Jakarta, INTI - The surge of generative AI has led to the creation of new startups almost every minute. However, over time, two AI startup business models that were once highly popular are now predicted to struggle to survive and may lose investor interest in the future.
LLM Wrappers Struggle Without Real Differentiation
These two models are LLM wrappers and AI aggregators. An LLM wrapper refers to startups that build products on top of existing AI models such as GPT, Claude, or Gemini, but lack meaningful intellectual property (IP) of their own. One example includes startups that use AI tools to help students with learning.
Darren Mowry, Vice President of Google Cloud who oversees the company’s global startup programs across Google Cloud, DeepMind, and Alphabet, stated that if a business model simply “wraps” an already powerful AI model with a new interface but offers no real differentiation, the industry’s interest will gradually fade.
“To grow and succeed, a startup must have a strong and broad competitive advantage, either through horizontal differentiation or by building something highly specific for a vertical market,” Mowry said, as reported by TechCrunch.
According to him, two LLM wrapper startups that demonstrate strong differentiation in this category are Cursor, a GPT-based coding assistant, and Harvey AI, an AI assistant designed for the legal industry.
Why AI Aggregators May Lose Momentum
The second AI business model predicted to lose traction in the future is the AI aggregator. This model brings together multiple AI systems into a single interface or API.
Companies operating in this space typically provide an orchestration layer that includes monitoring, governance, or evaluation tools. Examples include Perplexity, an AI-powered search startup, and OpenRouter, a developer platform that enables access to various AI models through a single API.
In the early stages, this model appeared promising. However, Darren Mowry noted that such an approach alone is not enough for companies aiming to scale their businesses.
The reason is simple: users are now looking for more than just access to multiple AI models. They expect intelligent routing that combines contextual understanding and domain expertise, rather than a platform that merely aggregates models. While many platforms of this type continue to emerge, Mowry’s message to new startups is clear: avoid building pure aggregator businesses.
In general, aggregators have shown limited growth recently because users increasingly demand integrated intellectual property that can guide them to the most appropriate AI model at the right time based on their specific needs.
Mowry has spent decades in the cloud computing industry. Before joining Google Cloud, he gained experience at AWS and Microsoft, giving him deep insight into the evolution of technology business models.
According to Mowry, the current AI landscape resembles the early days of cloud computing in the late 2000s and early 2010s, when Amazon’s cloud business began expanding rapidly. At that time, many startups emerged to resell AWS infrastructure, positioning themselves as easier entry points that offered tools, billing consolidation, and customer support.
However, as Amazon developed its own enterprise tools and customers became more capable of managing cloud services directly, most of those startups eventually disappeared. The ones that survived were those that added real value, such as security solutions, cloud migration services, or DevOps consulting.
Conclusion
The rapid rise of generative AI has created countless startup opportunities, but not all business models are built for long-term success. According to Google Cloud’s Darren Mowry, startups that rely solely on wrapping existing AI models or aggregating multiple systems without developing their own intellectual property may struggle to remain relevant. As the AI ecosystem matures, the companies most likely to thrive will be those that offer clear differentiation, deep domain expertise, and real technological value rather than simply providing access to existing tools.
Read more: New GPT-5.4 Model by OpenAI is Its Most Capable Model Yet