Foundational Models Will Cannibalize the SaaS Built on Top of Them
LLMs absorb the functionality of apps built on top of them
The common assumption is that an entire SaaS ecosystem will blossom atop foundational models, much like mobile apps atop iOS or web services atop AWS. But this analogy breaks down under the gravity of model centralization and scope creep.
Here's the deeper structural insight:
Foundational models aren’t just platforms. They’re agglomerative engines. As they improve, they tend to absorb the functionality of many thin layers built above them. Unlike an operating system, which mediates access to hardware, or a cloud API, which offers modular services, LLMs are universal function approximators. Their improvement doesn’t just expand capability. It obsoletes intermediaries.
Historical analogy:
Think of Google Search. At first, it sent traffic to weather sites, dictionary sites, and flight trackers. Over time, it consumed them by surfacing weather directly, defining words, showing flight prices. GPT-4-class models do something similar: as their ability to reason, retrieve, and interact improves, the demand for specialized wrappers evaporates.
This is especially true when:
The app adds only a thin UI or prompt layer over the model.
The value is in domain-specific RAG or structured output, which foundation model providers can integrate directly (see GPTs, agents, or tool use).
Implication:
SaaS AI is increasingly a race against integration. If your product is a thin abstraction over ChatGPT or Claude, the question is not if OpenAI or Anthropic will replicate it, but when, and whether their integration will outperform your UX, latency, or customization.
The only defensible moats are:
Deep domain coupling (e.g., EMR systems, CAD workflows, industrial control)
Proprietary data flywheels
Offline/edge deployment needs
Trust, compliance, or explainability mandates
Integration into legacy systems closed models won’t touch
But in general: foundational models are increasingly predatory in their horizontal spread. “Build on top” is not a stable strategy unless your layer adds something truly irreplicable or your distribution moat is massive.