The surge in generative AI has birthed numerous startups, but as the market stabilizes, two popular business models appear to be heading toward obsolescence: LLM wrappers and AI aggregators. Darren Mowry, head of Google’s global startup organization across Cloud, DeepMind, and Alphabet, shared insights indicating that startups employing these models may be facing significant challenges.
LLM wrappers are organizations that integrate existing large language models, such as Claude, GPT, or Gemini, into a product or user experience aimed at solving specific issues. For example, a startup might develop AI tools to assist students in their studies. However, Mowry cautions that relying solely on these back-end models without unique offerings is becoming untenable. He emphasized, “If you’re merely white-labeling the model, the industry is losing patience.”
To thrive, startups must cultivate robust differentiation—either through broad applications or a clear focus on niche markets. Successful examples of LLM wrappers include Cursor, a coding assistant powered by GPT, and Harvey AI, which focuses on legal assistance.
The landscape has shifted since mid-2024 when OpenAI first introduced its ChatGPT store. Startups can no longer simply add a user interface on top of existing models; they must deliver genuine value.
AI aggregators, a subcategory of LLM wrappers, compile multiple models into a single interface or API, enabling users to query different models based on their needs. While some have gained traction, Mowry warns newcomers to avoid the aggregator space, noting that this segment is currently stagnant. According to him, users seek intrinsic value—a compelling intellectual property layer—to ensure efficient routing to the appropriate model.
Drawing parallels to the evolution of cloud computing, Mowry notes that early cloud startups, which resold AWS infrastructure, struggled to compete as Amazon began to provide its own enterprise tools. Only those that offered significant additional services, such as security or DevOps consulting, managed to survive.
Currently, AI aggregators are encountering similar pressures as model providers continue to enhance their own capabilities, potentially marginalizing intermediaries.
In contrast, Mowry remains optimistic about the future of vibe coding and developer platforms, citing successful startups like Replit, Lovable, and Cursor as models that attract significant investments and customer engagement. He also identifies a promising future for direct-to-consumer technologies, highlighting opportunities for students in film and television utilizing Google’s AI video generator, Veo.
Beyond AI, Mowry sees considerable potential in biotech and climate tech, driven by increasing venture capital investment and unprecedented access to data, enabling startups to generate value in innovative ways.
