Google VP warns that two types of AI startups may not survive

The rapid expansion of generative artificial intelligence spurred a surge in startup creation, but a period of reassessment is now underway. According to Darren Mowry, who leads Google’s global startup organization across Cloud, DeepMind, and Alphabet, two business models – LLM (Large Language Model) wrappers and AI aggregators – are facing increasing scrutiny and potential challenges.

The Rise and Potential Fall of LLM Wrappers

LLM wrappers involve building a product or user experience layer around existing large language models like Claude, GPT, or Gemini to address a specific problem. An example cited is a startup utilizing AI to assist students with their studies. However, Mowry suggests that startups relying heavily on these underlying models, essentially “white-labelling” them, are encountering resistance.

Did You Know? In early 2024, OpenAI launched a ChatGPT store, signaling a shift towards direct access and competition for LLM-based startups.

Mowry emphasized the need for “deep, wide moats” – significant differentiation – whether through horizontal innovation or specialization in a specific vertical market. Successful examples of LLM wrappers with strong differentiation include Cursor, a GPT-powered coding assistant, and Harvey AI, focused on legal applications. The ability to build sustainable product value, rather than simply adding a user interface to an existing model, is now critical.

AI Aggregators Face Headwinds

AI aggregators take a different approach, combining multiple LLMs into a single interface or API to route queries and provide access to various models. Companies like Perplexity and OpenRouter fall into this category. Mowry advises startups to “stay out of the aggregator business,” as these platforms are not currently demonstrating substantial growth.

Expert Insight: The current situation echoes the early days of cloud computing in the late 2000s/early 2010s, where companies reselling AWS infrastructure were largely displaced as Amazon developed its own enterprise tools.

The core issue, according to Mowry, is a lack of built-in intellectual property. Users desire intelligent routing to the most appropriate model based on their specific needs, not simply access to multiple models. This mirrors the evolution of cloud computing, where initial resellers were overtaken by providers offering comprehensive services.

Looking Ahead

Mowry expressed optimism about “vibe coding” and developer platforms, citing record-breaking investment in companies like Replit, Lovable, and Cursor in 2025. He also anticipates growth in direct-to-consumer AI applications, such as Google’s Veo video generator for film and TV students. Beyond AI, Mowry highlighted biotech and climate tech as promising areas, driven by increasing data availability and investment.

Frequently Asked Questions

What are LLM wrappers?

LLM wrappers are startups that build a product or user experience around existing large language models like Claude, GPT, or Gemini to solve a specific problem.

What is the primary concern with AI aggregators?

The primary concern is that they lack sufficient intellectual property and users want intelligent routing to the right model, not just access to multiple models.

What types of AI startups does Mowry see as promising?

Mowry is bullish on vibe coding, developer platforms, and direct-to-consumer AI applications.

As the generative AI landscape matures, what strategies do you think will be most crucial for startups to establish a sustainable competitive advantage?