AI hasn’t killed SaaS — but it has removed its illusion of inevitability.

The SaaS premium didn’t disappear overnight. It began to erode when inevitability stopped being priced as certainty.

For most of the past decade, SaaS felt structurally unstoppable. Recurring revenue. Negative working capital. High gross margins. Expanding TAM. The Rule of 40. Forward multiples of 20x, 30x, even 60x were not anomalies — they were framed as the mathematical outcome of compounding subscription economics in a low-rate world. In 2021, the BVP Nasdaq Emerging Cloud Index peaked at roughly 20x forward revenue. Analysts spoke of a “SaaS premium” as confidently as they spoke of gravity.

Today, that logic is being repriced. And the repricing is not just about rates. It’s about AI — and a deeper shift in how markets think about inevitability.

The question is no longer whether software grows — but whether the economic model that powered its premium survives AI intact.

The End of the “Recurring Revenue Premium”?

Recent market data tells a stark story. US software forward P/E ratios have fallen sharply from their post-pandemic peaks and now trade near decade lows for many categories. The BVP Cloud Index has given back the majority of its 2020–2021 gains in real terms. Meanwhile, companies that were once viewed as structurally superior — pure-play SaaS names with high NRR and negative churn — now trade on mid-teens earnings multiples that look increasingly like those of industrial businesses.

This isn’t merely a macro unwind. Rate normalisation explains some of the compression. It does not explain everything. The narrative itself has changed.

The evidence is accumulating across categories. Klarna described reducing its headcount from 7,000 to below 3,000 — without raising additional capital — because AI enabled it to ship more with fewer people. Duolingo announced in early 2024 that it was replacing contract translators with AI, framing it explicitly as a structural workforce decision rather than a cost-cutting measure. Salesforce, in its fiscal 2025 results, acknowledged that some customers were scrutinising seat-based pricing models in light of AI productivity gains — a previously unthinkable conversation for the company that invented the subscription seat.

These are not isolated anecdotes. They point to a common mechanism. If AI enables faster product iteration, lower engineering headcount, reduced customer support cost, and automation across back-office functions, then the scarcity value that underpinned traditional SaaS operating leverage begins to shift. The old model assumed specialised human capital was the bottleneck. AI challenges that assumption directly — and when the bottleneck moves, the premium attached to solving it moves too.

Systems of Record vs Systems of Intelligence

The SaaS boom was built on systems of record — Salesforce for CRM, Workday for HR, NetSuite for finance. They were sticky because switching costs were high, data migration was painful, workflows were deeply embedded, and integrations created inertia. The genius of the early SaaS model was that it turned workflow dependency into a durable revenue stream.

But as agentic AI evolves, an uncomfortable question emerges: if AI agents sit above systems of record and orchestrate workflows across tools, where does the value migrate? Does the system of record become commoditised infrastructure — the plumbing — while the system of intelligence captures incremental economics?

Markets increasingly behave as though intelligence — not workflow — will be the new centre of gravity for valuation.

This is not a hypothetical. Microsoft’s Copilot strategy is explicitly an attempt to insert an intelligent orchestration layer across Office, Dynamics, and Azure without fully cannibalising the underlying revenue base. ServiceNow, Salesforce Agentforce, and HubSpot’s Breeze are all variations of the same defensive move: embed AI deeply enough that the system of record becomes a system of intelligence.

Markets are beginning to price the difference between companies that make that transition successfully and those that do not — selectively and unevenly, but in the right direction.

The Switching Cost Inversion

Klarna’s CEO made a subtler point worth dwelling on: “The next thing that’s going to hit everyone bad is the switching cost of data.” Historically, switching costs protected incumbents. Data lived inside a vendor’s system, formatted to that vendor’s schema, queryable only through that vendor’s API. Leaving meant losing history, or paying eye-watering migration costs.

In an AI world, this may invert. Data portability, API abstraction, and model fine-tuning are reducing dependency on any single workflow layer. If an AI agent can ingest structured and unstructured data from multiple sources and dynamically reconstruct context, UI-level stickiness weakens considerably.

The moat shifts. Workflow embedding and interface familiarity matter less. What matters more is proprietary datasets that cannot be easily replicated, embedded AI usage that generates compounding feedback loops, ecosystem control at the platform or infrastructure level, and continuous model learning that improves with scale.

In boardrooms, this shift shows up less as panic and more as a slow repricing of certainty — the recognition that defensibility may no longer sit where it used to.

The TAM Illusion

One of the most underappreciated themes in the SaaS repricing is how thoroughly Total Addressable Market narratives shaped the boom — and how quietly that lens is being discarded.

For years, expanding TAM justified expanding multiples. Every workflow was framed as a category waiting to be “software-ised.” Analysts would accept a 30x revenue multiple, partly based on a slide showing a vast addressable market that the company had barely begun to penetrate.

The problem with TAM-led valuation was always that it was circular. Markets defined themselves by ambition rather than economic reality. When tools become easier to build, replicate, or layer onto existing infrastructure, TAM stops behaving like a linear expansion story.

We are moving from TAM-led valuation to earnings-durability-led valuation. The question is no longer “how big could this get?” but “how certain is this cash flow in five years?” That is a profound psychological shift for an industry that spent a decade rewarding optimism.

Ambition used to be about expanding multiples; durability now defines them.

Why This Is Not 2022 Again

It is tempting to frame this as another post-zero-rates hangover. But 2022 was primarily about discount rates. The mechanism was straightforward: rates rise, long-duration cash flows get discounted more heavily, growth stocks de-rate.

2026 is different. The question is no longer whether future cash flows should be discounted at 5% or 10%. The question is whether those cash flows will be realised at their assumed magnitudes.

Capital has already begun to reflect this. Rotation toward industrials, defence, commodities, and physical asset plays is not simply a value trade. It reflects a view that companies with tangible assets, regulated moats, or geopolitically driven demand are more durable than those whose moats depend on the continued relevance of a software interface.

Software used to be the disruptor. Parts of it are now viewed as potentially disrupted.

Where Multiples Likely Settle

Rather than a single “SaaS multiple,” markets are beginning to differentiate across three tiers.

Commodity Workflow SaaS — low differentiation, minimal AI moat, limited pricing power — likely settles in the 10–15x earnings or 3–6x revenue range.

Embedded, Data-Rich Platforms — mission-critical integration, proprietary datasets, high switching friction derived from data rather than workflow — likely sustain an 18–25x earnings multiple or a high single-digit revenue multiple.

AI-Native Infrastructure and Enablement — companies that enable compute, orchestration, or critical AI infrastructure — will attract premium valuations but with significantly higher volatility.

The key shift is from blanket software optimism to layered discrimination.

The Rule of 40 Isn’t Dead — It’s Incomplete

The Rule of 40 once functioned as a clean shorthand for balancing growth and profitability. AI compresses both sides of that equation simultaneously in ways it was not designed to capture.

Growth may decelerate as AI lowers barriers to competition. Margins may simultaneously expand through automation. A company could score well on the Rule of 40 while its competitive position quietly erodes.

Increasingly, the lenses I find most useful are shifting away from classic SaaS dashboards. I care less about seat expansion and more about economic flow: how much revenue is tied to AI execution, how much growth comes from consumption rather than users, whether proprietary data compounds advantage, and which layer of the stack captures value regardless of which applications win. The market is slowly learning to analyse software less like a subscription business and more like an economic system.

The Intangible Asset Paradox

For two decades, research consistently showed that intangible-heavy businesses outperformed. Software companies dominated those screens.

Yet today, markets are favouring tangible assets while repricing intangible-heavy software. The explanation is not that intangibles have lost value. It is that AI has introduced uncertainty around which intangibles will compound and which will be rendered obsolete.

When duration becomes uncertain, markets shorten it. That is rational. It may also be where opportunity begins.

What Happens Next?

Three plausible scenarios frame the medium-term path.

Scenario A — AI Accelerates SaaS Profitability

Automation expands margins faster than revenue compresses. Multiples stabilise and modestly re-rate.

Scenario B — SaaS Becomes Utility Infrastructure

Growth slows structurally. Valuations converge toward industrial-style earnings multiples.

Scenario C — A New Layer Captures the Value

AI orchestration platforms are becoming the premium asset class, while legacy SaaS trades at a persistent discount.

The market today appears to be pricing a combination of B and C, with Scenario A as the recovery path for companies that earn it.

The Core Question

SaaS is not dead. But the era of automatic premium valuations is over.

The question for 2026 is no longer “Is this company SaaS?” It is:

“Where does it sit in the AI value stack — and how defensible is its data?”

AI hasn’t killed SaaS. It has forced the market to decide which software businesses are products — and which are infrastructure for economic activity.

Inevitability once drove multiples. In 2026, proof does. Have a great week!