AI Application Startups Face Rapid Obsolescence by Foundational Model Providers

In the rapidly evolving landscape of artificial intelligence (AI), a compelling investment thesis is emerging that suggests most AI application startups will struggle to survive and scale. This thesis, articulated by Yishan Wong, highlights the formidable challenge posed by foundational model providers-large technology companies with extensive resources and agile capabilities. According to Wong, these big players are not slow incumbents but rather fast-moving entities that can quickly incorporate new functionalities into their existing platforms, rendering many AI application startups obsolete.

Why it Matters

The implications of this thesis are significant for investors and entrepreneurs in the AI space. Traditional models of startup growth, which rely on a period of relative stability to build and scale a business, may not apply here. The rapid pace of innovation and the dynamic nature of foundational AI technologies mean that startups have a narrow window-approximately 12-18 months-to generate substantial cash flow or secure an acquisition by one of the major players.

Key Risks

1. Rapid Obsolescence: Foundational model providers are continuously advancing their capabilities, which can quickly render new applications obsolete. This rapid pace of change leaves little room for startups to establish a sustainable business model.

2. High Competition: The large technology companies have significant resources and can swiftly replicate or improve upon the functionalities offered by smaller startups. This competition makes it difficult for startups to maintain a competitive edge.

3. Unstable Foundation: Unlike previous technological waves, AI foundational technologies are still in flux. The lack of stability means that applications built on these foundations may need constant updates, making it challenging to create long-term value.

The Opportunity

Despite the challenges, there are still opportunities for AI application startups to succeed. According to Wong, the best strategy is to focus on highly specialized fields with unique and specific data barriers. These niches can provide a protective moat against the rapid advancements of foundational model providers. Examples include:

• Hardware and Real-World Data: Applications that rely on hardware or real-world data, such as those in healthcare, manufacturing, or robotics, may have a better chance of survival due to the complexity and uniqueness of their data sets.

• Acquisition Potential: Startups can aim to create applications that are good enough to attract the interest of large technology companies. An acquisition by one of these players can provide a viable exit strategy for founders and investors.

Case Studies and Data

Wong's thesis is supported by several observations from the AI industry:

• 12-18 Month Window: Startups have a limited time frame to generate significant cash flow before their applications become obsolete. This window is much shorter than in previous technological waves, such as the PC or mobile internet revolutions.

• Specialized Data Barriers: Applications that leverage unique data sets, particularly those related to hardware or real-world phenomena, are more likely to survive and thrive. For example, a startup focusing on AI-driven medical imaging may have a competitive advantage due to the specialized nature of its data.

Conclusion

The rapid advancement of foundational AI technologies presents both challenges and opportunities for startups in the application space. While the risk of obsolescence is high, focusing on highly specialized fields with unique data barriers can provide a path to success. For investors, this thesis suggests a need for careful consideration of the timing and focus of investments in AI application startups.