The rapid rise of generative AI has led to the emergence of numerous startups, but as the market evolves, two business models are emerging as potential pitfalls: LLM wrappers and AI aggregators.
Darren Mowry, head of Google's global startup organization encompassing Cloud, DeepMind, and Alphabet, warns that these types of startups may be facing significant hurdles. According to Mowry, those relying on existing large language models (LLMs) without substantial differentiation are at risk. "If you're merely relying on the backend model to perform all the functions, the industry is losing patience," he stated.
LLM wrappers are startups that utilize existing models like Claude, GPT, or Gemini, layering a product or user experience to address specific needs. For instance, a startup that employs AI to enhance student study habits exemplifies this approach. However, Mowry emphasizes that simply applying a user interface over a model won't suffice in today's competitive landscape.
"Wrapping thin intellectual property around models like Gemini or GPT-5 indicates a lack of unique value," he noted. To thrive, startups must establish robust, distinctive advantages that serve either broad markets or specific niches. Examples of successful LLM wrappers include Cursor, which offers a coding assistant powered by GPT, and Harvey AI, focused on legal assistance.
Looking ahead, Mowry suggests that startups can no longer expect to gain traction by merely adding a UI to existing AI models, as was possible in earlier stages of AI development. The key now lies in building sustainable value into their offerings.
AI aggregators, which compile multiple LLMs into a single interface or API for user queries, face similar challenges. These platforms often provide orchestration layers that include monitoring and governance tools. However, Mowry advises new entrants to avoid the aggregator model, as these startups are currently struggling to grow. Users desire integrated intellectual property that ensures optimal model routing based on their needs.
Reflecting on his extensive experience in cloud services, Mowry draws parallels between the current AI landscape and the early days of cloud computing. In the late 2000s, many startups emerged to resell AWS infrastructure, only to be outpaced as Amazon developed its own enterprise solutions. The startups that survived were those that offered genuine value-added services.
Today, AI aggregators are similarly squeezed as model providers enhance their offerings, potentially sidelining middlemen. Nevertheless, Mowry expresses optimism about the future of vibe coding and developer platforms, citing a record year for startups like Replit and Cursor, which have attracted significant investment.
He also foresees strong growth in direct-to-consumer tech, particularly in applications that empower users with advanced AI tools, such as Google's AI video generator, Veo, which offers exciting opportunities for film and TV students.
Beyond AI, Mowry believes that sectors like biotech and climate tech are poised for expansion, driven by substantial venture investments and the availability of vast data resources that enable startups to create unprecedented value.
