Is SaaS dead? Why big companies are opting to build their own in-house

No, SaaS is not dead. Businesses are on track to spend about $1.44 trillion on software in 2026, up 15.1% from last year¹. That is not what a dying industry looks like. Still, "is SaaS dead" keeps getting asked for a reason: the build-versus-buy math that held for twenty years has shifted, and some closely watched companies are now building tools in-house that they would have bought without a second thought in 2020. (Figures are accurate as of publication; treat fast-moving specifics as a snapshot.)

The short version: SaaS (software you rent by subscription) is not dying. The undifferentiated per-seat subscription, a generic tool rented forever for one generic job, is what's under pressure.

What did Klarna actually do when it "shut down" its SaaS?

Klarna is the story everyone cites, and most people cite it wrong. In 2024, CEO Sebastian Siemiatkowski told analysts the fintech had shut down Salesforce, would shut down Workday within weeks, and was "shutting down a lot of our SaaS providers, as we are able to consolidate." The internet heard "AI replaced Salesforce."

Six months later he clarified what really happened². Klarna did not swap its CRM (customer relationship management software) for a chatbot. It consolidated scattered corporate data into an internal knowledge stack built on a graph database, then used AI coding tools to spin up its own interfaces on top of that unified data. In his words: "So no, we did not replace SaaS with an LLM." The license savings were "nice," he said, but the real prize was standardizing the company's knowledge in one place.

Two details matter. Klarna is a bank with a deep engineering culture, and even it needed a serious internal platform effort, not a weekend of prompting. And Siemiatkowski's own prediction is not the death of SaaS but consolidation: fewer vendors doing more, absorbing the capabilities of the tools they replace.

Why is building in-house suddenly realistic?

Because AI collapsed the cost of the top layer of software. Interfaces, workflows, internal dashboards, glue between systems: things that used to take a dev team a quarter can now be produced in days with AI coding tools, which is exactly how Klarna deployed new interfaces on its internal stack. A bought tool serves everyone's workflow approximately. A built tool serves yours exactly, and for the first time "exactly" is affordable.

There is an honest boundary, though. What got cheap is the application layer. What did not get cheap is the infrastructure underneath: systems that must be correct every single time, under load, with an audit trail (a verifiable record of every change). Anyone who has tried vibe coding a point of sale has met this wall. The demo takes an afternoon; then inventory that survives two simultaneous sales, reports that reconcile (match to the penny), and card payments turn out to be a different class of problem.

What should you still buy instead of build?

Buy anything where a silent error is catastrophic. Payment processing and PCI compliance (the card industry's security standard), payroll tax tables, double-entry ledgers, certified card-present hardware. These fail rarely, invisibly, and expensively, and no tool built in a sprint should own them. Even executives at the big SaaS vendors concede that companies have built homegrown HR and finance systems for decades²; the argument was never that building is impossible, it's that specialized infrastructure is somebody's full-time job.

So the pattern big companies are converging on is a split: build the layer where your business is different, buy the layer that has to be correct every time. Klarna built knowledge tools and interfaces. It did not build its own card network.

What does this mean for a small business?

You are living the same problem at a smaller scale. A typical merchant stack is a POS subscription, a loyalty app, a scheduling tool, and a marketing platform, each with its own monthly fee and none of them fitting quite right. You cannot hire a platform team the way Klarna did. That is why the interesting shift in commerce software is platforms splitting themselves along exactly the line above: sell the infrastructure (payments, inventory, reporting) and open the top layer for you, or an AI working on your behalf, to build. Final runs on that model: the commerce infrastructure is the product, the software on top is prompt-based and yours to shape, and there is no monthly software subscription. Expect more platforms to follow as AI agents become retail software's second user.

So, is SaaS dead?

No. Software spending is growing faster than almost any other corporate line item, and the companies "killing" SaaS are mostly consolidating it into fewer, deeper platforms. What is dying is renting a generic tool for a job your own AI-built software could now do better, and paying per seat for it forever. The rule of thumb: build what makes you different; buy what has to be correct every time. To see what that split looks like at the checkout counter, start with the difference between a POS with AI and a POS an AI can build.