Token Costs: From Free-As-You-Go to Pay-As-You-Go to "Do You Really Know?"

Headlines are dominating the news: "When are we going to start to see profits from AI?" As token pricing is changing, people are feeling the pinch. However, it's well known that the cost of AI does not accurately reflect the true cost of AI. Even with new metered pricing, the true cost of the tokens you use is far greater than the actual cost you are paying.

General users are getting the deal of a lifetime even if it doesn't feel that way in recent months. Businesses that are experimenting with AI-assisted coding are starting to feel the pain of token bills despite the fact those costs are far less than the true cost. Companies looking to integrate agentic AI are holding back due to rising costs, among many other reasons such as actual need and privacy/security concerns. Frontier labs are shelling out cash in record amounts while pondering how to generate profits.

Now we are starting to see competition with more and more open-source (cheaper) LLMs hitting the market, be they global LLMs from China or smaller, task-specific models from various labs. This is leading to speculation that the "charge for it later" strategy will fall on its face, causing frontier labs (mainly OpenAI and Anthropic) to also fall on their faces.

The biggest problem with all this AI stuff is the way it was all initially rolled out - free! Get them addicted and they will pay later. At the end of the day, AI may be another example of how free stuff does nothing but cause a bunch of problems.

I wanted to share a video I found in the Blackline YouTube feed that explains the AI cost crisis. It's properly titled "The AI Cost Crisis Has Started." It's a really good explanation of all of this and it explains the polarity between feeling taxed to death for a service that should be costing far more and looming debt and uncertainty for not charging enough for your service.