LLM wrappers and aggregators under pressure

The generative AI gold rush produced a wave of startups almost overnight. But according to a senior Google executive, not all of them are built to last.

Darren Mowry, who leads Google’s global startup organisation across Cloud, DeepMind and Alphabet, says two once-popular models now have their “check engine light” on: LLM wrappers and AI aggregators.

Speaking on the Equity podcast, Mowry cautioned that thin layers built on top of foundation models are no longer enough to win.

The problem with thin wrappers

LLM wrappers are startups that package existing large language models — such as OpenAI’s GPT or Google’s Gemini — into user-friendly tools aimed at specific use cases.

In the early days of the AI boom, that was often sufficient. Companies could bolt a clean interface onto a powerful model and quickly attract users. But Mowry argues that the market has matured.

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

To survive, startups need defensible advantages. That could mean:

  • Proprietary datasets
  • Deep vertical specialisation
  • Embedded workflows that create switching costs
  • Strong intellectual property beyond prompt engineering

Examples of wrapper models with stronger positioning include:

  • Cursor, a coding assistant built around GPT
  • Harvey AI, focused on legal services

The difference, Mowry suggests, lies in building moats that extend beyond the base model itself.

Aggregators face margin squeeze

AI aggregators — platforms that combine multiple models under a single interface or API — are also under scrutiny.

Companies like Perplexity AI and OpenRouter provide access to different LLMs while offering orchestration tools, monitoring, or governance layers.

But Mowry’s advice to new founders is blunt: “Stay out of the aggregator business.”

As model providers expand their own enterprise features, including tooling and governance capabilities, third-party aggregators risk being squeezed out. The dynamic echoes the early days of cloud computing, when startups resold infrastructure from hyperscalers only to be displaced once those platforms built direct enterprise solutions.

In short, middlemen without differentiated services may struggle to maintain pricing power.

Lessons from the cloud era

Before joining Google Cloud, Mowry held leadership roles at AWS and Microsoft, witnessing firsthand how early cloud resellers faded when customers learned to work directly with providers.

The survivors were those that added tangible value:

  • Security layers
  • Migration services
  • DevOps expertise
  • Industry-specific integrations

The parallel to today’s AI market is clear: orchestration alone may not be enough.

Where growth still looks strong

Despite the caution, Mowry remains optimistic about several AI segments.

Developer platforms and so-called “vibe coding” tools had a breakout year in 2025. Startups like:

  • Replit
  • Lovable
  • Cursor

have drawn significant investment and customer momentum.

He also sees promise in direct-to-consumer AI applications, including creative tools powered by Google’s video generator Veo, as well as in data-heavy sectors like biotech and climate tech.

In these areas, startups are not merely packaging models — they are embedding AI into workflows that unlock new forms of value.

Summary

Google executive Darren Mowry has issued a clear warning to AI founders: thin LLM wrappers and pure-play aggregators may struggle as the market matures and model providers move up the stack. Startups that survive will need durable moats, vertical depth and genuine intellectual property. As the AI sector enters its next phase, differentiation — not access — is emerging as the real currency.