Icecat Capital’s Investment Thesis as Software Code is Commoditized by AI

From the perspective of our invest arm, Icecat Capital, the “SaaSpocalypse” – in which the valuations of SaaSs steadily decline – is not a false narrative but a structural reality: AI has fundamentally commoditized the creation and implementation of software, rendering traditional, seat-based and expensive SaaS models highly vulnerable.

The Commoditization of Code

The barrier to entry for enterprise software has collapsed. As demonstrated by developers using tools like Claude to replicate complex software suites in weeks, code is no longer a defensible moat. This is playing out at scale with major corporations realizing that renting software is no longer economically sound when AI-assisted generation makes owning it feasible. Starbucks, for instance, is actively developing AI-powered in-house software to replace systems from entrenched vendors like Microsoft, IBM, and Oracle as part of a massive $2 billion cost-reduction initiative. Icecat itself proved this internally by ditching Atlassian’s Jira and Confluence to build and deploy its own AI-driven alternative, Icecat Projects, in a matter of weeks.

The Incumbent Denial

ServiceNow’s CFO recently dismissed the AI disruption narrative, arguing that enterprise software requires strict security, guardrails, and governance that AI alone cannot provide, stating that “every business and industry needs a harness for AI”. From an Icecat Capital perspective, this is a textbook example of the innovator’s dilemma. Incumbent SaaS giants are retreating to compliance, integration, and security as their final moats because the core functionality of their products is easily replicable. While enterprise governance is critical, it is a feature, not a long-term defense against zero-marginal-cost software generation and implementation by disruptors.

The Icecat Capital Investment Thesis

Icecat Capital’s decision to exit traditional SaaS and invest solely in AI-resilient startups reflects the belief that simply charging for a digital workflow interface is a dying business model. A truly AI-resilient startup must possess moats that AI cannot code away:

• Proprietary Data: Companies sitting on unique, highly specialized data sets that serve as the necessary ground truth to train and refine domain-specific models.
• Agentic Services over Software Seats: Startups that sell outcomes rather than software licenses, using AI agents to replace outsourced human labor (e.g., automated legal review, robotic process automation) rather than just providing a dashboard for human workers.
• Physical or Hard-Tech Integration: Businesses where AI interfaces with the physical world, hardware, or complex logistical networks that cannot be simulated or replaced by a purely digital model.
• Regulatory Hurdles: Businesses that depend on government licensing, regulatory approval and oversight, such as exchanges, banks or critical infrastructure will benefit from AI, but can’t be replaced by it.

Ultimately, traditional SaaS companies are selling a digital abstraction of a process. When AI can write that abstraction on demand, the premium value of the SaaS vendor evaporates. Zero marginal costs will translate to abundancy in freemium or open source software functionalities. Software might just be a feature to promote AI-resilient products and services, which have a sustainable, unique value for its clients.