Is AI about to empty the (half full) software glass?

Plenty has been written about the recent collapse in valuations

This collapse seems to have been triggered by fears that rather than software “eating the world”, AI will instead “eat software” (to plagiarise Marc Andreessen). There is a furious debate going on as to whether AI will have a seriously damaging effect on all incumbent software providers, or only on some of them, and if so, which ones; what the effects will be; and how long they will last.

We think it is too early to conclude that the whole software sector has been permanently downgraded from always-growing stock market darlings, to low-growth or ex-growth “also-rans”, but  the recent collapse in valuations has certainly increased the level of doubt around what has been a highly-valued segment of the market. The downward re-pricing of the sector has been substantial: for example, the Bessemer Venture Partners Nasdaq Emerging Cloud Index (which includes industry titans such as Salesforce, Servicenow and Adobe, as well as smaller, faster growing but less profitable companies such as Snowflake and Uipath) has fallen by about 25% since January 7th.

The debate is swirling between the “glass half full” side and the “AI is going to empty the glass eventually” side. The former claim that the incumbents will adapt, and in fact may even benefit from new AI technology. This group claim that AI will enable them to code more cheaply and efficiently, adding value to their software products, which cannot easily be replaced or sidelined as they embed unique ‘system of record’ data, core work processes, experience, market infrastructure and trust. Also, like the other major evolutions in tech such as the internet, cloud and mobile computing, which created new use cases and demand for computing power, connectivity and data, AI will eventually grow the whole market, rather than cannibalise existing vendors. And incumbents should be best placed to participate in that growth.

The gloomier set say that AI (particularly agentic AI) will enable customers to take more software development in-house, disintermediating the incumbents, or at the very facilitating a pushback on pricing. This could force a transition away from subscription-based “number of seats” models to usage-based pricing - which is less predictably recurring and therefore less valuable.

A simplistic LBO model quantifies the impact of one scenario over the other. Imagine a theoretical software buy-and-build, backed by ‘Megafund’, a fictional private equity group.

Until recently, a reasonable set of assumptions for the buyout model would include 15% recurring organic growth and the ability to bolt on smaller companies growing at a similar pace, perhaps at slightly lower multiples, reflecting lower cash generation. An entry multiple of 25x EBITDA, supported by 7.5x senior debt, and add-on acquisitions at 18x EBITDA and similarly leveraged, would generate 5-year returns for Megafund of 21%, assuming the same EBITDA multiple of 25x at exit.

Now let’s tinker with just a few of the assumptions above. After the initial acquisition of the platform asset, let’s assume “the AI effect” drives the organic growth down to 5% a year, for both the platform company and the bolt-ons. Let’s be generous and assume that cash EBITDA margins are not affected – maybe because the company can use AI to code more cheaply or reduce marketing costs using AI-powered CRM systems, or otherwise reduce operating costs. Let’s also assume that the (AI-spooked) credit committees at the senior lenders get a bit uncomfortable with leverage levels of 7.5 times and require that new acquisitions can only be financed at 5 times. However, the owners of the smaller add-on businesses, who already feel that they were being acquired at a discount, accept a small reduction but refuse to sell out at anything less than 15x.

Now the model shows that Megafund’s IRR drops to 12%. Note that this still assumes that the exit multiple remains 25x EBITDA, like the entry multiple – a highly questionable assumption, given the less bullish 5% growth and lower leverage multiples.

In order to preserve IRR’s at an acceptable level, it is inevitable that in-prices have to change. The million-dollar question is, of course, by how much. Interestingly, if you assume that an acceptable IRR for a large buyout is 15%, you can generate this with a 7.5% organic growth rate in our example with both the entry and exit multiple at 15 times EBITDA.