There is a particular kind of confidence that accumulates around an industry in its prime; a collective belief that what has worked will continue to work and that the mechanisms underpinning the success are not merely functional but structural. This has been the foundational investment view of software-as-a-service, or SaaS, where the moat could not be spanned and order prevailed.
Ones and zeros, the language of the software technology stack, truly found its stride when the cloud was born. No more discs and shrink wrap – the industry had found a frictionless distribution model to provide predictable revenue, rising margins, and the power of subscription.
The proposition was elegant in its simplicity: take a capability that many businesses required, develop it once, and then distribute it endlessly across a global customer base through the internet. Building robust software was genuinely difficult. It required engineering talent that was scarce, expensive, and hard to coordinate. It demanded years of iteration and user feedback, resulting in something that appeared simple despite its complexity.
The killer application is still elusive, but AI is sending vibrations through multiple industries, with Anthropic’s Claude LLM, a suite of autonomous AI agents, showing its teeth. Anthropic first launched Claude in 2023, and this month released Claude Opus 4.6, a new model designed for complex reasoning, high-level coding and agentic workflows.
Reports of how advanced this model was, especially in software coding, accelerated the rotation away from software in a market shift nicknamed ‘SaaSpocalypse’. This wiped out hundreds of billions of dollars of market capitalisation as Opus’ release gave rise to new agents and new capabilities. SaaS/software, cybersecurity, legal, IT services, and logistics companies have all been hit hard and exacerbated the decline already underway.
SaaS has been particularly hard hit by the phenomenon now known as “vibe coding”. The idea is deceptively simple. AI-assisted development tools have reached a point of maturity where individuals with little or no formal engineering training can direct an AI system to build functional software. The “vibe,” or the specification of intent, replaces the technical labour of programming and opens up the world of software to the many. The implications for SaaS are significant, and recent volatility in SaaS related companies suggests the threats are real.
Traditional software business models rely on a durable asymmetry: building software was hard, and most organisations lacked the capability to do it themselves, and it is that asymmetry that justified the licence fee. If vibe coding erodes that asymmetry, then the rationale for paying a monthly fee to a SaaS vendor becomes less compelling.
The question of whether this disruption represents genuine structural change or merely the latest cycle of technology euphoria has not been resolved. For months, investors have debated whether the vast sums directed into AI infrastructure would ever generate a commensurate return, and whether the absence of a dominant commercial application meant a reckoning was inevitable.
The observation that bubbles are harder to sustain once widely identified offers some comfort, and recent derating of technology valuations may have reduced the risk of a more violent correction. But the capital structures financing this buildout – circular, leverage-laden, and premised on future payoffs – are not without fragility. The money is not free, and the casualties, when they arrive, are unlikely to be few.
The disruption to SaaS pricing models does not stop at build-versus-buy. A further pressure arrives from the structure of how enterprise AI is being deployed. Anthropic’s Cowork (an AIpowered desktop assistant), for example, represents a category of tool that sits not above an application but, in a sense, beside the user – capable of navigating, operating, and extracting value from software environments.
When an AI agent can use a CRM, a project management tool, or a data platform as efficiently as a skilled employee, the seat-based pricing model that underlies much of SaaS – charging per user, per month – begins to look incoherent. Why pay for a seat for an agent? And if agents multiply, as they are already doing at remarkable speed, the pricing architecture of the entire industry will require a rethink.
The financial markets have registered all of this, and the damage has been substantial. Estimates of value destruction across the SaaS and software sector run to the thick end of a trillion dollars, understating the full picture when collateral damage to adjacent sectors is included. Unable to identify where safety lies within software, capital has begun rotating towards hardware, infrastructure, and physical assets. The so-called “atoms over bits” trade reflects an instinct that tangible things are harder to commoditise than digital ones. There is
some logic in that instinct, although the market’s capacity to absorb a large rotation from software to hardware investments remains limited.
This creates an uncomfortable situation for investors who must remain exposed to
technology. The mega-cap platforms have begun to diverge in how the market perceives their vulnerability. Google retains the characteristics of a store of value within technology, owing to its advertising dominance and deepening AI infrastructure presence. Amazon’s positioning across retail and cloud offers diversification that pure software companies cannot match.
Microsoft, by contrast, has come to wear a rotating ‘sorting hat’ of concern – an AI strategy that has not yet produced the narrative clarity investors need. Indeed, Microsoft is heavily investing in training initiatives for its Copilot AI, driven by a need to ensure high adoption rates and demonstrate value to corporate customers. That does not make Microsoft a structurally impaired business, but it illustrates how even dominant software incumbents are being viewed with scepticism that would have seemed excessive a few years ago.
The argument that SaaS moats are being eroded runs into a reasonable counterpoint: plenty of software businesses have large, deeply embedded customer bases, complex integrations, and genuine switching costs. These things do not evaporate overnight. A business that sits at the centre of a company’s operational workflow, touching payroll or supply chain or regulatory compliance, is not easily replaced by a vibe-coded alternative, however capable.
Nor does every SaaS category face the same degree of AI disruption. Products that are genuinely difficult to build – large, complex, brittle, tightly integrated with regulated environments – may prove far more resilient than lighter-weight tools aimed at discretionary workflows.
The more honest framing is not that SaaS is dying but that the easy growth is over. The harder question is which businesses have the substance to justify their valuations in a world where software capability is no longer scarce. The companies that will fare best are likely those that have built something genuinely structural – deep data assets, network effects that compound with use, or platforms that become more valuable precisely because AI agents need somewhere reliable to operate.
Cloudflare, for example, has positioned itself not as a software vendor but as infrastructure for the agentic internet: the network through which AI agents pass, the platform on which they compute, and the control layer through which their activity is managed and secured. That positioning is coherent in a world where agents are the new users.
The broader lesson may be this: the SaaS era is not failing because the model is wrong. Recurring revenue, customer retention, and expanding product suites remain genuinely valuable business characteristics. What is being corrected are the assumptions that a) these characteristics are sufficient insulation against structural change and b) once a SaaS company had achieved scale, it had also achieved permanence.
One analogy that has gained traction is the comparison between SaaS businesses and movie studios; the economics make the parallel instructive. A film generates significant upfront value through theatrical release, followed by a long tail of recurring income through syndication, licensing, and franchise extension. For the major studios with established franchises – the Disney model, in effect – residual value is high and the revenue base is durable. The intellectual property continues to earn long after the credits have rolled.
For studios reliant on individual productions without franchise continuity, the economics are more precarious. Each release must justify itself, the residual floor is low, and survival depends on a sequence of hits that is difficult to engineer. The independent studio, operating without the buffer of accumulated IP and recurring income streams, is an increasingly marginal proposition – not because the craft has diminished, but because the business model offers little protection against a run of misfortune.
The parallel for SaaS is pointed. A software business that is not genuinely embedded in its customers’ operations – one whose product can be replicated, substituted, or displaced without significant switching cost – faces a structurally similar challenge. It must run harder to stand still, shipping product at a tempo that AI will increasingly dictate. The consolidation that reshaped the studio landscape over decades may be arriving in software in compressed form. It is already predicted that in a short period, AI will have written more code than all
humans over the history of computing, while some predict that in three to six months from now, AI coding will account for 90% of all code written.
What about the investments?
For years, it has been assumed that a mature software business likely had the
characteristics of a perpetual income stream that would underpin a high valuation for a growing business. However, this rests on two key assumptions: that retention rates would remain high and stable, and that the terminal value was not zero in the out years. Whether you subscribe to the possibility that vibe coding or agentic workers will derail the software train – a narrative that perhaps deserves more scepticism than markets are currently suggesting – or not, it is difficult to imagine an explosion of cheaper, more abundant software does not take market share.
With the marginal cost of software creation cratering, there will be commoditisation of markets with similar-looking product, and the software buyer will have choices. With that the case, no matter what you expect the stability or terminal rate to be, the discount rate has risen. While, for some, the decline may have been overdone, it is understandable.
What would restore confidence?
The honest answer is that evidence, and time, will be required – evidence that retention rates are holding, that pricing power has not permanently shifted to the buyer, and that the productivity gains from AI are accruing to software vendors rather than merely flowing through them to their customers. Some of that evidence will emerge in quarterly earnings. Some will take longer, as the full consequences of vibe coding and agentic deployment work their way through enterprise technology stacks.
In the meantime, the more durable question for investors is not whether SaaS survives, but which businesses within it have built something that compounds rather than commoditises. The SaaS model was never the problem. The assumption of permanence was. Those businesses that understood the difference, and built accordingly, may yet find that the architecture, though tested, was sound.
Tim Chesterfield is CIO of the Perpetual Guardian Group and the founding CIO and Director of its investment management business, PG Investments. With $2.8 billion in funds under management and $8 billion in total assets under management, Perpetual Guardian Group is a leading financial services provider to New Zealanders.
Disclaimer
Information provided in this publication is not personalised and does not take into account the particular financial situation, needs or goals of any person. Professional investment advice should be taken before making an investment. The information provided in this article is not a recommendation to buy, sell, or hold any of the companies mentioned. PG Investments is not responsible for, and expressly disclaims all liability for, damages of any kind arising out of use, reference to, or reliance on any information contained within this article, and no guarantee is given that the information provided in this article is correct, complete, and up to date.
