The economics of intelligence-as-a-service

What happens when reasoning becomes a measurable, tradable product?

The idea that “intelligence” — via LLMs or agents — can be priced and consumed like electricity changes how we think about value creation. We move from paying humans for cognitive work to paying APIs for cognitive units. To really grasp this shift, you need to unpack how cost, value, and scalability interact.

Key questions:

• What is the correct unit of intelligence : token, inference, reasoning step, or successful task?
• How does the cost curve of intelligence behave over time? (e.g., cost per 1K tokens, cost per solved benchmark task)
• What drives these costs down? model reuse, algorithmic efficiency, hardware improvements, or scale economies?
• Where is the value capture today? in the model, the data, or the workflow that embeds it?
• How does commoditization affect innovation: does cheaper intelligence expand use cases or erode differentiation?
• How do elasticity and marginal cost behave when intelligence is abundant? does demand scale faster than cost declines?