AI startups are facing challenges as the market matures, and business models based on LLMs are becoming unsustainable.

The generative AI boom has led to the rapid emergence of numerous startups, but not all business models based on large language models (LLMs) prove to be viable.

Darren Mowry from Google emphasizes that startups that rely too much on others' models, such as LLM wrappers and AI aggregators, may face challenges. LLM wrappers, which build interfaces on existing models, do not provide defensible value, and simply white-labeling a model is no longer sufficient. Instead, startups need to develop deep expertise, proprietary data, and workflow integration. AI aggregators, which combine multiple LLMs, also face pressures as model providers develop enterprise features that reduce their margins. Mowry compares this situation to the early days of the cloud, when startups reselling AWS services disappeared in favor of those adding real value. However, there are opportunities in platforms for developers, direct-to-consumer applications, and the biotech and climate tech sectors, which benefit from access to data and AI analytics.

The message for founders is clear: the future success of AI startups will depend on the smart integration of models into real workflows and industry-specific problems.