Get your free personalized podcast brief
We scan new podcasts and send you the top 5 insights daily.
Despite rapid revenue growth selling training and post-training data to foundation model labs, data companies face questionable terminal value. Kanu Gulati points out that once large models learn and synthesize external datasets, they often no longer require ongoing input. Furthermore, open-weight models rapidly distill advances from frontier labs, eroding the proprietary value of that training data for enterprises unless founders continually invent new data categories.
Related Insights
AI companies built to fill feature gaps on top of foundation models are at high risk. As core models rapidly improve, they often absorb these adjacent features, disintermediating the "wrapper" companies. Their early-adopter customers are also the quickest to switch to better tools.
Startups building on OpenAI or Anthropic APIs face a major platform risk. Their usage data trains the underlying foundational models, enabling the platform owners to eventually absorb their features natively and make the startups obsolete.
Creating frontier AI models is incredibly expensive, yet their value depreciates rapidly as they are quickly copied or replicated by lower-cost open-source alternatives. This forces model providers to evolve into more defensible application companies to survive.
The pace of AI development means a startup's competitive advantage can be erased overnight by the next model release from a major lab like Google or Anthropic. Dr. el Kaliouby stresses that true defensibility now requires more than just a proprietary algorithm; it demands unique data, distribution, or IP that cannot be easily replicated.
A vertical AI startup is extremely vulnerable if its core offering can be easily replicated by the foundational model it's built upon. True defensibility comes from integrating unique, proprietary data sources or solving non-obvious workflow problems that the base model cannot simply be prompted to do.
As powerful foundation models like GPT become commodities, a company's defensible moat is no longer its algorithm but its proprietary, hard-to-replicate dataset. The value lies in the unique data you can feed into these common models, as it's the one thing that is not easily found or replaced online.
As AI application layers become easier to clone, the sustainable competitive advantage is moving down the tech stack. Companies with unique, last-mile user interaction data can build proprietary models that are cheaper and better, creating a data flywheel and a moat that is difficult for competitors to replicate.
Frontier model providers like OpenAI and Anthropic are under immense pressure to monetize beyond tokens, forcing them to compete at the application layer. Startups building on their platforms are providing valuable data that will be used to create competitive products, effectively training their own replacement.
While data labeling companies show massive revenue growth, their customer base is often limited to a few frontier AI labs. This creates a lopsided market where providers have little leverage, compete on price, and are heavily dependent on a handful of clients, making the ecosystem potentially unstable.
AI products in easily verifiable domains (like coding) will be dominated by large labs. Startup defensibility lies in generating unique data where success is hard to verify automatically, like genuine human learning. This requires real user session data to create a data flywheel that frontier models cannot replicate.
