Two business models that surged during the generative AI boom are now facing tougher questions, as Darren Mowry, who leads Google’s global startup organization across Cloud, DeepMind, and Alphabet, said LLM wrappers and AI aggregators show warning signs as the market shifts toward products with clearer differentiation.
Why LLM Wrappers Are Under Pressure
LLM wrappers are startups that take existing large language models such as Claude, GPT, or Gemini and add a product or user experience layer to address a specific task, such as helping students study. Mowry said on this week’s episode of Equity that startups relying on the back-end model to do most of the work, and effectively white-labeling it, are meeting less patience from the industry than before.
He said that wrapping “very thin intellectual property around Gemini or GPT-5” signals a lack of differentiation. In his view, startups need “deep, wide moats” that are either horizontally differentiated or focused on a specific vertical market to “progress and grow.” He cited Cursor, a GPT-powered coding assistant, and Harvey AI, a legal AI assistant, as examples that fit this description.
Mowry said the environment has changed from mid-2024, when OpenAI launched its ChatGPT store and new products could gain traction by placing a simple interface on top of an existing model. He said the challenge now is building sustainable product value.
Aggregators And The Limits Of Model Access
AI aggregators are a subset of wrappers. These startups combine multiple models into a single interface or API layer that routes queries across systems and gives users access to more than one model. They often include orchestration features such as monitoring, governance, or evaluation tools. Examples include the AI search company Perplexity and the developer platform OpenRouter, which offers access to several models through one API.
While many of these platforms have gained ground, Mowry said incoming startups should “stay out of the aggregator business.” He said aggregators are not seeing much growth or progression because users want “some intellectual property built in” to ensure they are routed to the right model at the right time based on their needs, rather than relying on compute access or behind-the-scenes constraints.
A Parallel With Early Cloud Computing
Mowry said the current situation resembles the early days of cloud computing in the late 2000s and early 2010s, when Amazon’s cloud business began to scale. At that time, a group of startups emerged to resell AWS infrastructure, offering tooling, billing consolidation, and support as easier entry points.
When Amazon later built its own enterprise tools and customers learned to manage cloud services directly, many of those resellers were pushed out. The companies that remained were those that added services such as security, migration, or DevOps consulting. Mowry said AI aggregators now face similar margin pressure as model providers expand into enterprise features themselves.
Where Mowry Sees Growth
Mowry said he remains optimistic about developer platforms and what he called vibe coding. He said 2025 was a record year for that category, with companies such as Replit, Lovable, and Cursor, which he said are Google Cloud customers, attracting investment and customer traction.
He also said he expects growth in direct-to-consumer technology that places AI tools in the hands of users. As an example, he pointed to film and TV students using Google’s AI video generator Veo to produce projects.
Beyond AI, Mowry said biotech and climate tech are also seeing increased activity, both in venture investment and in the amount of data available to startups, which he said enables new ways to create value.
Featured image credits: Wikimedia Commons
For more stories like it, click the +Follow button at the top of this page to follow us.