Usage Limit

A usage limit is a fixed upper bound on the use of an online service. The provider determines how much a customer may draw on the service within a certain period. This can be a number of requests per hour, an amount of text per month, or a number of images generated per day. Once the limit is reached, the service either refuses further requests or continues working more slowly. Usually a message then appears indicating when it will be available again. Anyone who needs more must wait or switch to a more expensive plan.

Why providers throttle at all

Behind every AI response lies real computational work in a data center. This work costs electricity and requires expensive specialized chips, of which only a limited number exist worldwide. A customer who asks one question after another therefore continuously incurs costs. Without a limit, a few heavy users could consume the capacity meant for everyone else.

Limits also protect against abuse. Otherwise, automated programs could fire off requests around the clock to scrape data or bring the service to its knees. An upper bound per account makes such attacks unattractive. It also ensures that a bug in someone else’s software doesn’t overload the entire system.

For companies building on top of an AI, limits are therefore a genuine business risk. Anyone developing an app needs to know how many requests they’re allowed to make per second. If the provider changes the limits, this can render their own app unusable. Such changes are a recurring topic in tech news.

How limits are counted and enforced

The simplest approach is for a service to count requests per time window. For example, 60 requests per minute might be allowed. Any further request is rejected until the window resets. Experts call this a rate limit — that is, a restriction on speed.

For language AIs, simply counting requests is often not enough. A short question is far cheaper than an uploaded book. That’s why counting is usually done in tokens — small chunks of text roughly four characters long. A limit might then read, for example: one million tokens per day. A single book chapter alone can already consume a noticeable portion of that.

A common misconception is that a limit shuts the service down hard. Often, instead, it’s merely throttled: requests still get through, but more slowly. Some providers also work with a credit-based model, where each request is deducted from a quota. This can be compared to a prepaid phone card. Once the credit runs out, you have to top up.

Usage limits in everyday life and subscription plans

Limits are most visible in the subscription tiers of the major chatbots. Free accounts are often allowed to use the most powerful model only a few times per day. After that, the service switches to a weaker variant. This very limit is the main argument providers use to advertise their paid versions.

The principle is also familiar outside the AI world. The data allowance in a mobile phone contract is a usage limit, as is the number of simultaneous streams on a video service. What’s new with AI, above all, is how strongly the cost per use can fluctuate.

In reports about AI companies, usage limits tend to come up in two situations. Either a limit is raised because new data centers come online. Or it is lowered on short notice because a service is being overwhelmed. Both are good indicators of just how tight computing capacity actually is at that moment.