpest illustration in his own words: application software trades at roughly twice Nvidia's multiple, even though "Nvidia is trading about 13x earnings for 70% growth, the most essential thing in AI, and [software is] at twice the multiple."¹ An investor pays roughly 22 to 23 times earnings for application software carrying real replacement risk, against roughly 13 times for the company producing the compute that threatens it, which is the price-action version of a broader argument that the market still over-values the layer most exposed to disruption relative to the layer producing the underlying capability.

The "too hard" basket

Gerstner invokes Warren Buffett's three-basket framework, easy yes, no, and too hard: "for me, software today is generally in the too-hard basket." He has held the view for months and warns explicitly against the temptation, once a correction stabilizes, to think you can outsmart the market, since "you still could be catching a falling knife."¹ The discipline is to decline to underwrite a business whose replacement timeline cannot be genuinely handicapped, rather than to guess at a bottom.

The buyer on the other side of the trade

The most direct challenge to that posture comes from someone actually buying the assets Gerstner is declining to underwrite. Robert F. Smith of Vista Equity Partners reads the identical repricing as an entry point: the 2020 to 2021 run-up "has now come down in enterprise software, so it's been a wonderful buying opportunity, especially as you're transforming businesses to become agentic."²

The disagreement is not about the price, both men agree software got cheaper, it is about whether the replacement question is answerable at all. Gerstner treats the sector as one distribution with an unforecastable replacement date, so the honest response is to decline to handicap it and call the whole category too hard. Smith treats it as three separate distributions and argues the too-hard problem is a failure to disaggregate: companies with genuine ownership of their workflows and data expand into new revenue as agents take over routine tasks, similar companies with cost compression on top re-rate on margin, and companies simply repackaging public data get eaten and, in his own words, deserve to. Two real caveats cut back toward Gerstner's side, though: Smith is a buyer describing his own entry price to an audience of prospective investors in his own fund, and if his three categories were as cleanly separable as he claims, public markets already containing plenty of category-one software companies should be sorting them apart from category-three ones, which is precisely what an undifferentiated sector-wide derating suggests the market currently finds difficult to do.