OpenAI, as a representative of large AI model companies, reflects to some extent the future of other AI model companies. The relationship between OpenAI's applications and models also reflects potential industry trends in AI's future.
Current AI represented by large models has made significant progress compared to 2010, with the core difference being that there are now truly profitable products. Among overseas SaaS tools, Heygen is a representative example, reportedly earning about $20 million annually without having its own model.
The most profitable product is estimated to be Microsoft Copilot. Although it's also a wrapper product without its own model, its revenue scale is already considerable. According to some predictions, Copilot's revenue might be about twice that of OpenAI. Such products have other benefits for Microsoft, promoting sales of other products.
The core components of Copilot include:
- Entry point and account
- Microsoft Graph (representing the full set of domain data)
- Large AI model
Above this is an overall systemic approach, corresponding to user experience including response speed, function fluidity, etc. The large model plays the role of an engine, with the entire system responsible for fitting its output to user data and returning it to the user.
In this structure, the product side (Microsoft) has more say. For large customers, model companies need to provide on-site services such as private deployment.
The IT industry has long had an invisible operating rule: technology creates new value, products take the profits. Or: hardware creates a new world, software takes the profits.
From this perspective, looking at OpenAI, various applications are like making cars, while OpenAI is making engines. But OpenAI faces challenges:
- Technology based on models may depreciate rapidly
- Huge investments are difficult to reduce
- Limited channels make monetization difficult
OpenAI's fundamental problem is its poor industry position, one that can only seek patronage. The core driving force is to always stand in the leading position. Once slowed or weakened, after several twists and turns, it may eventually be swallowed by giants.
For OpenAI, perfecting large AI models is just the starting point; the ultimate industry position depends on whether it can break through this starting point. OpenAI clearly recognizes this, hence continuously launching new products and services, but overall its product strength seems weak, putting it in a dangerous position.
OpenAI needs to make breakthroughs in products, but its team seems more suited for R&D rather than product development. To break through the limitations of its industry position, the key lies in products, but the current team configuration may struggle to develop products capable of breaking through.
In choosing between ToB and ToC, OpenAI tends to choose the C-end. Strategically, this is undoubtedly correct, because on the B-end, companies like Microsoft have already built complex product stacks that are difficult to disrupt with a single product.