The initial frenzy of the generative AI boom, which seemed to mint a new startup every minute, is giving way to a more sobering reality. According to Darren Mowry, who leads Google's global startup organization across Cloud, DeepMind, and Alphabet, two once-hot business models now have their 'check engine light' on: LLM wrappers and AI aggregators. His warning signals a critical inflection point where novelty is no longer enough, and sustainable, differentiated value is the only path to survival.

LLM wrappers are startups that build a product or user experience layer on top of existing large language models like Claude, GPT, or Gemini to solve a specific problem. A classic example would be a company using an AI model to help students study. Mowry argues that the industry's patience for this approach has worn thin. 'If you’re really just counting on the back end model to do all the work and you’re almost white-labeling that model, the industry doesn’t have a lot of patience for that anymore,' he stated on the podcast Equity. He emphasized that wrapping 'very thin intellectual property' around a foundation model is no longer a viable strategy for differentiation or growth.

For a wrapper startup to succeed, Mowry contends it must build 'deep, wide moats' through either horizontal differentiation or deep specialization in a specific vertical market. He points to examples like Cursor, a GPT-powered coding assistant, and Harvey AI, a legal AI assistant, as companies that have managed to create substantial proprietary value on top of the underlying models. The era where a startup could simply 'slap a UI on top of a GPT' and gain traction, as some did following the launch of OpenAI's ChatGPT store, is effectively over. The new imperative is to build durable product value that cannot be easily replicated.

A related and equally vulnerable category is the AI aggregator. These are startups that aggregate multiple LLMs into a single interface or API layer, routing user queries across different models and providing access to a suite of them. These platforms often include orchestration features like monitoring, governance, or evaluation tooling. Companies like the AI search startup Perplexity or the developer platform OpenRouter, which offers a single API for multiple models, fall into this category. Despite some early success, Mowry's advice to new entrants is blunt: 'Stay out of the aggregator business.'

He explains that aggregators are struggling to grow because users increasingly demand 'some intellectual property built in' to intelligently route requests to the optimal model based on specific needs, not just based on backend compute costs or access constraints. As the major model providers like OpenAI, Google, and Anthropic expand their own enterprise feature sets and tooling, these aggregator platforms face intense margin pressure and the risk of being squeezed out of the market.

Mowry draws a direct parallel to the early days of cloud computing in the late 2000s and early 2010s. When Amazon Web Services began to take off, a wave of startups emerged to resell AWS infrastructure, offering simplified entry points with consolidated billing and support. However, as Amazon built out its own enterprise tools and customers became more sophisticated in managing cloud services directly, most of those intermediary startups vanished. The only survivors were those that added genuine, value-added services like security, migration, or DevOps consulting. Mowry, who has decades of experience in the cloud sector with stints at AWS and Microsoft before Google, sees history repeating itself in the AI landscape. The message is clear: in a market dominated by tech giants with vast resources, startups must offer more than just convenience or access; they must build defensible, proprietary technology and deep domain expertise to endure.