There was a time when sending a text made you do quick math. You had 200 a month. You’d watch the number. You’d think twice before firing one off to a friend, because message 201 cost you extra.
Nobody does that anymore. And almost nobody remembers doing it.
I bring this up because I’m watching the AI industry walk straight into the same wall the telecom companies hit thirty years ago. Open up Copilot, Cowork, or a dozen other AI tools right now, and you run into metering. Pay per agent run. Pay per premium request. Burn through your credits and the tool quietly slows down or stops. You’re back to counting your texts.
Here’s what the telecom companies learned the hard way. Per-minute and per-text pricing was never really a pricing model. It was a way to ration something scarce. Spectrum was limited, switching capacity was limited, so they charged by the unit and made a fortune doing it. Texting was the most profitable product per byte in human history. They held onto that meter for as long as they possibly could.
Then one carrier offered unlimited. Customers stampeded. Everyone else matched within a year and a half. Not because the carriers got generous, but because the meter was costing them customers faster than it was making them money.
Now, the obvious objection. AI isn’t telecom. When a carrier’s network got bigger, the cost of one more text dropped to basically zero. AI doesn’t work that way. Every response burns real compute and real electricity. A heavy user genuinely costs the provider far more than a light one. So you can’t just wave the telecom story around and declare that flat pricing wins because it won before. The economics underneath are flipped.
But that misses the actual lesson.
The reason flat pricing won wasn’t that texts became free. It was that the meter changed how people behaved. The moment you make someone do math before they act, they act less. Every text you didn’t send because you were saving your count was a text the carrier never made a cent on anyway. The meter wasn’t protecting revenue. It was suppressing the very behavior the whole business depended on.
That’s the part that should terrify anyone metering AI right now. The entire value of these tools comes from people reaching for them constantly, without thinking. Thirty times a day. Building the reflex. The second someone hesitates before running an agent because they’re watching their credit balance, you’ve killed the habit you were trying to build. Metered pricing trains people to use your product less. For AI, that’s not a leak. That’s the whole boat.
So no, I don’t think pure flat pricing is where this lands, because the compute really does cost something. What I think wins is pricing that feels flat even when it isn’t. Generous limits most people never reach. Pooled usage across a team. A meter that still exists in the back office for the rare abuser, while everyone else just uses the thing and never thinks about a bill.
That’s actually what telecom’s “unlimited” always was. It had fair-use clauses and throttling buried in the fine print. It was never truly infinite. But it felt infinite, and feeling is what drives behavior.
Metered in the basement. Flat in the showroom. The companies that figure that out will own the next decade, and the ones still making you reach for a calculator before you try a feature will watch their users drift to whoever stops making them count.
We’ve seen this movie before. We already know how it ends.