All-you-can-eat model

The term comes from the world of dining. At a buffet, you pay a fixed price and may help yourself as many times as you like. This is exactly the principle the all-you-can-eat model describes in business: the customer pays a fixed amount, usually per month, and may then use an offering without limit. The bill stays the same whether they use it once or a hundred times. In the tech industry, such offerings are also called flat rate or fixed-rate plans. The counterpart is pay-per-use billing, where every single use costs money.

Who wins and who loses at the buffet

For customers, the big advantage is predictability. At the start of the month, you know exactly what will be on the bill at the end of the month. Those who use it a lot often come out cheaper than with per-unit billing. And because there’s no need to think about costs, people tend to use such offerings far more intensively.

For the provider, it’s a bet. They calculate based on an average: the many light users are meant to subsidize the few heavy users. As long as the mix is right, the business works. If it tips because an above-average number of extreme users show up, the plan turns into a loss-maker.

This is exactly a real problem with AI services. Every request to a language program like ChatGPT costs the provider computing time in large data centers, and that’s expensive. In 2025, OpenAI CEO Sam Altman publicly stated that his company loses money on its most expensive subscription because customers use it more than expected. With a streaming service, the hundredth movie played costs almost nothing extra. With AI, the hundredth request costs just as much as the first.

Blended pricing and its hidden limits

Mathematically, the model is a blended calculation. The provider estimates how much an average customer will use and sets the price above that. At 20 euros a month and estimated costs of 8 euros per customer, there’s room to spare. Individual customers may then even rack up 100 euros in costs, as long as there are enough frugal users to balance it out.

To make the math work, providers almost always build in brakes. Common examples are caps per hour or per day, so-called rate limits. Or speed drops once someone uses the service very heavily. Sometimes the unlimited aspect only applies to the basic features, while the best tools are billed separately.

That’s why “unlimited” in advertising is rarely meant literally. The terms of service then include a clause against abusive use. Another typical misconception is equating all-you-can-eat with a subscription. Every all-you-can-eat model is a subscription, but many subscriptions have fixed quotas and are therefore precisely not a flat rate.

From the mobile phone flat rate to the AI subscription

In everyday life, this model is everywhere. The mobile plan with unlimited texts, the streaming subscription with all the movies, the monthly pass for bus and train, the gym membership: all the same idea. Germany’s Deutschlandticket is also a flat rate, just one that is politically intended and subsidized with tax money.

In tech news, you currently encounter the term mainly in connection with AI subscriptions. Providers like OpenAI, Anthropic, or Google sell monthly packages for 20 to 200 euros. Business customers, on the other hand, usually get billed per processed piece of text. That’s why analysts are looking closely at whether these flat-rate plans actually turn a profit.

As a user, it’s worth doing a simple calculation. Estimate how often you actually use a service and compare that to the per-use price. For occasional use, pay-per-use billing is cheaper. For daily use, the flat rate almost always wins.