AI Will Ultimately Make Money from Commissions

Tokens measure the cost of AI, not the value it creates.

The same number of tokens can draft a trivial email or help a company make a consequential decision. The value created is entirely different, yet AI companies currently charge almost the same price for both. They are selling the computation used to produce an output rather than the outcome itself.

This structure resembles electricity. A power company does not distinguish between a kilowatt-hour used to light a room and one used to manufacture semiconductors worth billions of dollars. Electricity enables enormous value, but the power company does not capture that value itself.

The same will be true if AI remains only infrastructure. Even if the entire economy uses AI, model providers may collect only the price of tokens. Model performance will keep improving while competition pushes token prices down. The value created by AI could grow even as the margins of the companies building it shrink.

The real business model of AI will therefore likely move from generating answers to completing work.

An AI that recommends an itinerary sells tokens. An AI that actually books the flights and hotels creates a transaction. An AI that drafts a customer-support reply is software; one that resolves the issue reduces labor costs. An AI that writes code sells usage, while one that fixes a bug and deploys the change sells an outcome.

Once AI controls the outcome, the basis of pricing changes. Instead of charging for tokens consumed, a company can charge for reservations completed, support cases resolved, costs reduced, or revenue increased. It can capture a portion of the economic value it creates rather than a few dollars per million tokens.

Ultimately, this is a commission-based business.

Traditional commission businesses make money by controlling the path of a transaction. Card networks process payments, travel agencies connect reservations, and advertising platforms connect purchase intent with sellers. If AI moves beyond recommendations and performs the search, comparison, negotiation, and purchase itself, AI becomes the route through which the transaction must pass.

Who pays the commission matters. If sellers pay, an AI has an incentive to recommend the product offering the highest commission rather than the best product. The structure of search advertising simply reappears. If users pay directly, or the AI receives a share of the money it saves them, the interests of the AI and the user become more closely aligned.

Pricing outcomes is harder than pricing tokens. A company must determine how much of an outcome was caused by AI and assume some responsibility for failure. The shift will therefore begin where results are easy to verify. A completed purchase, a resolved support case, or a realized cost saving can be measured relatively clearly.

Not every AI company can adopt this model. Foundation-model providers may remain like electric utilities that sell tokens. AI systems that directly perform customers’ work and transactions can collect commissions on top of them. The company with the best model may not necessarily be the company that makes the most money.

Competition in AI will ultimately be less about who generates the most tokens than about who controls execution at the moment an outcome is created. To earn in proportion to the value AI creates, companies must sell outcomes rather than tokens.