The Software as a Service business model is under pressure, and parts of it are nearing the edge of death. AI has made applications faster to build, common features easier to replicate, and companies increasingly capable of creating their own systems without remaining dependent on third-party vendors.
The pressure is already visible in real companies. Several large and mid-sized SaaS brands have reported declines in paying users, subscription revenue, or annual recurring revenue. This does not prove that the entire SaaS industry will collapse, but it does show that the traditional model is no longer immune.
Declining SaaS Subscriptions Are No Longer Just a Prediction
Several SaaS companies have reported falling customer counts or subscription revenue, although the causes and severity differ across businesses.
Dropbox lost paying users as ARR declined
Dropbox shows that even a major brand can contract when its core product operates in a mature market filled with comparable alternatives.
In its fiscal 2025 results, Dropbox reported 18.08 million paying users at the end of 2025, down from 18.22 million a year earlier. Fourth-quarter revenue fell 1.1 percent year over year, while total annual recurring revenue declined 1.9 percent.
The drop from 18.22 million to 18.08 million is not a sudden collapse. Dropbox also continued to generate strong profits and cash flow. The numbers nevertheless reveal an important reality: scale and brand recognition do not automatically protect a SaaS company from saturation, churn, pricing pressure, and product substitution.
Chegg lost subscription revenue as AI became an alternative
Chegg provides a more severe example of how AI can erode a subscription product that once held a strong market position.
In its 2025 annual report, Chegg said that students were increasingly turning to generative AI products, including ChatGPT, as alternatives to vertical education services such as Chegg. The company warned that this behavior was accelerating the decline in new subscriber sign-ups.
Chegg's Academic Services revenue fell by $235.4 million, or 43 percent, in 2025 compared with the previous year. The decrease was primarily driven by a $223.8 million decline in subscription revenue caused by lower traffic and fewer subscribers.
Chegg's experience cannot be applied directly to every SaaS category because digital education has its own market dynamics. The threat pattern is still relevant elsewhere: when AI can deliver much of the outcome users want through a simpler interface, the reason to keep an older subscription becomes weaker.
The Subscription Model Is Losing Its Immunity
A decline at several companies does not mean that every SaaS category is shrinking. It does show that software subscriptions can no longer rely on customer habit and technical barriers as their primary defenses.
For years, SaaS benefited from three major advantages:
- building software internally was too expensive;
- integrations required specialized technical teams;
- paying a monthly fee was easier than funding a development project.
AI is eroding all three advantages. Companies can now turn requirements, workflows, data structures, user roles, and integration needs into working applications far more quickly.
When the cost of a prototype or narrowly scoped internal system can fall from thousands of dollars to hundreds, the decision to buy SaaS needs to be recalculated. Those figures are not universal benchmarks. Production systems handling sensitive data, financial transactions, compliance obligations, or high scale can still require substantial investment.
The direction of change is clear, however. Software that was once too expensive to build internally is becoming increasingly practical to create.
AI Is Turning SaaS Features into Commodities
AI weakens feature lists as a competitive defense because common software patterns can be replicated quickly.
Dashboards, forms, authentication, approvals, notifications, user management, report exports, payments, and API integrations are recurring patterns. With clear requirements, AI can help generate database schemas, interfaces, APIs, tests, documentation, and initial configurations.
There is evidence that AI can accelerate specific programming tasks. In a controlled experiment published by Microsoft Research, developers using GitHub Copilot completed a JavaScript HTTP server task 55.8 percent faster than the control group.
That study measured one limited programming task, not the development of an enterprise system from concept to production. Its result should not be used to promise that every project will finish 55.8 percent faster. It does support the narrower conclusion that parts of implementation work can be accelerated materially.
If the main value of a SaaS product can be described as a short list of standard features, the product lacks a durable defense. Competitors can reproduce it. Customers can also build a smaller version that fits their internal process more closely.
Companies Can Become Their Own Software Vendors
A company with a clear business architecture can now build internal applications without outsourcing its entire process to an external vendor.
An organization can create systems for budget approvals, asset management, vendor portals, production scheduling, specialized CRM, or project tracking. These systems can connect directly to existing databases, payment services, document storage, and communication channels.
The critical capability is no longer typing code quickly. Value has shifted toward defining:
- business processes and exception conditions;
- data structures and ownership;
- permissions for each role;
- validation and failure-handling rules;
- integration requirements;
- security, auditing, backup, and recovery standards;
- measurable success criteria.
With that foundation, AI can become a very fast implementation engine. Companies do not always need a large SaaS suite containing dozens of features when only two or three functions matter to their operations.
Why Are Customers Cutting SaaS Subscriptions?
Customers are reducing subscriptions not only because of economic conditions. They are also questioning products that overlap, remain underused, or can be replaced by AI.
1. Too many tools perform overlapping functions
Companies often pay for separate applications covering communication, documents, analytics, automation, and project management. When one platform or AI agent can execute work across systems, some subscriptions become difficult to justify.
2. Per-seat pricing no longer matches AI-driven work
Per-seat pricing assumes that people open an application and complete tasks directly. When an AI agent can perform work for many people through APIs, seat count becomes a poor measure of value.
IDC predicts that by 2028, purely seat-based pricing will be obsolete and 70 percent of software vendors will restructure pricing around consumption, outcomes, or organizational capabilities. This is an industry forecast, not a certainty, but it illustrates the pressure on the traditional model.
3. Customers can build a good-enough version
Customers do not always need the most comprehensive product. They need a system that completes a specific process, connects to internal data, and follows the way their organization works.
An internal application does not need to match every capability of a commercial SaaS product. It only needs to replace the portion that the customer actually uses. This is why a large product can be threatened by a much smaller system.
4. AI reduces the value of the interface
Much of SaaS value has historically lived in dashboards and click-based workflows. AI agents are turning those interactions into natural-language instructions and cross-application automation.
When a user can simply say, “Create the sales report, check for anomalies, and send it for approval,” individual applications may become background services. Vendors that control only the interface risk losing their direct relationship with users.
5. Generic products are easy to compare and replace
The more generic a product's function, the easier it is for customers to find an alternative, combine several services, or build an internal version. Differentiation based on appearance and a feature checklist is no longer enough.
A Few Security Keywords Do Not Make an Application Secure
AI can accelerate application development, but security is not solved by adding words such as encryption, authentication, and role-based access control to a prompt.
In an evaluation covering more than 100 language models across Java, Python, C#, and JavaScript, Veracode reported that 45 percent of samples failed security tests and introduced vulnerabilities from the OWASP Top 10.
This vendor study used curated tasks, so its results do not represent all AI-generated code deployed in the real world. Its findings are still sufficient to show that functional code is not automatically secure.
AI-assisted applications still require human review, automated tests, dependency scanning, secrets management, access controls, monitoring, and risk-appropriate penetration testing.
AI lowers the cost of creating a first version. It does not eliminate the cost of making software secure and trustworthy.
Which SaaS Products Are Nearing the Edge of Death?
The most exposed SaaS products are generic tools whose features are easy to explain, easy to copy, and unsupported by unique assets beyond their code.
Warning signs include:
- most features can be rebuilt with common components;
- customers use only a small portion of the available functions;
- the product has few critical integrations;
- customer data can be moved easily;
- switching costs are low;
- pricing depends heavily on seat count;
- product value is difficult to connect to business outcomes;
- AI or a larger platform can absorb its core function.
Products with these traits may not fail immediately. Their companies may survive through cash flow, cost reductions, price increases, or bundling. Growth can still come under pressure while customer retention becomes increasingly fragile.
The SaaS Companies That Can Still Survive
Surviving SaaS companies will rely on defenses that cannot be reproduced easily through prompts and AI-generated code.
Those defenses include:
- proprietary data and deep domain context;
- complex integrations that must remain reliable;
- proven compliance, auditing, and security;
- user, partner, or marketplace networks;
- strong distribution and brand recognition;
- an operating track record for critical processes;
- measurable business outcomes rather than access to features.
The future of SaaS is unlikely to be a collection of dashboards priced per user. It is more likely to consist of modular services, APIs, data, infrastructure, and agents priced according to usage or outcomes.
Conclusion: SaaS Is Not Disappearing, but Its Old Model Is Dying
A SaaS business approaches the edge of death when its only advantage is that software used to be difficult and expensive to build.
Dropbox shows that a major company can lose paying users and experience declining ARR. Chegg shows a more severe version of the threat, where AI became a direct alternative and subscription revenue fell sharply. These two cases do not prove that every SaaS company will collapse, but they show that the pressure is real.
When customers can build much of what they need, code and features are no longer sufficient defenses. Vendors need data, distribution, integrations, compliance, trust, or responsibility for outcomes that cannot be copied easily.
The question is no longer whether SaaS will die. The sharper question is this: how many customers will continue paying for a subscription when they can build a good-enough prototype for a limited need within days?
FAQ
Is the SaaS business model really nearing death?
Parts of the SaaS model are approaching a breaking point, especially generic products that are easy to copy and depend on per-seat pricing. SaaS products with strong data, integrations, compliance, distribution, or operational value can still survive.
Which SaaS companies have reported declining customers or subscriptions?
Dropbox reported that paying users fell from 18.22 million to 18.08 million at the end of 2025. Chegg reported a $223.8 million decline in Academic Services subscription revenue in 2025, primarily because of lower traffic and fewer subscribers.
Does AI always make building an application cheaper than buying SaaS?
No. AI can reduce initial development costs, especially for prototypes and narrowly scoped internal applications. Total ownership costs still include security, hosting, maintenance, support, integrations, monitoring, and data migration.
Why are generic SaaS products the most vulnerable?
Generic SaaS products often consist of common patterns such as dashboards, forms, authentication, reports, and notifications. AI and ready-made components make those features increasingly easy to reproduce.
What is the strongest defense for a SaaS business?
The strongest defenses exist outside the codebase, including proprietary data, domain expertise, deep integrations, compliance, trust, distribution, community, and measurable business outcomes.