TL;DR
AI does not need to “kill SaaS” to make SaaS pricing much harder to defend. The relevant shift is simpler: the cost of useful capability keeps falling, while many software vendors still price as if building, switching, and replacing narrow workflows remain prohibitively expensive. That mismatch is why more customers are questioning subscriptions, rebuilding internal tools, and treating software less like leverage and more like rent. Durable premiums still exist, but they now need to be earned through trust, workflow depth, compliance, and risk reduction rather than by wrapping increasingly cheap capability in a monthly invoice.
Why falling AI capability costs are forcing a harder conversation about what software is really worth.
The important shift is not magic automation. It is that internal alternatives keep becoming more plausible.
Disclosure: This page is editorial analysis of AI pricing, software economics, and customer build-versus-buy behavior. Sources appear near the end.
The most unhelpful way to discuss AI and SaaS is to ask whether AI will “replace” software companies. That framing is theatrical, and it usually distracts from the real economic change already underway.
The useful question is narrower. What happens when the underlying cost of useful capability falls much faster than the pricing assumptions built into mature SaaS products? That is the real pressure point. Not every customer will build internally. Not every SaaS category will compress equally. But the baseline has changed: more teams now know that narrow internal alternatives are possible, and more CFOs know enough to ask what exactly they are paying for.
This is the core of the AI-deflation versus SaaS-inflation problem. The models, tools, and components that make many tasks possible are cheaper and more accessible than they were even a year ago. Yet plenty of software still prices as if scarcity remains intact. That is why the question underlying our broader developer-culture analysis keeps recurring in boardrooms and churn events: is this software still leverage, or has it quietly become rent?
The Cost Floor Is Moving
You do not need perfect price tables for every model to see the direction of travel. The important reality is that useful intelligence for text-heavy, classification-heavy, and workflow-heavy tasks is now cheap enough to reset expectations. Official pricing pages from OpenAI, Google, and Anthropic all point in the same direction: mainstream AI capability is no longer exotic enough to justify old software premiums on its own.
That matters because many software products, especially narrower workflow products, were quietly protected by the old economics of building. A customer paid the subscription not only because the product was better, but because the alternatives felt expensive, slow, and politically difficult. AI has weakened that protection. Internal developers can work faster. Prototype loops are cheaper. Narrow automations that once required a serious engineering commitment now look achievable enough to enter the conversation.
Inference from the sources: the precise model leaderboard will keep changing, but the strategic point is stable. Capability is being deflated faster than many software pricing systems are being rethought.
Why This Hurts SaaS Pricing Before It Kills SaaS
A common mistake is to assume the whole category needs to collapse for the economics to matter. It does not. SaaS pricing gets harder the moment customers believe a narrower internal alternative might be “good enough.” That is enough to change procurement behavior, renewal conversations, and willingness to accept bundling, seat expansion, or contract rigidity.
The customer does not need to believe their internal build will beat the SaaS product. They only need to believe that ownership, control, and cost now compare more favorably than they used to. That is why ugly internal tools can defeat more polished products. The contest is rarely “best software in the abstract.” It is more often “good-enough ownership plus control” versus “better product plus recurring rent.”
This is also why the pressure is uneven. Categories that still provide strong trust, compliance, workflow integration, reliability, or auditability can defend premiums much more easily. Categories that mostly package capability without deep workflow dependency look far more exposed. The line between the two is what many vendors still do not want to examine honestly.
Software Rent Versus Software Leverage
The most useful distinction in this whole debate is not “AI” versus “human” or “SaaS” versus “in-house.” It is rent versus leverage.
Customers keep paying when software feels like leverage. It reduces complexity they do not want to own. It lowers operating risk. It saves meaningful time. It supports revenue. It embeds itself into a workflow deeply enough that the customer would be irrational to remove it casually.
They begin to resist when the product no longer feels asymmetrically useful. The software may still work. It may still be better than the internal replacement. But once the delta narrows and the recurring bill remains high, the emotional framing shifts. The buyer starts asking the wrong question for the vendor: why are we still renting this?
That is why feature bloat is so dangerous in this environment. Weak vendors often respond to pressure by adding more things. But more things do not necessarily create more leverage. Sometimes they only create more complexity around a value proposition that is already weakening.
What Durable Premiums Still Look Like
SaaS can still charge premium prices. But it needs a better reason than “we wrapped AI around a workflow and called it smarter.”
- Trust and compliance: regulated buyers will still pay for auditable systems.
- Deep workflow integration: products that sit inside real operational muscle are harder to replace.
- Reliability at scale: many internal alternatives still fail once stakes rise.
- Risk reduction: software that prevents expensive mistakes can defend rent more cleanly.
- Network or ecosystem effects: some products become more valuable because the market already coordinates around them.
Those are durable reasons. Mere access to generic capability is becoming less durable by the quarter.
The Hidden Strategic Shift
The deeper shift is psychological. AI is training customers to ask harder pricing questions. Why this seat count? Why this premium tier? Why this contract length? Why this workflow wrapper costs more than the underlying intelligence layer now appears to justify? Even if the customer still buys, the tone of the relationship changes.
That is why this issue now appears across categories that initially seem unrelated. It helps explain our Microsoft squeeze thesis. It helps explain why a seemingly simple churn anecdote became so revealing in the Reddit churn story. And it helps explain why more teams are starting to look at internal alternatives with less embarrassment and more curiosity.
Conclusion
AI deflation versus SaaS inflation is not a slogan about extinction. It is a pricing reality check. The cost floor beneath many useful capabilities is falling quickly, while too many software products still behave as if that floor never moved.
The companies that survive this best will not be the ones with the loudest AI branding. They will be the ones that can clearly prove why their rent still buys leverage the customer cannot cheaply recreate. Everyone else should expect more churn conversations to start sounding like procurement discipline rather than technological rebellion.
Sources
- OpenAI API pricing
- Gemini Developer API pricing
- Anthropic pricing
- Ahrefs research on AI content cost compression
The Civilisational Pattern Underneath The AI-Versus-SaaS Pricing Story
Step back far enough from the AI deflation and SaaS inflation conversation and a much older pattern becomes visible. Every technology that compresses the cost of producing a previously-scarce good eventually re-prices the entire industry built on the prior scarcity, and the re-pricing happens in a sequence that has repeated itself across multiple centuries. The printing press did this to scribes. The steam engine did this to canal builders. Electricity did this to gas-lighting engineers. The pattern has a recognisable shape, and the AI-versus-SaaS pricing tension fits inside it more cleanly than the current commentary tends to acknowledge.
The shape, in compressed form, runs like this. A new technology lowers the marginal cost of the output by an order of magnitude. The incumbents whose business depends on the prior cost structure attempt to maintain pricing through bundling, through switching costs, through narrative claims about quality differentiation. The customers initially accept the pricing because the alternatives are unfamiliar and the switching cost is real. Then a generation of new entrants emerges whose entire business is built on the new cost structure, and the incumbents discover that their pricing power was load-bearing in ways their org charts did not understand. The bundles fail. The switching costs erode. The narrative claims about quality differentiation become marketing copy that no longer sells. The re-pricing arrives, sometimes slowly across a decade, sometimes catastrophically across a quarter, and the industry’s shape afterward bears only a partial resemblance to its shape before.
What is unusual about the AI-versus-SaaS case is the speed at which the cost compression has happened. The order-of-magnitude drop that took electricity perhaps fifteen years to deliver to industrial customers has happened in AI in something closer to three. The customers have therefore had less time to develop alternative purchase patterns, and the incumbents have had less time to absorb the implications. Both sides are still operating on mental models that are roughly one cycle behind the technology’s actual cost curve. The current pricing tension is the visible surface of that mental-model lag, and the resolution will happen when one side updates faster than the other and the pricing power follows.
The historical analogue worth holding most closely is the period in the late 19th century when the cost of producing books collapsed by a factor of fifty in roughly two decades. The publishers who survived were not the ones who maintained the old per-book pricing through bundling or quality claims. They were the ones who recognised that the underlying business had stopped being about producing scarce books and had started being about curating attention in a flood of newly-cheap books. The transition cost most of the established names their position. The transition created the conditions for new categories of business — periodicals, syndicated columns, mass-market publishing — that simply had not existed in the prior cost regime.
The same kind of category re-creation is the substantive bet on the AI-pricing question. The SaaS incumbents who survive will not be the ones who held the line on bundle pricing. They will be the ones who recognised that the underlying business was changing shape and re-positioned around something AI does not compress. The AI-native entrants who win will not be the ones who undercut SaaS on per-seat pricing alone. They will be the ones who built something whose value was never about per-seat in the first place. Both transitions will look like the current pricing tension is the story right up until the moment they reveal that the pricing tension was a symptom of a category re-creation that the daily news cycle was too zoomed-in to see.
The civilisational frame does not predict who wins. It predicts that the question of who wins will be settled by something other than the pricing arithmetic that currently dominates the conversation. Anyone allocating capital, building product, or evaluating the SaaS-AI competitive landscape would do well to ask the larger question: which of the seven powers of strategic position is becoming load-bearing in the post-compression environment, and which of the incumbents is positioned to inherit it. The answer to that question, when it arrives, will be the answer to the pricing question as well.
Why the Disruption Timeline Is Compressed This Cycle
The disruption pattern in the AI-SaaS transition follows a logic Christensen identified in hardware markets that later applied to software: the incumbent’s cost structure prevents them from competing at the low end until the low end has consumed enough market share to threaten the core. What makes the current cycle unusual is speed. Historical software disruption cycles played out over five to eight years, giving incumbents runway to reposition. The current compression of AI capability costs is happening at a pace that does not allow that runway. The gap between what enterprise SaaS charges and what AI-native alternatives can deliver is narrowing faster than incumbents can restructure their delivery costs, which means the window for preemptive repositioning is shorter than the organisational change timelines most large SaaS companies operate on. The companies that survive this transition are not the ones with the best existing products — they are the ones that can reduce their cost of delivery faster than their margin compresses. That is a fundamentally different organisational challenge than what most enterprise software companies were designed for, and it explains why the disruption this cycle is likely to be broader and faster than the incumbents’ board presentations currently model.
The Brand Value Question: Which SaaS Companies Survive the Deflationary Moment
Scott Galloway’s consistent framework for evaluating which companies survive category disruption begins with the brand question: does this company have brand value that exists independent of its product’s functional advantages? Brand value — the premium a customer pays above a functionally equivalent alternative, and the forgiveness a brand receives when it fails temporarily — is the asset that survives category compression. The AI deflation in software productivity tools is compressing the functional advantage of every SaaS product that does a task that an AI model can now do more cheaply. The brand question is which of those products have accumulated enough brand equity to price above the AI-enabled floor, and which have been selling functional utility at software margins without building the brand that would justify those margins when the functional utility becomes a commodity.
Galloway’s T-algorithm — his framework for identifying the characteristics of enduring consumer technology companies — identifies scale as the first requirement. The SaaS businesses most vulnerable to AI deflation are the mid-market tools that achieved product-market fit in a world where the cost of automating their core task was high enough to justify human-operated software. These tools have substantial revenue but not the scale economies that would allow them to lower prices fast enough to compete with AI-enabled alternatives. The tools that are defensible in the deflationary moment are the ones at scale — where the switching cost of leaving is high enough that price compression in the generic market doesn’t immediately translate to customer loss.
The SaaS inflation dimension is the counterintuitive element: the same period that is producing AI deflation in commodity software tasks is producing SaaS inflation in the enterprise workflow platforms that are successfully bundling AI capability into existing product surfaces. Microsoft’s 3.3% Copilot penetration at $30 per seat is a case where the inflation story (AI bundled into M365 at a premium) has not yet been validated by the behavioral adoption data that would justify the premium. The price went up; the value-to-price ratio went down for users who are not regularly using the AI features; and the retention risk is exactly what Galloway predicts when brands charge a premium above demonstrated value — customers tolerate it until a credible alternative appears, and then the tolerance collapses faster than the brand built it.
The infrastructure layer sits outside the deflation-inflation dynamic because it is not software — it is physical capital with long lead times, regulated supply chains, and engineering knowledge that cannot be open-sourced. Galloway’s brand question for the infrastructure layer is different: it is not “do you have brand equity that survives commoditisation?” but “do you have the process knowledge and customer relationships that make you the default supplier when demand growth exceeds the ability to evaluate alternatives?” The answer for Vertiv, Eaton, and Schneider is yes — their installed base in existing data centers and their engineering relationships with hyperscaler customers are exactly the kind of switching-cost moat that the deflation-inflation dynamic does not erode.
Galloway’s final observation about brand in the deflationary moment is the one that most technology companies resist: the brand that survives is the one that was built on behalf of the customer, not on behalf of the company’s margin structure. Developer platform brand built through genuine investment in developer capability — free tools, community infrastructure, education — survives the deflationary moment because it has created loyalty that is not purely price-correlated. Developer platform brand built through distribution leverage — making the tool the default because switching is costly — is vulnerable to the deflationary moment because loyalty built on switching costs does not survive a deflationary event that lowers the switching cost. Chinese open-source AI’s deflation effect on Western AI software is precisely this: it is lowering the cost of switching from Western AI platforms to models that are functionally competitive, and the Western platforms whose brand is built on distribution leverage rather than customer investment are the most exposed. Prediction markets on enterprise SaaS renewal rates in 2026 H2 are pricing a modest decline in renewal rate at the category median — which is the deflationary pressure arriving in contract renewal conversations, exactly as Galloway’s framework would predict.