The Flat Rate Era of AI Is Over. The Meter Is On.

Share

For the past couple of years, AI has been sold the way almost every new technology is sold at first: as a flat, unlimited promise. Pay a subscription, get a well with no bottom. The pitch was seductive precisely because it insisted there was no meter anywhere, that intelligence had somehow escaped the ordinary economics of usage. I think that pitch always had a short half-life, and this week the market stopped pretending.

Consider what happened in the span of a few days. OpenAI announced that starting in December its Plus, Pro and Business plans will share one token measured allowance across ChatGPT and Codex, which means talking to the assistant and using it to write code will now draw from the same finite pool. Then on November 9, five companies set or changed an AI price on the same day, and every single one of them billed by usage: Meta and Anthropic by token, Google by call, ClickUp by credit, Apple by a prepaid pack. Taken together it reads less like five separate decisions and more like the industry quietly agreeing on a rule that everyone already knew to be true.

This is the normal history of technology, not a sign that AI has failed. Virtually every unlimited offer in business is a subsidy, a price held artificially low in order to build a habit, not a reflection of the real cost of delivery. The internet was sold that way. Mobile data plans were sold that way. Rented software was sold that way before the meter eventually came on. Unlimited is a marketing posture. Metered is the thing that actually has to be true for the numbers to work out. AI was never going to be the exception, no matter how much the early advertising suggested otherwise.

There is a real irony here, and I notice it every time I read the announcements. The founding promise of consumer AI was liberation from exactly this kind of counting. The whole aesthetic was frictionless and careless, an infinite tap you could leave running. And yet the moment the products reached scale, the counting came back with a vengeance, all the way down to the token. We have built the most capable technology ever put inside an ordinary product, and the unit it ends up trading in is embarrassingly granular. You can be holding something that feels like magic and still have the distinct sense of watching a taxi meter on a long ride.

For someone running a business, the first consequence is plain. The cost of AI becomes a budget line you have to actively manage instead of a fixed number you set once and forget. A flat subscription was predictable by construction. It promised no surprises, which is exactly what a sensible finance person wants, and there is real value in that. Usage pricing brings back the oldest problem in software buying, the surprise on the bill. And the OpenAI change sharpens the point considerably, because chat and coding now compete for the same pool. Every question you ask the assistant this morning is a question you cannot spend this afternoon on the thing that actually ships.

It is also worth noticing where the balance of power sits in a metered world. The person who sets the meter is the toll keeper on a river everyone has to cross, which is a comfortable place to be. That is a concentration risk, and I would not pretend it away. The more AI moves to usage pricing, the more the big providers control the single most visible number in the customer relationship, the monthly bill. The meter gives them a lever on cost that a flat rate never could. In that sense metering is not just an accounting decision, it is a strategic one that quietly transfers a little more control to the platform.

But I do not think metering is a purely grim development, and it is worth saying that plainly rather than defaulting to skepticism. Usage pricing has an honesty to it that the flat rate never had. It ties what you pay to what you use, and in a healthy market that kind of alignment tends to pull prices toward the real cost of delivery rather than away from it. A metered model rewards products that deliver clear value per unit and punishes the ones that were coasting on an assumption of abundance. In a strange way the meter is the sign that the category is growing up, an admission that AI has real costs and real winners, that it is an industry and not a party.

If that is right, then the metering wave quietly redirects where the durable advantage sits. When everything is measured, the business that holds its customers is the one that makes the meter run less painfully for them: the product that delivers predictable total bills, that turns a raw token cost into a legible outcome, that stops the customer from having to think about units at all. It is the difference between selling someone a gallon and selling them the confidence that the tank will not run dry at the wrong moment. The companies that make metered intelligence feel abundant, even while it is not, will be the ones that keep the relationship. That has always been where the value lives anyway, in the reassurance and the workflow, not in the raw fuel.

There is an honest counterargument, and I want to give it room rather than wave it off. Not every agreement is going to lapse into raw metering anytime soon. The premium tiers still sell latitude, and most casual users will never brush against a limit. The pressure is heaviest where usage is heaviest, on the power users and the companies that run real workloads day in and day out. For that group the reaction worth watching is whether the meter becomes a reason to own the switch. Once the marginal cost of every call is visible and growing, the incentive to run the predictable, repeatable parts of your work on something you control becomes sharper, and the value of a workflow that keeps your own run cost low rises. The meter does not just collect money. It teaches people, for the first time, to care about unit economics.

The deeper point is that the flat rate was never a technology decision. It was a customer acquisition decision. It built the habit, got the assistants into every phone and dashboard, and made unlimited feel like the natural order of things. Once the habit is established, the subsidy can be politely withdrawn, and that is what the past few weeks have been. This is why the same conclusion keeps showing up from different angles. The intelligence itself is getting cheaper and more ordinary by the day, and the value is moving to the layers that were never going to be free, the context, the workflow, the relationship, the trust. The meter is just the most visible sign that the party is over and the economic reckoning has begun.

Let me end on a measured note, because there is no clean villain here. The companies putting the meter on are not behaving cynically. They built enormous infrastructure and it genuinely costs something real to run, and metering is the honest way to make that cost visible to the people who create it. A product that is priced by use can also be priced cheaply at the low end, which widens access even as it charges the heaviest users more. AI was always going to end up a metered utility once it stopped being a miracle. The only surprise is how long we all got to pretend otherwise. That is the quiet fact worth sitting with. Flat was the honeymoon. Usage is the marriage.