Google VP warns that two types of AI startups cannot survive

The amount of AI resulted in a one-minute startup. But as the dust settles, two once-hot business models are looking like cautionary tales: LLM wrappers and AI aggregators.

Darren Mowry, who leads Google’s global startup organization for Cloud, DeepMind, and Alphabet, says startups and hookups have a “light engine”.

LLM wrappers are basically frameworks that wrap existing major languages, such as Claude, GPT, or Gemini, with an object or UX layer to solve a specific problem. An example would be a starting point for that uses AI to help students learn.

“If you’re just relying on the background color to do all the work and you’re just writing in white, the industry doesn’t have a lot of patience anymore,” Mowry said of this. Equity’s weekly session.

Wrapping up “very little information surrounded by a Gemini or GPT-5” isn’t a sign that you’re not distinguishing yourself, says Mowry.

“You have to have deep, wide channels that are separated horizontally or something real in the vertical market” to start “moving forward and growing,” he said. Examples of deep LLM wrapper models include Cursor, a GPT-powered coding assistant, or Harvey AI, a legal AI assistant.

In other words, developers can no longer expect to slap a UI on top of GPT and start adopting their products, as they could, perhaps within 2024 when OpenAI. launched his ChatGPT store. The challenge now is to build a sustainable value for money.

AI aggregators are a subset of wrappers – they are startups that combine multiple LLMs into a single interface or API layer to provide queries across all models and give users access to multiple models. These companies often offer an orchestration suite that includes management, governance, or monitoring tools. Consider: AI Perplexity Research or the OpenRouter platform, which provides access to multiple types of AI through a single API.

Although many platforms have started, Mowry’s words are clear for upcoming startups: “Don’t get caught up in the aggregator business.”

In general, integrators do not see growth or progress today because, they say, users want “some built-in intelligence” to ensure that they are sent to the right model at the right time according to their needs – not because of behind-the-scenes calculations or constraints.

Mowry has been in the cloud game for years, cutting his teeth at AWS and Microsoft before setting up shop at Google Cloud, and he’s seen it done. He said that what is happening today reflects the early days of cloud computing in the late 2000s/early 2010s when Amazon’s cloud business started.

At the time, startups began reselling AWS infrastructure, marketing themselves as a simple entry point that offered tools, including billing, and support. But as Amazon developed its own business tools and customers learned to manage cloud services directly, many of the startups were dropped. The only survivors are those who have added specific functions, such as security, migration, or DevOps consulting.

Today’s AI integrators face similar challenges as model providers expand into their own businesses, potentially putting out middlemen.

For him, Mowry is focused on platforms for creating vibes and software developers, who had a consistent year in 2025 with startups like Replit, Lovable, and Cursor (all Google Cloud customers, on Mowry) that attract more money and attract customers.

Mowry also expects strong growth in direct-to-consumer technology, in companies that put powerful AI tools in the hands of customers. He also mentioned the opportunity for film and TV students to use Veo’s AI video generator to bring news.

Beyond AI, Mowry also thinks that the natural and climate sciences are having a moment – ​​given the amount of money flowing into these two industries and the start-ups with an “incredible track record” they could reap real benefits “in ways we couldn’t before.”