Is SaaS Dead? Why That’s the Wrong Question

In short: SaaS as a business model, meaning software delivered as a service with recurring revenue, is here to stay. What’s under pressure is the architecture most SaaS companies grew up on: per-user pricing, customer lock-in through the interface and competition on features. How hard any one company gets hit depends on how it’s built. No study will tell you that. Every executive team has to work it out for itself.

Since spring 2026, I’ve heard the same question in many versions: is SaaS dead? It comes from executive teams, advisory boards and investors, sometimes with concern, sometimes with an undertone of “we saw this coming.” My answer is no. But that’s the less interesting half. The more interesting half: over the next two to three years, it will be decided which software companies come out of this shift stronger and which come out weaker. That decision is made inside each executive team.

What do the numbers say about the future of SaaS?

In July 2026, Gartner put a number on the debate.

$234B

By 2030, up to $234 billion in enterprise software spending will be exposed to what Gartner calls “agentic arbitrage.”

That’s roughly 20 percent of enterprise SaaS spending. The mechanism is simple. AI agents get work done across multiple systems, and people need to operate individual interfaces less and less (Gartner press release, July 1, 2026).

The same source, however, pushes back on the doomsday scenario. Gartner explicitly speaks of a metamorphosis rather than an apocalypse and sees significant revenue opportunities for vendors that enable agentic workflows across business functions. Roughly four-fifths of spending isn’t in the danger zone, and the exposed fifth doesn’t disappear. It moves to other vendors and other services.

The limits of such figures matter just as much. A forecast for 2030 is not a measurement. The Bitkom study “Softwarewelt 2036” is based on qualitative interviews with executives in the German software industry and makes no claim to statistical representativeness. None of these sources makes the decision for an executive team. They describe a direction, not the situation of any single company.

What exactly is under pressure: the business model or the architecture?

The debate suffers from a lack of precision. “SaaS” usually refers to two things that went hand in hand so closely for twenty years that nobody ever had to pull them apart.

One is the business model: software as a service, paid for by subscription, with recurring revenue. The other is the architecturemost SaaS products were built on: designed around the interface, priced per user, developed along a feature list and shipped in fixed release cycles.

It’s mainly the architecture that’s under pressure. I think it’s very likely that a software company will still be paid on a recurring basis for a service five years from now. Whether it can still charge per head, whether its customers still work through its interface and whether its organization will still look the way it does today is an open question.

What unspoken assumption does the SaaS model rest on?

Behind the success of the SaaS model lies an assumption so self-evident that nobody ever wrote it down: the value of software grows with the number of people who work with it. More users mean more value, and more value justifies more licenses.

That assumption has long since moved beyond the price list. It’s baked into sales compensation, which rewards newly sold licenses. Into customer success, which tracks active users. Into product development, which improves the user’s workday. And into the business plan, which projects growth from new users at existing customers. Each of these habits makes sense on its own. But they all hinge on the same point. As soon as software does part of the work itself instead of just helping with it, that point comes under strain.

Which five assumptions hold up a classic SaaS company?

I like to describe a business model as a load-bearing structure. As long as the loads stay the same, nobody needs to run the numbers again. When they change, you have to check which walls are load-bearing. In my analysis, classic SaaS companies have five.

Features secure the lead.With AI, the effort needed to copy features drops sharply. Key question: what about our product is hard to copy?

The user is the unit of value.When agents take over work, the customer needs fewer licenses for the same output. Key question: what is our customer paying for when an agent does the work?

The interface keeps the customer.When agents access systems through APIs, the day-to-day relationship with the user can shift to a third party. Key question: who owns the customer relationship?

Additional growth costs almost nothing.Every AI interaction comes with compute costs. Key question: how do we make money when our marginal costs are no longer zero?

Software is built by large teams along a feature roadmap.A different product calls for a different organization. Key question: is our organization built for the product we want to sell tomorrow?

These assumptions are interconnected. A new pricing model without adjusted sales compensation fails in the field. An in-house agent layer without cost transparency eats into margins. That’s why no single measure is the answer, neither a bolted-on AI assistant nor a quick switch to outcome-based pricing.

How do you turn the findings into decisions?

At the end of the review, there’s a plan for every component. I work with three zones for this.

Keep covers what still holds up under new loads: customer data built up over years, industry expertise, an installed customer base, proven reliability on compliance. Keeping means deliberately protecting these assets and making sure they stay inside your own company.

Rebuild covers what’s showing cracks: pricing metrics, contract clauses that define a user as a natural person, compensation models, APIs. This is the most demanding zone, because it happens while the business is running and almost always takes something away from someone who has gotten used to it.

Build new covers what doesn’t exist yet: agents that do the work instead of supporting it, offerings billed by outcome, services nobody provides today because they would be too labor-intensive for people.

The practical value of this split is that it rules out two comfortable stances: defense (“our numbers are fine”) and demolition (“we have to reinvent ourselves”). Both spare the leadership team the distinction that actually matters.

What should executive teams review now?

  • At your next leadership meeting, ask what share of your revenue would remain if your customers did the same work with a third of the people three years from now. Have each member write down a number on their own first. The spread of the answers is the real finding.
  • Have someone calculate what share of recurring revenue grows directly with headcount at the customer, broken down by customer size.
  • Talk to the person responsible for your standard contracts. What applies today when another vendor’s agent accesses your customer’s data through your API? If the answer takes more than two sentences, you’ve found a gap.

Frequently asked questions

Are AI agents killing SaaS?

No. Software as a service with recurring revenue remains a viable business model. What’s under pressure is per-user pricing, lock-in through the interface and competition on features.

What is agentic arbitrage?

The term comes from Gartner. It describes how AI agents get work done across multiple applications, so people use the interfaces of individual programs less often. As a result, software spending shifts away from user-based licenses.

Which SaaS companies are hit hardest?

The hardest hit are vendors whose revenue depends mostly on user licenses, whose value lies mainly in the user interface and who have no hard-to-copy data or deep industry expertise. Vendors of systems where companies keep their system-of-record data are better positioned.

Further reading: From tool budget to headcount budget: pricing for AI agents · Who owns the customer when the agent does the work? · Managing two curves at once: leading the transformation

Sources: Gartner press release, July 1, 2026 · Bitkom (2026): Softwarewelt 2036

Similar Posts