"Which software and AI categories are building the most durable moats — and which are most at risk of being absorbed by the hyperscalers and foundation-model providers over the next 18 months?"

The fundamental nature of software moats is being re-evaluated, with a consensus that while artificial intelligence is a powerful tool for differentiation, it is not a durable moat in itself due to its replicability [1, 14]. Traditional moats such as network effects, deep workflow integration, and becoming a system of record remain critical for defensibility [1, 2, 5]. However, there is tension regarding the durability of other historical moats; some sources argue that complex backend logic remains a significant barrier to entry for incumbents like Stripe and Rippling [3, 4], while others contend that AI weakens the defensibility of complex proprietary software and is actively eliminating moats like data lock-in and migration pain [22, 30]. A more extreme view posits that the concept of a durable moat in dynamic software sectors has never been true . In this evolving landscape, proprietary, live data has emerged as a more powerful and durable form of defensibility for AI-native companies [6, 9, 10].

High-quality, incumbent SaaS companies with mission-critical platforms are considered the most durable category, largely insulated from the "SaaSpocalypse" narrative that gained traction in early 2024 . Firms like Salesforce, ServiceNow, and Adobe are well-positioned to integrate AI as an enabler, enhancing their existing offerings [2, 29]. Their resilience stems from established moats beyond code, including customer trust, governance, security, liability management, and vast distribution networks [2, 8]. For new entrants, the most durable path involves building on the application layer by focusing on **complex, industry-specific workflows** and proprietary datasets [9, 26]. This strategy allows startups to create defensibility against foundation model providers by solving nuanced problems that require specialized data and logic, thereby aggregating value from an increasingly commoditized model layer .

Conversely, software categories at the highest risk of absorption are single-function "point solutions" and thin AI wrappers built directly on top of foundation models [2, 7]. These companies are vulnerable to being made obsolete as model providers like OpenAI and Anthropic expand their native functionalities, turning what was once a product into a feature [7, 12, 17]. Companies with no proprietary data, quick installation, and no regulatory overlay face significant disruption, with some experts predicting they may not have a right to exist **within five years** [18, 25]. A critical vulnerability for many established SaaS players is the per-seat pricing model, which is threatened as AI reduces customers' need for human employees, thus cannibalizing the incumbent's revenue base [13, 16, 23]. This creates an urgent need for these companies to pivot to new pricing models based on outcomes or value delivered .

The future role of hyperscalers and foundation-model providers remains a subject of debate, creating uncertainty for the entire software ecosystem. One perspective holds that they will become an infrastructure layer, partnering with application companies in a manner similar to AWS's role in cloud computing . An opposing view suggests they will compete directly, aggressively expanding into the application layer and adopting enterprise sales models to capture large customers [12, 20]. This uncertainty extends to where long-term value will accrue, with credible arguments being made for the foundation models themselves , the application layer that aggregates them , and the core infrastructure providers who benefit from the proliferation of all AI-built software . This transition is fundamentally expanding the total addressable market for software from IT spend to **the much larger category of labor spend**, reframing the entire opportunity around automation and labor replacement [1, 15, 21].