Is AI Token Revenue More Durable Than Expected?

Jeff Lawson says AI token revenue has proven a lot more durable than he would have guessed three or four years ago. At the very beginning of the era he expected more of a race to the bottom. Token prices over that period did fall that way. Demand so far outstripped the pricing dynamics that the drop did not matter, and what carried the revenue was use case expansion.

He frames selling intelligence in a utility model as, in some ways, the ultimate usage model, and he calls the token business incredible product-market fit for one of the ultimate markets on the planet, intelligence. The same usage model had already turned communications into something people buy by the unit. Stripe did it for payments, which sit even closer to the core than communications. AWS did it for compute and storage, so revenue becomes a function of human activity built on the basic building blocks of technology. Intelligence is the next step, and it is no longer about apps.

Twilio was already used to selling in a utility way, passing carrier costs through. Tokens would have fit both the technology and the business model, and he wanted Twilio in that market. Companies that resell, mark up, or embed tokens have a hard time making the shift, and a structurally lower gross margin from reselling tokens makes it harder still. None of the companies are public, which is why durability was the open question.