Software Company Valuations: What Has AI Really Changed?

How a software company is valued today, and where AI fits in: multiples with sources, the real premium, and what the buyer looks at before making an offer.

Those of us who work in corporate transactions are seeing a situation this year that, at first glance, doesn't add up. Never have so many software companies been bought: 2025 closed as the year with the most deals on record, with 2,698 acquisitions of SaaS companies —subscription software firms— worldwide, 28 percent more than the year before. And at the same time, the median multiple for listed software hit a more-than-a-decade low. More appetite than ever, reference prices colder than they've been in years. Both at once.

The short explanation has two letters: AI. The full explanation takes a little longer, and it's worth having, because it affects anyone who owns a software company, is thinking of selling one, or is considering buying one. That's what this article is about: how a software company is valued today, what artificial intelligence has done to those valuations, and where the price is really decided when buyer and seller sit down at the table.

How a software company is valued (the method hasn't changed)

Let's start with the basics, because once the basics are clear, everything else makes sense. A software company is valued, fundamentally, like any other: by its capacity to generate profits in the future. To avoid arguing about the future blindly, the market uses shortcuts called multiples. You take a metric from the company —usually revenue or EBITDA (operating profit, before interest, taxes, and amortization)— and multiply it by a number that captures what is being paid for comparable businesses. If companies similar to yours change hands at four times their revenue, there's your first reference for what yours might be worth.

In software, that number rewards one thing above all others: recurring revenue. Billing ten million by selling projects one at a time is not worth the same as billing it through subscriptions that renew on their own. That's why the SaaS model enjoyed multiples higher than the rest of the sector for a decade. And that's why buyers watch the churn rate so closely —the share of customers who cancel each year. A multiple is, at bottom, a bet on how long that revenue will last.

There are frameworks that capture the bet. The most cited is the Rule of 40: the sum of annual growth and profit margin should exceed 40 percent. It still works as a yardstick, and the market still pays for it: according to the Aventis Advisors series, at the end of 2025 each ten-point improvement in that rule was associated with slightly more than an additional point of revenue multiple.

None of this has been touched by AI. The methods are the same, and multiples are still the language in which offers are written. What AI has changed is the variable that feeds the entire calculation: confidence that today's revenue will still exist five years from now.

What has happened with the numbers

Let's put figures to that loss of confidence, because it's one of the few times a change in mindset can be measured this clearly.

The median multiple for listed SaaS went from 7.3 times revenue at the start of 2025 to 3.4 times in March 2026. Less than half, in little more than a year. An aggressive discount "due to fears of AI disruption." In the private market —the one for deals that don't make the papers— the fall was gentler: from a median of 3.8 times revenue in 2025 to 3.1 at the start of 2026. Our M&A report on the software and IT services sector, with data through the end of 2025, describes growing consolidation: industrial and technology groups leading deals, private equity active through build-up strategies, and interest concentrated in recurring-revenue models, vertical ERP, and cybersecurity.

The stock-market episode even has a name of its own. In early February 2026, the new AI agent capabilities unveiled by Anthropic acted as the catalyst for a massive sell-off in software shares that the financial press christened the SaaSpocalypse. There were aftershocks for weeks: in a single April session, Cloudflare fell 12 percent, Snowflake 9, and ServiceNow 7. The sector's cumulative correction since the start of the year has been put at around two trillion dollars in market value, according to financial-press tallies. The share-seller's reasoning was simple: if AI agents can do the work of applications, who will pay for software subscriptions five years from now?

And yet, while share prices were falling, the M&A market was setting records. Counting nearly 2,500 enterprise software acquisitions in 2025, its all-time high, with 147 billion dollars in disclosed value. Private equity —the funds that buy companies to grow them and sell them on— took part in close to 60 percent of SaaS deals. There were mega-deals with AI as the central argument, such as Alphabet's 32 billion dollars for Wiz or Palo Alto's 25 billion for CyberArk. And there were, in parallel, acquisitions of plain old software at control prices: Thoma Bravo took Dayforce, the workforce-management platform, private for 12.3 billion dollars, paying a 32 percent premium. Its founder, Orlando Bravo, capped the sequence this very week with a phrase unusual for a cautious investor: the SaaSpocalypse, he said, "is over," and AI is an enormous tailwind for software.

So where does that leave us? Does AI sink the value of software or send it soaring? The figures in those deals are a long way from the typical Spanish software company, but the question weighs just as heavily on a company with six million in revenue as it does in Silicon Valley.

The premium that exists and doesn't exist at the same time

Both, and it's not wordplay: it depends on which market you're looking at.

At one extreme, companies whose product is AI itself. The median revenue multiple for AI companies stood at 24.2, with extreme cases far above that, while the general listed-SaaS index closed 2025 at 4.8 times. These are the multiples that make headlines. They should be read with a caveat that the source itself provides: much of that figure comes from financing rounds, not full company sales, and anyone selling their AI startup "will probably receive a significantly lower multiple in the acquisition offer."

At the other extreme, the market in which the vast majority of software companies operate, the Spanish ones included: the middle market. There, the premium for "using AI" no longer exists. Almost a third of the companies bought in the previous 30 months incorporated AI, and their median valuations had converged with those of the rest. Acquirers have come to see AI as an inherent part of software, not as an extra to be paid for separately. Only companies with genuine integration, measurable efficiencies, and sustainable monetization attract the buyer's full interest.

The translation is direct. Saying AI no longer moves the multiple; slapping a layer of AI on the product for the photo —what the industry calls AI-washing— doesn't either. Proving it does. And the word "prove" has three specific surnames, on which the sources and our own experience in deals agree: proprietary data that competitors cannot replicate, a presence in a critical customer process, and the tool's ability to execute tasks, not just suggest them.

Not every company absorbs that discount the same way, and that's where the real dividing line lies. Three profiles hold up better. The one with high switching costs, with the software so embedded in the customer's operation that ripping it out would cost months and money, so it isn't replaced even when an alternative exists. The one that rests on proprietary data that no competitor —not even someone who can now build a substitute with AI tools— is able to reconstruct. And the one that lives in the depth of a specialized workflow, where the value is not the code, which is already replicable, but knowledge of the process. At the other extreme, software that merely digitizes simple, standardized tasks carries the steepest discount: it's exactly what a customer can rebuild in-house. The buyer, whether we see it in the offer or not, is sorting every business along that scale.

The two tables

Everything above comes to a head at the decisive moment: the negotiation. And there we see a scene repeat itself that explains this market better than any statistic.

On one side of the table, the seller arrives with the headline multiple in mind. They've read about the 33 times revenue, or they remember the 2021 multiples, and they anchor their expectations there. On the other side, the buyer arrives with a new analysis tucked under their arm. Bain, which has assessed more than a thousand companies in due-diligence processes (the in-depth review of the business that the buyer commissions before closing), describes it bluntly: most acquirers acknowledge that an AI-risk analysis has led them to walk away from a deal. Its framework classifies each business into three categories according to what AI can do to it: revolution, when it threatens the entire model (less than 10 percent of cases); transformation, when it demands substantial changes; and augmentation, when it adds efficiency without redefining anything (around half of cases).

And here is the nuance that, in our experience, explains more breakdowns in a negotiation than any other: that analysis almost never appears in the offer as a line that says "AI discount." It comes in disguised. It's in the churn the buyer projects for the coming years, in the growth scenarios it accepts or trims, in which companies it chooses as comparables, in how much cash it requires the business to withstand without growing. The seller sees an offer lower than expected and doesn't always know where the cut comes from. It comes from there: from doubt about the durability of their revenue.

In Europe, moreover, the table has had a new question for some time now. The European AI regulation, the AI Act, is already part of due diligence in technology deals: legal advisers verify how the business's AI systems are classified by risk level and request specific contractual guarantees on compliance. In our experience it doesn't usually kill deals, but it adds homework, and homework left undone is paid for in price or in terms.

What a seller can do (and what a buyer looks at)

The good news is that almost everything that decides the outcome at that table can be worked on before sitting down at it.

For anyone thinking of selling, preparation begins before the process, ideally one or two years out. It consists of building, with your own data, the answer to the question the buyer is going to ask anyway: why this revenue will keep existing. Having churn and retention measured and documented, rather than estimated on the fly. Being able to show what data the business accumulates that no one else has, and which customer process would break if the product disappeared tomorrow. If the product incorporates AI, being able to show measurable efficiencies and not slides. A company that arrives at the table with that story assembled answers the questions before they hurt, which is the most elegant way to defend a multiple. And on the "when": with a market buying at record pace but choosing with a magnifying glass, the right moment is set less by the calendar than by the state of that preparation.

For the buyer, the exercise is the exact mirror image: separating the businesses with momentum of their own from those that were merely riding the sector's wave. Academia offers the same reading. Christopher Stanton, a professor at Harvard Business School, argues that AI's impact on software will be uneven: the risk is concentrated in tools that digitize simple flows, the ones a customer could rebuild internally with assisted-coding tools, and far less in specialized platforms with critical data and processes. Buying a business in the first category on the cheap can turn out very expensive. Paying well for one in the second can be the best deal of the decade.

There's a final reading that strikes us as the fairest to this moment in the market. Median multiples have come down, it's true. But the median is a point, not a destination: what has moved is the distance between what's paid for defensible software and what's paid for replaceable software. The market keeps buying software, and at a record pace; what it has stopped paying for is the label, and what it pays for better than ever is the substance. For the owner of a software company, the question worth money has stopped being where multiples are trading and has become another: how long will what your business does well last, and what can you show to prove it. Whoever has a good answer will find, today as always, buyers willing to pay for it.

Mario Senra, Partner at NORGESTION