The generative AI gold rush created a startup nearly every minute. But as the market matures, two once-trendy business models — LLM wrappers and AI aggregators — are increasingly under pressure, according to Darren Mowry, who leads Google’s global startup organization across Cloud, DeepMind, and Alphabet.
Speaking on the podcast Equity, Mowry said startups built primarily as thin layers on top of existing large language models have their “check engine light” on.
The Problem With Thin LLM Wrappers
LLM wrappers are companies that build a user interface or product experience on top of existing models such as Claude, GPT-5, or Gemini to address specific use cases — for example, study tools for students.
According to Mowry, simply white-labeling a backend model with minimal proprietary technology is no longer enough.
“If you’re really just counting on the back-end model to do all the work … the industry doesn’t have a lot of patience for that anymore,” he said.
Startups need strong competitive moats — either horizontal differentiation across industries or deep specialization within vertical markets. Companies like Cursor, a GPT-powered coding assistant, and Harvey AI, which focuses on legal workflows, represent examples of LLM-powered products with deeper defensibility.
The days of launching a simple GPT interface — as many did during the excitement following OpenAI’s ChatGPT store debut in 2024 — are fading. Sustainable product value now requires proprietary workflows, data advantages, or domain expertise.
Why AI Aggregators May Struggle
AI aggregators — a subset of wrappers — combine multiple models into one interface or API, routing queries across providers. Examples include Perplexity and OpenRouter, which provide unified access to different AI models along with orchestration tools like monitoring and governance.
But Mowry’s advice to new founders is blunt: “Stay out of the aggregator business.”
He argues that aggregators face margin pressure as model providers expand their own enterprise features. Customers increasingly expect built-in intelligence that routes tasks to the right model based on specific needs — not simply access to multiple models behind a unified interface.
Lessons From the Cloud Era
Mowry compares today’s AI ecosystem to the early cloud computing boom of the late 2000s and early 2010s. Back then, startups emerged to resell infrastructure from Amazon Web Services and other providers, promising easier onboarding and consolidated billing.
When cloud giants built their own enterprise tools and customers became more sophisticated, most resellers were squeezed out. The survivors were companies that layered on high-value services like security, migration expertise, and DevOps consulting.
AI aggregators, Mowry suggests, could face a similar fate if they fail to add meaningful proprietary value beyond access.
Where the Opportunity Lies
Despite the cautionary tone, Mowry remains optimistic about AI-driven innovation. He is particularly bullish on “vibe coding” and developer platforms, pointing to startups such as Replit, Lovable, and Cursor as standout performers in 2025.
He also sees strong potential in direct-to-consumer AI tools, especially creative applications like Google’s AI video generator Veo, which can help film and TV students bring stories to life.
Beyond AI, Mowry highlighted biotech and climate tech as sectors benefiting from massive data availability and renewed venture investment. In his view, startups that combine domain depth with data-driven intelligence — rather than thin abstraction layers — are best positioned for long-term growth.
As the AI boom matures, the message from Google’s startup chief is clear: differentiation and defensibility matter more than ever.