The Changing Landscape of AI Startups: A Word of Caution
With the current boom in generative AI, startup companies are being formed in the blink of an eye. However, as the initial excitement settles, the viability of certain business models is being questioned. Two such models that were once considered revolutionary, LLM wrappers and AI aggregators, are now being viewed as cautionary tales.
As highlighted by Darren Mowry, the head of Google’s global startup organization for Cloud, DeepMind and Alphabet businesses, startups operating based on these particular models might have their “check light” on.
Understanding LLM Wrappers and Their Limitations
LLM wrappers are startups that leverage existing large language models like Claude, GPT or Gemini. They wrap these models with a product or user experience layer to tackle a specific problem. A classic example would be a startup that uses AI to aid students in their studies.
However, Mowry warns that this model may not be sustainable. “If you’re really just relying 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 opined on a recent episode of Equity.
According to Mowry, this approach lacks differentiation, which is key in today’s competitive tech landscape. “You have to have deep, wide moats that are either horizontally differentiated or something really specific to a vertical market” for a startup to “progress and grow,” he added. Examples of startups that have managed to differentiate include Cursor, a GPT-powered coding assistant, and Harvey AI, an AI legal assistant.
The Rise and Fall of AI Aggregators
AI aggregators are a subset of wrappers. These startups package multiple large language models into a single interface or API layer. This arrangement allows users to access multiple models at the same time. These companies typically provide an orchestration layer that includes monitoring, governance, or assessment tools. Examples include AI research startup Perplexity and development platform OpenRouter.
Despite initial traction, Mowry’s advice to new startups is clear: “Stay out of the aggregator business.” He explains that users prefer startups with “some built-in intellectual property” that can direct them to the right model at the right time based on their needs, rather than computational constraints or backend access.
Lessons from the Past
Based on his extensive experience in the cloud game with leading companies like AWS, Microsoft, and now Google Cloud, Mowry sees a potential parallel between the current state of AI aggregators and the early days of cloud computing. Back then, certain startups were set up to resell AWS infrastructure with additional support and tools. However, when Amazon introduced its own enterprise tools, many of these startups were squeezed out. The survivors were those who added real value with additional services, such as security, migration, or DevOps consulting.
AI aggregators face similar risks today as model providers start expanding their enterprise capabilities, thereby potentially marginalizing the intermediaries.
The Future of AI Startups
Despite these challenges, Mowry remains optimistic about certain sectors of the AI startup industry. He is particularly bullish on coding and development platforms, which had an impressive run in 2025. Startups like Replit, Lovable, and Cursor attracted major investment and customer interest.
Direct-to-consumer technology also shows promise, with companies leveraging powerful AI tools to enhance user experience. For instance, film and television students can now use Google’s AI video generator, Veo, to bring their stories to life.
Looking beyond AI, Mowry also identifies biotech and climate tech as sectors poised for growth. The substantial amounts of data available to startups in these sectors can be used to create real value in novel ways.
As the world of AI startups continues to evolve, the key to survival lies in differentiation, innovation, and the ability to deliver lasting value to users. To learn more about this topic, click here.
