Challenges for AI Startups: A Google Executive's Warning on LLM Wrappers and Aggregators
A Google executive warns that startups building on large language models (LLMs) are facing significant challenges.
The AI startup landscape is going through a phase of consolidation, with some business models showing signs of obsolescence. In particular, startups that rely on large language models (LLMs) are facing significant challenges. Darren Mowry, Google's vice president responsible for global startups across Cloud, DeepMind and Alphabet, recently highlighted these difficulties, indicating that startups that merely "wrap" existing models or act as AI aggregators may not survive in the long term.
LLM Wrappers: Challenges and Limitations
Startups operating as "LLM wrappers" focus on adding user interfaces or product layers on top of existing language models, such as GPT or Gemini, to solve specific problems. For example, a startup might use AI to help students study. However, Mowry observes that if a startup merely "wraps" a foundation model without adding distinctive value, the industry has little patience for this approach. The key to success lies in building "deep, wide moats" that are differentiated horizontally or specific to a vertical market.
AI Aggregators: A Declining Model
Another declining business model is that of "AI aggregators". These startups aggregate multiple LLMs behind a single interface or API layer, routing queries across different models and giving users access to several models. However, Mowry warns that aggregators are not seeing much growth or progress, since users require "no embedded intellectual property" to ensure that queries are routed to the right model at the right time, according to their needs, and not because of underlying compute or access constraints.
Implications for AI Startups
These developments suggest that AI startups need to evolve beyond business models that rely solely on foundation models or on aggregating existing models. To thrive, it is essential to develop products with distinctive added value, own unique intellectual property and focus on specific market niches. Furthermore, deep integration into vertical workflows and offering complete solutions can provide a sustainable competitive advantage.
Conclusion
- Necessary innovation: Startups must go beyond using foundation language models, developing unique and differentiated solutions.
- Vertical focus: Concentrating on specific market niches and verticals can offer opportunities for sustainable growth.
- Intellectual property: Building and protecting intellectual property is essential to differentiate in the competitive AI market.
- Integration into workflows: Offering solutions that integrate deeply into existing workflows can increase adoption and customer loyalty.
- Market adaptation: Startups must be agile and ready to adapt to rapid changes in the AI market to remain competitive.