Firms face significant adjustment costs when switching between different AI models.
These costs stem from optimizing prompts, workflows, and internal tools for a specific model.
This friction helps incumbent closed-source models maintain higher prices.
The market for AI services is differentiated, not a simple commodity market.
Heterogeneous demand across various use cases prevents a single dominant AI product.
Examples
- A software firm generating code needs different AI features than a law firm summarizing depositions.
- A retailer translating product listings has unique latency and reliability requirements.
- Specific use cases like finance, health, and academia demand tailored AI solutions.
- No single AI creator dominates all categories, showing diverse market needs.
- The leading AI model in categories like legal and marketing changes frequently.
Do AI model adjustment costs really realize the diverse demands across LLM use cases, or is that a bit too absolute?
In this market, switching costs certainly play a role in provider lock-in, but the local context often dictates far more.
For example, even if technical friction was zero, a French company might still choose a specific LLM provider for data sovereignty reasons to comply with strict GDPR regulations, making general 'heterogeneous demand' less about technical switching and more about legal necessity.
So, while a factor, these costs are conditional, not always the decisive mechanism for varied demand.