As venture capital continues to pour into artificial intelligence, not all AI startups are built to last. Darren Mowry, who leads Google’s global startup organization across Cloud, DeepMind and Alphabet, has a blunt message for founders building on top of large language models: if your product is little more than a thin wrapper around someone else’s model, your “check engine light” is already on.

Speaking on this week’s episode of the Equity podcast, Mowry said the era of easy traction for simple LLM-based products is largely over. Startups that rely almost entirely on back-end models such as OpenAI’s GPT family, Anthropic’s Claude or Google DeepMind’s Gemini without building meaningful intellectual property of their own are unlikely to see durable growth.

The End of the “Slap a UI on GPT” Era

LLM wrappers — startups that package an existing large language model with a user interface or workflow to solve a specific problem — once attracted rapid user growth. In mid-2024, when OpenAI launched its ChatGPT store, it was possible to gain attention by layering a clean UX on top of a powerful foundation model.

But that window has narrowed.

“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,”

Mowry said. Wrapping “very thin intellectual property around Gemini or GPT-5” signals a lack of differentiation, he added.

AI Is Redefining Startup Economics, Says Microsoft VP

The implication is clear: sustainable AI startups must build deeper product value, proprietary data, or vertical specialization. Otherwise, they risk being commoditized as foundation models improve and expand their own feature sets.

Why AI Aggregators Are Under Pressure

A subset of wrappers — AI aggregators — may face even steeper challenges. These companies combine multiple large language models into a single interface or API layer, routing queries across providers while offering monitoring, governance or evaluation tools.

Examples include AI search startup Perplexity AI and developer platform OpenRouter, which gives developers access to multiple AI models through one API.

While such platforms have gained visibility, Mowry’s advice to incoming founders is direct:

“Stay out of the aggregator business.”

According to Mowry, aggregators are not seeing strong growth or progression because users increasingly expect built-in intellectual property — smart routing based on real understanding of context and need — rather than simple model switching driven by compute availability or access constraints.

In his view, startups must build “deep, wide moats that are either horizontally differentiated or something really specific to a vertical market” to progress and grow.

A Familiar Pattern from the Early Cloud Era

Mowry draws a parallel with the early days of cloud computing in the late 2000s and early 2010s, when Amazon Web Services began to take off. At the time, numerous startups emerged to resell AWS infrastructure, positioning themselves as simpler gateways that offered billing consolidation, tooling and support.

But as Amazon Web Services built its own enterprise tools and customers became more sophisticated in managing cloud services directly, many of those resellers were squeezed out. The companies that survived were those that added real, defensible services such as security, migration expertise or DevOps consulting.

AI aggregators today face similar margin pressure. As model providers expand into enterprise features — including governance, monitoring and orchestration — middle layers that do not offer strong differentiation risk being sidelined.

Where Mowry Sees Opportunity

Despite his warnings, Mowry remains bullish on several AI-driven categories.

One is the wave of “vibe coding” and AI-native developer platforms. In 2025, startups such as Replit, Lovable and Cursor— all Google Cloud customers, according to Mowry — attracted significant investment and customer traction. These companies go beyond basic model access, embedding AI deeply into developer workflows and building sticky ecosystems around their tools.

Cursor, for example, represents what Mowry considers a stronger type of LLM wrapper: a GPT-powered coding assistant tightly integrated into software development environments, with product depth that extends well beyond a simple interface layer.

He also points to vertical AI plays such as Harvey AI, which targets the legal industry with specialized functionality. These startups build domain-specific capabilities, workflows and datasets that create defensible advantages.

Mowry expects continued growth in direct-to-consumer AI as well. He highlights the opportunity for creative students to use Google’s AI video generator Veo to bring stories to life — an example of powerful generative tools moving into the hands of everyday users.

Beyond AI infrastructure and applications, he sees biotech and climate tech as having a moment. Both sectors are benefiting from increased venture investment and unprecedented access to large datasets, enabling startups to create value in ways that were previously impossible.

The Bar for AI Startups Is Rising

The broader message from Mowry is not that AI opportunity is shrinking — but that it is maturing. As foundation models become more capable and cloud providers move up the stack, superficial differentiation is no longer enough.

Startups that merely wrap existing LLMs without building proprietary technology, deep integrations or vertical expertise may struggle to justify their existence. Those that develop meaningful intellectual property, embed themselves into critical workflows or harness unique data, however, could still define the next generation of AI companies.

Material by Iva Abadjievа

Image: Darren Mowry’s LinkedIn profile