Google senior executive says these startups face a tough road ahead

Artificial Intelligence

• LLM wrapper startups face tougher market conditions
• AI aggregators face pressure as model providers enhance features
• Startups need proprietary data or deep integration to stand out

Did our AI summary help?

The generative AI boom minted a startup a minute. Slap a sleek interface on top of GPT, add a niche use case, and funding often followed.

That era, says Darren Mowry, may be winding down.

Mowry, who leads Google’s global startup organisation across Cloud, DeepMind and Alphabet, argues that startups built purely as LLM wrappers or AI aggregators are entering a tougher phase. Speaking on the Equity podcast, he said many such companies now effectively have their “check engine light” on.

The problem with thin wrappers

LLM wrappers are startups that build a product layer on top of existing large language models such as Claude, GPT-5 or Gemini. The idea is simple: take a powerful foundation model and tailor it to a specific audience, such as students or marketers.

That worked well in the early days of the AI surge. But Mowry believes investors and customers now expect more.

“If you’re really just counting on the back end model to do all the work and you’re almost white-labelling that model, the industry doesn’t have a lot of patience for that anymore,” he said.

In his view, startups need defensible moats. That could mean proprietary data, deep workflow integration, or strong vertical expertise. Simply wrapping “very thin intellectual property” around a frontier model is no longer enough.

There are exceptions. Tools such as Cursor, a GPT-powered coding assistant, and Harvey AI, which focuses on legal workflows, represent what he considers deeper, more defensible plays.

Aggregators under pressure

AI aggregators, a subset of wrappers, face even steeper challenges. These startups aggregate multiple models into one interface or API layer, routing queries across different providers. Examples include Perplexity AI and OpenRouter.

While such platforms offer orchestration, monitoring and governance tools, Mowry’s advice to new founders is direct: “Stay out of the aggregator business.”

His reasoning is structural. As model providers build their own enterprise features, governance layers and optimisation tools, they compress the margins available to middlemen. Users increasingly want embedded intellectual property and smarter routing based on their needs, not just access to multiple models.

A cloud déjà vu

Mowry has seen this pattern before. During the early days of cloud computing, a wave of startups emerged to resell AWS infrastructure, promising simpler tooling and consolidated billing. But as Amazon built more enterprise capabilities and customers grew more sophisticated, many of those intermediaries were squeezed out.

Only those that added genuine services, such as security, migration and DevOps consulting, survived. AI aggregators today risk a similar fate.

Despite his caution, Mowry remains optimistic about parts of the AI ecosystem. He is bullish on developer platforms and so-called vibe coding tools. Startups such as Replit, Lovable and Cursor have attracted significant investment and user traction.

He also sees strong potential in direct-to-consumer AI, particularly creative tools. Google’s AI video generator Veo, for instance, could open new possibilities for film and television students.

Beyond AI, Mowry highlights biotech and climate tech as sectors benefiting from unprecedented access to large, high-quality datasets.