First we paid for access. Next we will pay for every token
The big AI companies are already moving business customers from flat fees to paying for what they use. I think the rest of us are next. Here is why.
Here is a prediction. Within a few years, most of us will pay for AI the way we pay for electricity. A fixed fee to be connected, then a meter for everything we use.
The meter will count tokens.
A token is the unit an AI model reads and writes. It is a chunk of text, roughly four characters, or about three quarters of an English word. A hundred tokens is about seventy-five words. Every question you type and every answer you get back is counted in them.
For years, most people paid for AI with a simple monthly subscription. Twenty dollars and use it as much as you like, within reason. That model is starting to break, and you can see it in what happened this year.
Anthropic now bills its Enterprise plan in two parts. A seat fee for access, and separate usage charges based on actual token use, at the same rates developers pay.
Microsoft launched its Copilot Cowork agent in June 2026 with usage pricing on top of the Microsoft 365 Copilot licence. Each task is charged in credits worth one cent each. Microsoft's Charles Lamanna compared it to "filling up your gas tank at the pump".
GitHub's Copilot plans now include a set number of premium requests each month, then charge for each extra one.
The pattern is clear. Pay to get in. Then pay for what you use.
Why is this happening? Because the cost of running AI is not flat. A quick question costs very little to answer. An agent that works for twenty minutes, reading documents and writing reports, costs a great deal more. A flat price means light users pay for heavy users. As the tools get more capable and people use them for bigger jobs, that gap gets too wide to hide.
We have seen this before. Mobile phones went from pay-per-minute to unlimited plans, and data brought metering back. Electricity has always worked this way. Cloud computing was built on it.
So what does this mean for you?
First, AI will start to show up on budgets as a running cost, like power or fuel. Businesses will need someone watching the meter.
Second, efficiency will matter. The person who gets a good answer in one clear message will spend less than the person who takes twenty messages to get there.
Third, the price of a task will become visible. When you can see that a report cost three dollars to produce, you start asking whether it was worth three dollars. That is a healthy question.
I think the all-you-can-eat era of AI was a launch offer. The meter is coming. The sensible move is to learn how to use fewer tokens well, before you are paying for each one.
Sources: Claude Help Centre, Enterprise billing; Microsoft Copilot Cowork launch coverage, June 2026; GitHub Copilot plans; OpenAI Help Centre, "What are tokens".
