In the first week of February 2026, nearly one trillion dollars in software market capitalization evaporated. Atlassian dropped 35% in a single week. The iShares Expanded Tech-Software ETF entered bear market territory. Analysts scrambled to coin a name for what was happening. They settled on “SaaSpocalypse.”
The proximate cause was Anthropic’s release of Claude Cowork and a set of open-source plugins that demonstrated an AI agent capable of autonomously navigating interfaces, manipulating files, and executing multi-step business processes. When disclosure emerged that the legal plugin alone could automate 90% of standard NDA and compliance triage, the selling began. It did not stop for days.
But the sell-off was not caused by a single product announcement. It was the market finally pricing in a structural reality that has been building for eighteen months: the moats that protected traditional SaaS businesses are collapsing, and AI is the force doing the collapsing.
We have been building AI-native systems for clients across industries for years. We have watched this shift from the inside, working with enterprises that are actively replacing SaaS products with custom-built alternatives. What follows is our analysis of what is actually happening, why it matters, and what comes next.
The Moat Problem
Traditional SaaS businesses were protected by three moats: complexity of implementation, switching costs, and the economics of custom development.
Building software was hard. It required specialized talent, significant capital, and months or years of development time. Buying a SaaS product was almost always cheaper and faster than building the equivalent in-house. Even when the SaaS product was imperfect, even when it forced you to adapt your workflows to its assumptions, the build-versus-buy math overwhelmingly favored buying.
Once you bought, switching costs kept you locked in. Your data was in the vendor’s format. Your team was trained on the vendor’s interface. Your integrations were built to the vendor’s API. Leaving meant rebuilding all of that, and the cost of rebuilding reinforced the original decision to buy.
These moats created a $700 billion global SaaS market. They funded billion-dollar valuations for companies that built project management tools, CRM systems, analytics dashboards, and workflow automations. The moats were real, and they were wide.
AI is draining them.
Why the Moats Are Failing
The implementation moat depended on the cost and difficulty of building custom software. AI agents have reduced both by an order of magnitude. Tasks that required a team of developers working for months can now be accomplished by a single engineer directing agents over a long weekend. The Retool 2026 Build vs. Buy report, which surveyed over 800 respondents, found that 35% of teams have already replaced at least one SaaS tool with a custom build. That number will be higher by the time you read this.
The switching cost moat depended on the difficulty of migration. AI agents are remarkably good at data migration, format conversion, and integration rebuilding. The tasks that made switching painful, parsing legacy data formats, mapping field schemas, rebuilding API integrations, are exactly the kind of structured, well-defined tasks that agents excel at. Switching costs have not disappeared, but they have been reduced to a fraction of what they were.
The economics moat depended on the gap between the cost of building and the cost of buying. When building was expensive, even an expensive SaaS subscription was the rational choice. When building becomes cheap, the calculus inverts. A team that can build a custom tool tailored to their exact workflow for less than the annual cost of a SaaS subscription will build. And increasingly, they can.
Seat Compression and the Revenue Crisis
The moat collapse is only half the problem for SaaS companies. The other half is seat compression.
SaaS revenue models are overwhelmingly based on per-seat pricing. You pay per user per month. This model worked because there was a roughly linear relationship between the number of employees doing knowledge work and the number of software licenses needed to support them.
AI agents break this relationship. When an AI agent can do the work of five junior analysts, enterprises need fewer analysts and fewer analyst-seat licenses. The relationship between headcount and software consumption decouples, and SaaS revenue shrinks even when customer retention stays constant.
The data on pricing model migration is stark. Seat-based pricing among SaaS companies dropped from 21% to 15% in a single year. Hybrid usage-based models surged from 27% to 41%. Credit-based pricing models increased 126% year-over-year. The industry is scrambling to find pricing models that survive in a world where the number of human seats is an increasingly poor proxy for the value delivered.
Companies clinging to per-seat pricing for AI-era products are already paying the price: 40% lower gross margins and 2.3x higher churn compared to those that have adopted usage or outcome-based models. The market is sending a clear signal about which pricing models have a future.
The Build-Over-Buy Inflection
The Retool report quantifies what we have been seeing with our clients: the build-versus-buy decision has reached an inflection point.
78% of respondents expect to build more custom internal tools in 2026. The SaaS categories most at risk of replacement are exactly the ones you would predict: workflow automations (35% already being replaced), internal admin tools (33%), business intelligence tools (29%), CRMs and form builders (25%), and project management tools (23%).
The reasons for building are not just economic. Respondents cited frustration with SaaS products that do not fit their workflows, slow vendor support cycles, and the overhead of managing dozens of SaaS subscriptions that each solve 70% of the problem while requiring workarounds for the remaining 30%.
One respondent captured the sentiment precisely: it was faster to rebuild the product internally than to wait for the vendor’s support team to respond. When building takes hours and vendor support takes days, the rational choice is to build.
This does not mean all SaaS is doomed. Products with deep proprietary data moats, genuine network effects, or regulatory compliance frameworks that are expensive to replicate will survive and potentially thrive. Palantir’s stock price held steady through the sell-off for exactly this reason. The products at risk are the ones whose primary value was convenience rather than unique capability, the ones that were easier to buy than build. That ease-of-building advantage has evaporated.
The Shadow IT Explosion
The build-over-buy shift has a secondary effect that enterprise IT organizations are only beginning to grapple with: shadow IT at an unprecedented scale.
The Retool report found that 60% of builders created tools outside IT oversight in the past year. When anyone with an API key and basic technical literacy can direct an AI agent to build a functional application in a weekend, the traditional IT procurement and governance process becomes a bottleneck that gets routed around rather than reformed.
This is not a new dynamic. Shadow IT has existed since the first department bought Salesforce on a corporate credit card without telling the CIO. But the scale is different when the tool being built is not a SaaS subscription that IT can eventually discover and govern, but a custom application running on internal infrastructure that nobody outside the team knows exists.
The security implications are significant. Enterprise software procurement processes exist partly to enforce security standards, compliance requirements, and data governance policies. When those processes get bypassed, the organization loses visibility into where data is flowing, what security standards are being applied, and whether regulatory requirements are being met.
But the solution is not to crack down on shadow building. The solution is to create governance frameworks that accommodate the new pace of custom tool creation. Organizations that try to force all custom development through a six-week procurement review will find that the tools get built anyway, just without any governance at all. Organizations that create lightweight, fast governance processes, automated security scanning, approved deployment environments, standardized authentication patterns, will get both the speed of custom building and the oversight the enterprise requires.
The Hyperscaler Squeeze
While SaaS companies are losing revenue from below (customers replacing products with custom builds), they are also losing budget from above.
Hyperscalers are projected to spend over $470 billion on AI infrastructure in 2026. Enterprises are following the same pattern. Gartner estimates that 60% of 2026 software spending growth will merely offset rising AI delivery costs rather than fund new software purchases. The budget for traditional SaaS is being squeezed from both directions: customers need fewer seats, and the remaining budget is being redirected toward AI infrastructure.
This creates a compounding problem for SaaS companies. Less revenue means less investment in product development. Less product development means less differentiation. Less differentiation means more customers deciding to build instead of buy. The cycle accelerates.
What Survives
Not all SaaS dies in this transition. The categories that survive share common characteristics.
Deep data moats. Products that accumulate proprietary data that becomes more valuable over time, and that cannot be replicated by building a clone, retain their defensibility. A CRM is replaceable. A data platform with years of proprietary industry benchmarks is not.
Genuine network effects. Products whose value increases with each additional user in the network cannot be replicated by building a single-tenant alternative. Communication platforms, marketplaces, and collaborative tools with established networks retain structural advantages that AI does not erode.
Regulatory compliance. Products that have invested heavily in regulatory certification, SOC 2, HIPAA, FedRAMP, PCI-DSS, retain an advantage because compliance certification is expensive and time-consuming regardless of how the software is built. An AI agent can build a HIPAA-compliant application, but getting it certified still requires the same audit process.
Infrastructure and platform layers. Cloud providers, database systems, and developer platforms are not threatened by the build-over-buy shift because they are what custom applications are built on. The more people build custom tools, the more infrastructure they consume. AWS, GCP, and Azure benefit from the SaaSpocalypse.
Everything else, every SaaS product whose primary value proposition was “it was easier to buy this than to build it”, is in structural decline. The question is not whether these products will lose market share, but how fast.
What We Are Seeing
In our client work, the shift is already operational, not theoretical.
We are helping enterprises build custom internal tools that replace three or four SaaS subscriptions simultaneously, not because we are against SaaS, but because a single custom tool that fits the client’s actual workflow is cheaper to build and maintain than four SaaS products that each require workarounds and manual data synchronization.
The economics are striking. A typical enterprise SaaS stack for a mid-size team, project management, analytics, workflow automation, and internal tooling, runs $200,000 to $500,000 per year in licensing. A custom solution that covers 90% of the same functionality, built with AI-native methods, can be delivered in weeks and maintained for a fraction of the annual licensing cost.
This is not a case for building everything custom. It is a case for re-evaluating the build-versus-buy decision with current cost structures rather than assumptions from 2020. The right answer for most organizations is a hybrid: keep SaaS products where they provide genuine, irreplaceable value (deep data, network effects, compliance), and build custom where the SaaS product was primarily a convenience purchase.
The Consulting Opportunity
For firms like ours, the SaaSpocalypse is not a crisis. It is the largest expansion of addressable market in a generation.
When every enterprise is re-evaluating its software stack and asking “should we still be buying this, or should we build it?”, the demand for teams that can deliver custom solutions at speed and at quality is enormous. The consulting market for AI implementation is projected to reach $90 billion by 2035, growing at 26% annually.
But this opportunity requires a different operating model than traditional software consulting. Clients do not want to pay for six months of discovery and twelve months of implementation. They want outcomes delivered in weeks. They want fixed-price engagements tied to business results, not open-ended time-and-materials contracts. They want partners who can move at the pace AI enables, not partners who use AI to pad billable hours.
We will address the transformation of consulting economics in a follow-up piece. For now, the point is that the SaaSpocalypse is not a destructive event for the entire software industry. It is a redistribution. Value is moving from generic SaaS products to custom solutions, and from subscription revenue to delivery revenue. The organizations and firms that position for this redistribution will thrive. The ones that cling to the old model will not.
What Comes Next
The February sell-off was the market catching up to reality. It was not the end of the transition. It was the beginning of broad awareness that the transition is happening.
Over the next twelve to eighteen months, we expect several things to accelerate:
Enterprise SaaS consolidation. Companies with weak moats will be acquired or fail. The survivors will be the ones with genuine data advantages, network effects, or compliance certifications. The middle of the market, products that are good enough but not differentiated, will hollow out.
Platform plays by AI companies. Anthropic, OpenAI, and Google will continue releasing capabilities that directly compete with categories of SaaS products. They are not trying to become SaaS companies. They are building platforms whose capabilities make categories of SaaS unnecessary. This is a more fundamental threat than competition because it eliminates demand rather than competing for it.
Custom development as a core competency. Enterprises that have outsourced all software development for the past decade will begin rebuilding internal capability, not to hire armies of developers, but to maintain small, highly capable teams that can direct AI agents to build and maintain custom tools. The CTO’s job is changing from vendor management to orchestration management.
New pricing models. Surviving SaaS companies will aggressively migrate to usage-based and outcome-based pricing. Per-seat pricing will become a relic. The transition will be painful for companies whose financial models, investor expectations, and sales compensation plans are built around predictable per-seat revenue, but the alternative is irrelevance.
The SaaSpocalypse is not a momentary panic. It is the market processing a structural change in the economics of software. The era of SaaS as the default answer to every business software need is ending. What replaces it is a world where the build-versus-buy calculation is re-run continuously, where custom solutions are cheap enough to compete with subscriptions, and where the winners are the organizations that can deliver outcomes rather than licenses.
That is the world we are building for. It is already here.
