AI model developers are no longer content with the role of providers of access to their technologies. They are increasingly releasing their own products and beginning to take market share from those who were their customers just yesterday. Investor Chamath Palihapitiya, founder of the Social Capital fund, directly pointed to this trend as a structural shift in the economics of the AI business.
Selling tokens is no longer profitable
Labs, including OpenAI, are rapidly expanding the functionality of their chatbots. Previously, such companies mainly sold access to models via API, and others built applications on top of them. Now the developers themselves are embedding ready-made capabilities directly into their products.
The reason lies in the economics. Processing each new token — a fragment of text that the model reads or generates — costs less and less and brings in almost nothing. Making money from simply selling access to a model no longer works. The money is shifting to where there is a ready-made product.
Who will suffer first
The greatest risk is borne not by large SaaS companies, but by young AI startups. They currently spend significant funds on lab models while being completely dependent on them. Palihapitiya identified three areas that will suffer before the rest:
- legal services;
- sales tools;
- customer support.
Labs can clearly see how these startups use their models. Such data hints at which products are in demand, and replicating them is not difficult. Next in line are intermediary services that connect AI models to the programs people use every day — documents, cloud storage, email. But labs are already doing this themselves: editing Google Docs directly within OpenAI products and connecting Google Drive and Google Docs to chats turned out, according to the investor's assessment, to be very successful.
For investors, this is an alarming signal. A startup built entirely on someone else's model may at any moment face a competitor in the form of its own supplier.
Cryptalist analysis: The AI market is entering a phase of vertical integration, where infrastructure providers absorb the application layer and appropriate the added value. For venture capital, this means a reassessment of risks in portfolios of AI startups without their own data and technological moat. Only projects with unique datasets, distribution, or deep integration into industry processes that are inaccessible to base models will remain resilient.