Is AI destroying SaaS?
In a world where generative AI can increasingly do the work, what will happen to Saas, asks Shamillah Bankyia
The pricing of public software has undergone a drastic repricing. Over the last few weeks, investors have dumped a trillion in value across multiple software names and sectors, and begun to reckon with a more uncomfortable idea: the terminal value of classic software as a service (SaaS) might look radically different in a world where generative AI can increasingly do the work rather than merely help humans do it. As a result, software is trading at an average multiple of around 6x EV/LTM revenue, versus around 20x in 2021, a 70 per cent compression, with horizontal software compressing the most to around 3x multiples.
We’ve been investing in enterprise software since 2007 and today have a portfolio of almost 100 software companies. It is worth starting with a simple truth. Software, and SaaS in particular, has been a fantastic asset class. It grew at roughly 20 per cent year on year for the last two decades, and it did so with economics that were unusually attractive: low churn, high gross margins and the ability to grow with customers over time. While SaaS grew at around 20 per cent historically, it is now growing at around 12 per cent – an eight percentage point decline. Importantly, a lot of that slowdown began before any recent AI catalyst became an easy headline.
Why SaaS worked so well for two decades
The ROI case for SaaS was clear. Instead of buying physical servers and spending billions in capex and operational costs to run your own software, you could rent the service and focus your business on what mattered most. You could amortise costs and deploy modern tools far faster than most in-house IT teams could build. The friction to adopt better software fell dramatically, and even if switching was never as easy as we all liked to imagine, the direction of travel was obvious.
On top of that, software products were just better. Having a group of people focused on solving a single enterprise problem meant they built better products. Multi-tenancy meant improvements for one customer could be rolled out across all customers, and products could get better and better. Dashboards looked better. Information was democratised across organisations. Collaboration was made possible by software.
In theory, this meant companies were more productive and could focus their efforts on growing the most strategic components of their businesses, primarily revenue. With such significant ROI, buying SaaS was a no brainer. An army of SaaS companies was born, with around $1 trillion of enterprise software spend last year alone.
The dirty secret of SaaS growth
The dirty secret of SaaS is that part of the historic magical growth was contractual. Long-term contracts let you bake in growth every year for the lifetime of the contract, often in the range of 3-5 per cent via price uplifts and escalators – the best companies could raise prices by as much as 10 per cent annually! That meant that to deliver the historical growth rates the market came to expect, companies did not need to find all of it from scratch each year. There was a base layer of contracted expansion, and then there was the work of finding the rest through broader deployment inside the customer, selling additional capability into the base and adding new customers.
For many categories, the incremental engine was seat-based. But seat expansion is hard to come by once you have reached peak internal penetration, and it is especially hard in a world where headcount is not expanding quickly – Fortune 500 companies have had less than one per cent growth in number of employees over the last few years. Companies also improved their go-to-market, optimising processes, finding accelerator partners and tightening conversion. That deepens your penetration, but it does not magically grow the market you serve. At the same time, shifts in interest rates meant that growth at all costs was no longer rewarded, and software budgets became a line item that CFOs were suddenly willing to interrogate.
The most elite SaaS companies found ways around these limits. They did not rely on seats and pricing escalators forever. They consistently chased market expansion. They expanded their product lines, moved into adjacencies, found new buyers, created new categories, either organically or inorganically. In our portfolio, Collibra expanded what it could do and who it could sell to through a steady cadence of new products and platform capabilities. Dataiku did the same, broadening its footprint and increasing the set of problems it could credibly solve for the enterprise.
Generative AI has challenged the top line, even for the best
Generative AI has now challenged the top line of even the best SaaS companies, not because SaaS is dead, but because the mechanisms of compounding are being questioned. We are all marvelling at the growth of AI-native companies across the stack including AI hardware winners like Nvidia. The reality, though, is that enterprise budgets have yet to grow meaningfully. As we wrote last year, we remain early on enterprise ROI for generative AI.
Even public market leaders like Datadog and Servicenow, who are ruthlessly rolling out new products, have not seen significant topline expansion yet. In the case of Datadog, some of the growth has come from AI-native customers themselves. Datadog cites 12 per cent of their revenue as AI native (and growing 17 per cent quarter on quarter and up from six per cent 18 months ago), and Servicenow’s AI product, Now Assist, now has $600m of annual recurring revenue. However, AI revenues remain small versus broader revenues at many incumbents, so it is hard to see that growth shine through immediately in the top line.
In practice, enterprise budgets have not grown enough to fund everything at once, so the pressure shows up first in the marginal decisions: expansion seats get questioned, nice-to-have tools get consolidated and new multi-year commitments get delayed while buyers wait to see what AI makes possible.
Is this a cycle, or a structural change?
The key question is how long this lasts, and what exactly is changing. There is no doubt in our minds that generative AI will be pervasive in the enterprise. The unknown is the pace of transformation and, more importantly, what happens to the unit of value that software monetises. SaaS was priced on the assumption that software scales with headcount; AI introduces a credible world where value scales without headcount.
None of this guarantees the end of elite software businesses. Our view is that part of the current slowdown is cyclical. Enterprise buyers are waiting to see what AI can bring. There is no sense in committing to huge amounts of new SaaS spend in the short term when there is even a small risk of it being considered legacy technology in a few years. But this caution is building technical debt. Deferred upgrades, delayed migrations and ageing integrations do not go away, they compound. And software is only 1-2 per cent of revenue for Fortune 500 companies, so the immediate instinct is unlikely to be ripping out highly performant software that is difficult to replace.
History suggests enterprise transitions take time. The move from mainframe to SaaS took 25 years, and it is still happening (IBM mainframes still generate billions in revenue and are growing, a reminder that enterprise transitions are slow). So elite software companies likely have time to catch up, but not the luxury of moving slowly. Early data from Datadog and Servicenow suggests incumbents who move fast can absorb AI rather than be disrupted by it. They need to aggressively hunt every AI dollar in their use case and their customer base. They need to quickly understand whether the problem they are solving will exist in a completely agentic world. And they need to land AI native winners as customers, as the Fortune 500 could look different five years from now.
Where AI will genuinely disrupt, rather than enhance, is in categories where good SaaS alternatives never fully emerged. Software development is the clearest example: AI coding assistants are already reshaping how code gets written and reviewed. Customer service is another, where AI agents can handle interactions that previously required both software and headcount. In these greenfield categories, AI is not competing with entrenched SaaS, rather it is filling gaps that SaaS never closed. The winners in this next era will be the companies that do the hard things their customers cannot do themselves, and tie that capability to measurable value. SaaS is not dead. But the game has changed, and the companies that recognise that fastest will be the ones that define the next decade of enterprise software.
Shamillah Bankyia is a partner at Dawn Capital