AI Is Breaking SaaS Pricing, and Turning Software Back Into a Service (Pricing Series, Part 3)

SaaS made software a product. AI cost models will force a change in pricing orthodoxy that has held true for 20 years.

TL;DR: AI software is cheap to build but expensive to serve. Buyers expect it to be free and interchangeable. This combination destroys pricing power—so you need to earn it through context, guidance, and trust.

"If we want things to stay as they are, everything must change."
— Il Gattopardo, Giuseppe Tomasi di Lampedusa

In professional services, pricing power has never come from inputs. It comes from something deeper: contextual expertise, the ability to guide, and the trust that you can deliver outcomes in uncertain environments. As I argued in Part 2, clients don’t pay for activity. They pay for knowing what not to do. They pay for judgment.

AI is changing everything about how software is built, priced, and delivered. But that doesn’t mean the underlying logic of value creation is changing. In fact, the opposite is true. We are entering an era where software must once again justify its value like a services firm would. Not through features alone, but through embedded understanding, opinionated guidance, and real-world outcomes. In other words, everything about the delivery model is changing—so that the value logic can stay the same.

This is what most AI-first startups are missing. They’re reaching for SaaS-era pricing models (subscriptions, usage-based metering, seat licenses) without recognising that the thing they’re selling has a fundamentally different cost structure, differentiation model, and customer psychology.

To win in this environment, you need to stop asking how to price the product and start asking how to earn the right to charge at all. That means acting less like a software vendor, and more like a trusted expert.

AI Kills the SaaS Margin Model

For twenty years, the economics of software followed a remarkably stable formula. Build something expensive once. Sell it cheaply, repeatedly. Over time, scale eroded your marginal cost to near-zero, allowing you to reap compounding profits on a fixed base of engineering effort. You charged a subscription not because customers loved predictability, but because it smoothed your payback window and built margin into the machine.

AI blows that up. Every GenAI query costs something. Often a lot. There is no marginal cost collapse. There is no “write once, serve infinitely” model. AI is probabilistic, compute-intensive, and context-specific. It does not scale like SaaS. It scales linearly. More users, more queries, more tokens, more cost.

Worse, the most valuable use cases are the most expensive to serve. Summarizing legal documents, generating code, orchestrating customer journeys—these tasks demand deeper context, larger context windows, and longer sessions. All of that drives up cost.

So the AI-native unit economics are the inverse of SaaS: cheap to build and expensive to serve.

The Buyer Has Been Trained to Expect Free

The market doesn’t just misunderstand this. It has been actively taught to expect the opposite. Amusingly, the AI ecosystem did this to itself. OpenAI and Anthropic publish per-token pricing. Wrappers flood Product Hunt with free trials. Product-led go-to-market playbooks teach founders to ship fast, hook users, and charge later, if ever.

The result is a widespread belief that AI is infinite, swappable, and cheap. That it’s just API dust sprinkled on top of existing workflows. That value is downstream of clever prompt engineering, not embedded capability or integration depth that exists far before any customer gets their hands on an interface.

But we’re hitting an unpleasant moment in the AI hype cycle (particularly painful when valuations are sky high) - no one wants to pay for what they've been told is free.

The Infrastructure Trap

A big part of the problem is that AI startups have misread the game they’re playing. Too many believe that if they act like infrastructure and sell on being stable, foundational, low-level, they will become infrastructure. And that as a result, pricing power will come later, at scale.

But the actual infrastructure is already spoken for. AWS, GCP, Azure, OpenAI—these players already control the bottom of the stack. And they do not share margin. They charge you per token, per millisecond, per call, and they dictate the rules.

Make no mistake. If you are building an AI-first business, you are not building infrastructure. You are renting it, and if you price like infrastructure, you are hoping that value somehow trickles down. Competing here is like trying to outrun gravity. The more usage you win, the more cost you incur. The better your product works(to stay ahead of the competition, and earn more usage), the more it bleeds margin. Unless you have true scale advantages or proprietary models (and politely, you probably don’t), you are playing someone else’s game with their unit economics.

And yet, usage-based pricing has, thus far, been the preferred monetisation model. This is a trap.

Usage Pricing Looks Great But It Will Kill You

Usage-based pricing feels like the natural fit for AI. It aligns with actual compute cost. It makes billing transparent. It scales with adoption. And in principle, it keeps the business healthy by tying revenue to unit economics.

But logical is not the same as strategic. In practice, usage-based pricing often leads AI startups into a dangerous space. The moment you expose your cost structure—per call, per token, per workflow—you give your buyer a baseline. And if your competitors are pricing on the same basis, the buyer can now compare everyone on a single dimension: cost per unit of work.

This is fine if you're the cheapest. But if you're not, the conversation shifts from value to justification. Why are you more expensive than Alternative X? Why should we pay more tokens per call when the outputs look similar?

In commoditized markets, usage-based quickly becomes cost-based. And when buyers don’t perceive meaningful differentiation, they push you toward the lowest available rate. It’s the logic of infrastructure, without the scale or defensibility to sustain it.

The result is a race to the bottom. Every efficiency gain is passed through to the buyer. Every improvement invites comparison. Every token counted is a token questioned. This is why usage-based pricing is necessary but not sufficient. You need it to align with variable cost, but it cannot be the only lever. Because if your pricing is perfectly correlated with your inputs, your margin becomes a coin toss.

To escape the squeeze, you need something else: something that makes the buyer want to pay more. That means differentiation not just in performance, but in proximity. It means context. It means trust. And most of all, it means seriousness.

How to Escape the Pricing Trap: Relearn the Art of Embedded Value

In Part 2 of this series, I argued that professional services firms earn pricing power not from time spent, but from three things: contextual expertise, the ability to guide with a strong point of view, and the seriousness to deliver outcomes in uncertain environments.

AI-first software companies now have to rediscover exactly the same logic.

In a high-choice, low-trust market, where dozens of tools can wrap the same foundation model and ship similar features, buyers don’t pay for functionality. They pay for confidence. They want to know the tool they pick is grounded in real understanding. They want to avoid mistakes, delays, and rework. And they want someone who knows what not to do as much as what to build.

The most successful software companies in the AI era are already operating this way. Stripe helps enterprise customers redesign their entire pricing logic. Figma embeds customer success engineers to shape internal design systems and workflows. Palantir deploys entire teams to co-build with clients inside high-stakes environments. These are not professional services firms. But they all behave like trusted experts: opinionated, integrated, and accountable.

This is where contextual expertise becomes the foundation for price. When your product doesn’t just work, but works correctly for that customer, in that situation, with their workflows, their constraints, and their risks—you earn the right to price beyond usage.

The product itself is only part of the value. The rest comes from how it embeds into the customer’s world. How the AI model can be customised to bend itself around customer-specific workflows, how customer success, strategy and support teams navigate decision-making structures, and data environments. That is what creates lock-in. That is what justifies pricing that goes beyond compute. And that is what turns an AI feature into a business-critical system.

This is not about turning every AI startup into a consulting firm. But for AI companies, the lesson is clear, albeit ironic: do not sell a tool. Sell a transformation. Sell contextual expertise and confidence. The companies that command margin in this market will be the ones that make themselves indispensable, not through exclusivity, but through embeddedness in their customers’ worlds.

Rethinking GTM and Valuation Models

This shift doesn’t just change pricing. It rewires the go-to-market model, the funding model, and the way we measure progress. If you are building AI-native software, you cannot rely on CAC payback heuristics designed for SaaS. You cannot assume 80 percent gross margin. You cannot expect high ARPU from a product that looks like a browser plugin.

Your compute costs are linear. Your revenue is not guaranteed. Your buyers are skeptical, hard to reach and have been taught to be so price sensitive that $1 feels like too much. And your stack is unstable. So stop benchmarking to SaaS. Start asking: where does margin accrue in this market? Where does trust accrue? Who actually owns the customer relationship?

The answer is rarely at the infrastructure layer, where ‘trust’ means ‘it doesn’t break’, and is table stakes. It is almost always higher up, where software stops being a tool and starts becoming a habit, a dependency, a silent partner. That is the shift AI forces. Not just in technology, but in business model, pricing, value creation, and how you help customers define success.

Felix

P.S. If you are interested in how AI is reshaping growth models, pricing structures, and go-to-market execution, the Market-Led Growth book goes much deeper. It lays out a full framework for operating in this new era. Subscribe to the newsletter to follow along.