The AI Bubble Nobody Sees Coming

When people talk about an “AI bubble,” they picture something familiar. A dot-com-style crash with stock tickers in freefall. We’ve seen the film before, everyone loses their money except for a few very smart people everyone was betting against.

But what if the real AI bubble doesn’t follow the script at all? What if, instead of bursting because the technology fails to deliver, it bursts precisely because it succeeds?

The Weekend CRM Problem

The global Software-as-a-Service market was valued at approximately $399 billion in 2024 and is projected to nearly double to $819 billion by 2030, according to Grand View Research. It’s one of the most reliable growth stories in modern business. Companies like Salesforce, HubSpot, and ServiceNow have built empires on a straightforward value proposition: building enterprise software is expensive and complicated, so let us do it for you at a fraction of the cost, charged monthly.

Now imagine a small business owner, Sarah. She runs a 15-person marketing agency. She currently pays around £1,000 a month for HubSpot, plus several hundred more for a project management tool, a customer support platform, and a handful of integrations to stitch them together. Her annual SaaS bill sits comfortably north of £16,000.

One Saturday morning, Sarah sits down with an AI coding agent. She describes her workflows in plain English: “I need a CRM that tracks client communications, automates follow-up emails based on deal stage, and gives me a dashboard showing pipeline value by month.” By Sunday evening, she has a working system. It’s not as polished as HubSpot. It doesn’t have 500 integrations. But it does exactly what Sarah needs, and it cost her a weekend and an AI subscription.

This is not a hypothetical. The 2025 Stack Overflow Developer Survey found that 84% of developers now use or plan to use AI tools in their development process, with 51% of professional developers using them daily. According to Index.dev, around 41% of all code written in 2025 was AI-generated. We have crossed a threshold where building functional software is no longer the exclusive domain of professional developers with years of training.

And this is where things get interesting. Not for Sarah, but for every SaaS company whose business model depends on software being too hard for Sarah to build herself.

The Equation That Breaks

The traditional SaaS equation is elegantly simple. Building software is expensive, so you spread development costs across thousands of customers and charge each one a fraction of what custom development would cost. But AI is collapsing the numerator of that equation, the development cost, towards zero. When the thing that made SaaS valuable (the difficulty of building software) disappears, the entire pricing structure comes under existential pressure.

Think of it in layers. The development layer, months of developer time historically costing £300,000 to £1.2 million to build from scratch, can now be compressed into days or weeks at near-zero marginal cost. The maintenance layer, security patches, bug fixes, infrastructure management, is partially threatened, though production systems at scale remain complex. The integration layer, API connections to hundreds of other tools, is still valuable but not impregnable. And the data and network layer, the algorithms trained on billions of interactions, the ecosystems of consultants and apps, remains the strongest moat.

But here’s the thing about moats: they hold until they don’t. And the water is rising faster than most incumbents appreciate.

We’re already seeing the early tremors. The average company used 106 SaaS applications in 2024, down from 112 in 2023, according to BetterCloud’s annual State of SaaS report. That might seem like a small shift, but the consolidation rate dropped from 14% to just 5% year-over-year. Companies aren’t simply swapping one tool for another. They’re starting to question whether they need certain categories of tools at all.

The Deflationary Spiral

Here is the chain reaction that should keep SaaS executives awake at night.

AI tools allow businesses to build custom alternatives to off-the-shelf SaaS products. This forces SaaS companies into downward pricing pressure. Lower pricing means lower revenue per customer. Lower revenue means reduced ability to invest in the large engineering teams that built the product in the first place. Reduced engineering investment means the product stagnates. A stagnating product accelerates the shift to AI-built alternatives. And the cycle deepens.

This isn’t a theoretical spiral. In “The AI Review Tax,” we looked at the employment data: programmer roles fell 27.5% in two years, while junior hiring at the 15 biggest tech firms dropped 25% in a single year (IEEE Spectrum, December 2025; Stack Overflow Blog, December 2025). Those aren’t abstract statistics. Many of those lost roles were at SaaS companies already feeling the squeeze. When your customers can build their own tools and your engineering team is shrinking, the business model comes under pressure from both directions simultaneously.

Marc Benioff, Salesforce’s CEO, announced in early 2025 that the company would stop hiring new software engineers, citing AI-driven productivity gains and shifting investment towards sales roles instead (Salesforce Ben, December 2024). When the company that essentially invented modern SaaS tells you it doesn’t need to hire engineers any more, something fundamental is shifting.

The Uncomfortable Truth About What SaaS Actually Is

Strip away the branding and the marketing and the quarterly earnings narratives, and most SaaS products are, at their core, a database with a decent user interface. The layers are straightforward: a data model, CRUD operations (create, read, update, delete), business rules, a front-end interface, and cloud infrastructure. AI can now generate competent versions of every single one of these layers.

What AI cannot easily replicate, at least not yet, are the moats that entrenched SaaS companies have spent decades building. Trust and brand reputation: Salesforce handling your customer data carries implicit guarantees that a weekend project does not. Regulatory compliance: SOC 2, HIPAA, and GDPR certifications cost millions to obtain and maintain, and for many industries, they’re non-negotiable. Ecosystem effects: when your entire industry uses Slack, the switching cost isn’t technical, it’s social. And institutional knowledge: 10,000 companies working out the kinks over a decade creates a product that handles edge cases no AI-built tool would anticipate.

These moats are real. But they protect different segments of the market very differently. Enterprise SaaS, where contracts are multi-year, switching costs are enormous, and compliance requirements are strict, is substantially more defensible than the long tail of SMB-focused tools. The CRM for Sarah’s 15-person agency is vulnerable. The CRM underpinning a Fortune 500’s global sales operation is not. At least not yet.

The critical question for the industry is: how big is the vulnerable segment? The answer, I suspect, is larger than the incumbents want to admit. Roughly 80% of SaaS products are, by any honest assessment, generic tools that solve common problems with competent but unremarkable execution. They’re the ones in AI’s crosshairs.

The Snake That Eats Its Own Tail

This is where the bubble argument comes into focus. The conventional wisdom about an “AI bubble” assumes one of three burst mechanisms: the technology fails to deliver, the industry runs out of training data, or regulation intervenes. All three are plausible. None of them is the most likely.

The mechanism I believe is most dangerous, and most overlooked, is that AI works exactly as promised, and in doing so, it systematically destroys the business models of the very companies funding its development.

Consider the chain. SaaS companies are among the largest buyers of AI infrastructure and tools. They integrate AI to improve their products, reduce costs, and stay competitive. But the same AI capabilities they’re purchasing are simultaneously empowering their customers to build alternatives. As customers defect or demand lower prices, SaaS revenue falls. Falling revenue leads to reduced AI spending. Reduced AI spending hits the revenue of AI companies themselves. The snake eats its own tail.

And there’s a secondary loop. If AI-driven efficiency means SaaS companies need fewer employees, as we’ve already seen with the employment data, then those laid-off workers reduce overall consumer and business spending, further dampening demand across the economy. This isn’t just a technology disruption. It’s a potential deflationary engine.

Scale that pattern across thousands of companies and you begin to see the outline of something that looks less like a traditional bubble and more like a structural compression of the knowledge economy.

What Klarna Teaches Us About Speed

If you want to see what happens when a company goes all-in on AI-driven efficiency, look at Klarna. We covered the full arc in “The AI Review Tax,” but it’s instructive for a different reason here.

Between 2022 and 2025, Klarna shrank from roughly 5,000 to 3,000 employees while doubling revenue to £720 million quarterly (Computer Weekly, October 2025). That’s a success story for shareholders. But it’s also a preview of what the deflationary spiral looks like from the inside: fewer employees, lower operating costs, higher margins, and 2,000 fewer people spending money in the economy.

Now multiply Klarna by every mid-sized SaaS company making similar calculations. The World Economic Forum’s Future of Jobs Report 2025 warned that 40% of employers expect to reduce staff where AI can automate tasks. A Resume.org survey of 1,000 US business leaders found that four in ten plan to replace workers with AI by 2026. These aren’t fringe predictions. They’re mainstream corporate planning.

The timing question is crucial. If this transition takes ten or more years, SaaS companies have time to adapt, lower prices gradually, and shift business models. Messy but survivable. If it takes two to three years, it’s potentially catastrophic. Companies cannot restructure fast enough, and the deflationary spiral gains momentum before the market can adjust.

Current signals suggest we’re on the faster timeline. But enterprise sales cycles of 12 to 24 months, multi-year contracts, institutional risk aversion, and genuine switching costs will all act as brakes. The transition won’t be instant. But it may be faster than anyone holding SaaS equity is comfortable with.

The Historical Pattern That Offers Hope

Before we spiral into full-blown techno-pessimism, it’s worth remembering the counter-intuitive lesson that history keeps trying to teach us: when technology makes something dramatically cheaper, we don’t do less of it. We do astronomically more of it.

The open-source movement of the 2000s was supposed to destroy commercial software. Instead, the cloud happened, and the commercial software industry grew larger than anyone imagined. Low-code tools in the 2010s were supposed to eliminate developers. Instead, they expanded the total market for custom software by making it accessible to people who would never have built anything before.

What if AI-enabled software development follows the same pattern? What if the result is not fewer software products but a world where every dentist has a dozen custom internal tools, every restaurant has a bespoke ordering system, and every freelance consultant has software infrastructure rivalling a small company? In that world, the total addressable market for software explodes, even as price-per-unit crashes.

This is what I’d call the Expansion Paradox: HubSpot’s price might drop, but if its potential customer base grows from 100,000 to 10 million businesses, the maths might still work. Might. The key uncertainty is whether SaaS companies can reinvent themselves fast enough to capture that expanded market before the deflationary spiral erodes their ability to invest.

The Three Scenarios

If I had to map the most likely outcomes, I’d sketch three.

Scenario one: The Deflationary Cascade. Over 2026–2027, early adopters build custom tools with AI agents. By 2028–2029, word spreads and SMBs abandon SaaS subscriptions at scale. By 2030, SaaS companies face a classic Innovator’s Dilemma: they cannot lower prices without destroying their business model, but they’re bleeding customers to free or near-free alternatives. Mass consolidation follows. Only the most entrenched survive.

Scenario two: The Bifurcation. Consumer and SMB SaaS gets decimated. Enterprise SaaS survives through switching costs, compliance requirements, and institutional inertia. A new category, “AI-first SaaS,” emerges at a fraction of legacy pricing. Traditional SaaS companies become zombie businesses: profitable enough to persist, unable to grow.

Scenario three: The Expansion Paradox. AI makes software so easy that the total volume of custom software explodes 100-fold. Every company builds dozens of internal tools they never would have before. But they still need SaaS for compliance, audit trails, cross-company collaboration, and industry-specific expertise. Prices fall, customer counts soar, and the market restructures around volume rather than margin.

My best guess is that we get a blend of all three: the long tail of generic SaaS gets decimated (scenario one), enterprise holds but bifurcates (scenario two), and the overall market eventually expands once the disruption settles (scenario three). The order matters, though. Destruction comes first. Expansion comes later. And the gap between the two is where the pain lives.

The Ultimate Irony

There’s one more twist worth considering. The AI companies themselves may face the same dynamic they’re imposing on SaaS.

If AI agents can build software, why do we need multiple competing AI providers at premium pricing? Once AI capability is commoditised, through open-source models, local compute, and competitive pressure, the entire stack compresses. We might end up with hardware providers like NVIDIA at the bottom, utility-scale compute in the middle, a handful of differentiated AI companies with genuine moats, and everything else becoming free or near-free.

Anthropic, OpenAI, Google. They’re building tools that could eventually make their own products less valuable. Not immediately. Not this year. But the trajectory is there, and it’s the same trajectory they’re imposing on everyone else.

What This Means for Business Leaders

If you run a company that depends on SaaS products, and in 2026, that means virtually every company, this analysis suggests three practical implications.

First, audit your SaaS stack with an honest eye. For each tool, ask: could an AI agent build a “good enough” version of this in a week? If the answer is yes, that tool’s pricing power is eroding whether its vendor acknowledges it or not. This doesn’t mean you should replace it tomorrow. But it means you should negotiate harder at renewal and begin experimenting with alternatives.

Second, think carefully about where you want to be on the build-versus-buy spectrum. For the past two decades, “buy” was almost always the right answer for non-core software. AI is shifting that calculus. For processes that are genuinely core to your business and genuinely differentiated, building a custom AI-assisted solution may now be faster, cheaper, and more tailored than any off-the-shelf product.

Third, watch the pricing signals. When SaaS companies start offering significant discounts, free tiers, or “AI-powered” feature bundles, they’re responding to the competitive pressure this article describes. Those signals tell you how fast the deflationary spiral is moving in your sector.

The Bubble Won’t Burst. It’ll Deflate.

The AI bubble is real. But it won’t burst with a dramatic crash. It’ll deflate slowly, as revenue models compress, as pricing power erodes, as the gap between what AI companies promise investors and what the restructured economy can actually pay for becomes impossible to ignore.

The £320 billion SaaS industry is not going to vanish. But it is going to contract, restructure, and re-emerge as something leaner, cheaper, and more competitive. The companies that survive will be those with genuine moats: deep ecosystems, irreplaceable data, regulatory certifications, network effects. The rest will face a slow squeeze that begins with pricing pressure and ends with irrelevance.

Every technological revolution produces this dynamic. The printing press. The automobile. The internet. Each one destroyed incumbents who couldn’t adapt and created space for newcomers who could. AI will be no different.

The bubble is real. It’s just not the bubble anyone expected.