Google’s Mowry Flags Risk for Two AI Startup Models

Darren Mowry, vice president of Google Cloud’s global startup organization across Cloud, DeepMind and Alphabet, warns that “LLM wrapper” and “AI aggregator” startup models may not survive as the generative AI market matures.

Darren Mowry, vice president of Google Cloud’s global startup organization across Cloud, DeepMind and Alphabet, told TechCrunch that two categories of artificial intelligence startups show early signs of structural weakness and face viability challenges as foundational AI models become more capable and market expectations evolve.

Mowry identified startups that act as so-called “LLM wrappers” as one group under pressure. According to Mowry, these companies build lightweight product or user-experience layers atop large language models such as those developed by OpenAI, Anthropic and Google, without developing substantial proprietary technology or defensible differentiation. He said that relying on underlying models to deliver core functionality can signal a lack of sustainable competitive advantage, with the industry losing patience for products that are essentially branded interfaces on another company’s model.

Mowry explained that thin wrappers often fail to embed deep intellectual property or domain-specific expertise, which he described as necessary for a startup to “progress and grow.” He cited examples of companies that have managed to build deeper moats through specialized workflows or domain integration, such as Cursor, a coding assistant powered by GPT, and Harvey AI, a legal assistant, but highlighted that such cases are exceptions rather than the norm.

A second model Mowry flagged is that of AI aggregators, which typically provide platforms or application programming interfaces that compile access to multiple large language models and route queries across them. While these aggregators may add features such as monitoring, governance or evaluation tooling, Mowry said their value proposition is weakening because developers and users increasingly seek integrated intellectual property and meaningful differentiation rather than simple multi-model access layers. He characterized the current state of these businesses as having their “check engine light” on, indicating looming challenges.

Mowry explicitly advised new founders to “stay out of the aggregator business,” arguing that as major model providers expand enterprise capabilities, governance layers and optimization tools, the margins and growth opportunities available to intermediary aggregators are compressing. He said that users want systems that embed expertise and route tasks to the right model based on need, rather than undifferentiated compute or rudimentary routing logic.

In detailing his perspective, Mowry drew a parallel between the current situation in AI and the early days of cloud computing. He noted that a wave of startups emerged in the late 2000s and early 2010s to resell infrastructure from major cloud providers such as Amazon Web Services, bundling tooling and support services. Over time, as cloud providers built out enterprise features and customers grew more self-sufficient, many of those intermediary startups were squeezed out, with only those adding substantive services surviving.

Mowry’s remarks were made on the TechCrunch Equity podcast, hosted by TechCrunch senior reporter Rebecca Bellan, where he discussed broader trends in AI startups and the pressures facing different business models within the ecosystem.

Beyond the cautionary assessment of wrappers and aggregators, Mowry highlighted areas he sees as exhibiting stronger potential. He singled out “vibe coding” tools and developer platforms, noting that several such companies had strong performance in 2025 with significant investment and customer traction. He also pointed to direct-to-consumer applications of powerful AI tools, including creative use cases such as video generation, as well as sectors outside of core AI that may benefit from abundant data, such as biotech and climate technology.

Mowry’s position reflects a shift in expectations within the generative AI market from rapid proliferation of startups exploiting easy access to foundation models toward greater emphasis on products with measurable differentiation, defensible intellectual property and integration into specific workflows or domains.