Your AI Tool Has No Speed Limit. Set the Cap Before You Turn It On.

In plain terms: A lot of AI tools have quietly switched from a flat monthly price to billing by usage, so the more the AI works, the more you pay, with no ceiling unless you set one. The cautionary tale making the rounds: an AI consultant told Axios that one company ran up a roughly $500 million bill in a single month after never setting a usage limit. That exact number is unverified, but the pattern behind it is very real, and the fix takes about a minute.

For years, software pricing was simple. You paid a flat fee per seat per month, and you knew your bill before it arrived. A lot of AI tools have quietly stopped working that way.

The bill nobody saw coming

The story everyone is passing around: an AI consultant told Axios in May that one of their clients ran up a roughly $500 million bill in a single month, after forgetting to set usage limits on its employees' AI licenses. The company was never named and the figure has not been confirmed, so treat it as a story, not a fact. It spread because everyone in the industry found it completely plausible.

The confirmed cases are nearly as striking. Uber burned through its entire 2026 budget for AI coding tools by April, four months into the year, and had to put hard caps in place. At another company, a single engineer ran up $40,000 in usage in one month. Their CEO said he honestly did not know whether to stop the guy or tell everyone else to copy him.

Why the meter has no ceiling

Here is the part that catches people off guard. The price of AI keeps falling. The cost per unit of work has dropped something like 98 percent. So owners assume their bill should be shrinking too.

It does the opposite. The tools got cheaper per task, but the newer agent-style tools do far more tasks, and they run as fast as you let them. A person has a natural ceiling: they go home at five. An AI agent does not. A falling price per task, times a soaring number of tasks, adds up to a bill that climbs. One industry group reported companies running three times over their whole year's AI budget by April.

This is not just a big-company problem

It is easy to read those numbers and decide this is an enterprise issue. It is landing on small tools too. The ChatGPT add-in for Excel and Google Sheets just moved off its free preview and now bills by credits per task. Microsoft moved Copilot plans to usage-based billing this month. The flat fee you signed up for is increasingly a floor, not a ceiling.

The one move: set the cap before you turn it on

You do not need to avoid these tools. You need to put the speed limit on yourself, because the tool will not do it for you. Before you switch on anything that bills by usage, credits, or tokens:

• Find the spend limit and set it on day one. Pick a monthly number you would be fine paying even if it doubled, and cap it there. A real cutoff, not just an email alert.
• Use the cheaper setting for routine work. Most tools let you pick a smaller, cheaper model for everyday tasks and save the expensive one for the hard jobs. Default to cheap.
• Watch the meter weekly for the first month. Usage is lumpy at the start while you learn what the tool actually does. Check every week until it settles, then you can relax.

The honest part

AI billed this way is genuinely worth it, often many times over. The danger is never the tool itself. It is turning it on, forgetting about it, and meeting the number at the end of the month. The whole risk is a setting you can handle in about a minute.

This is the kind of thing we set up carefully for the businesses we work with, so the tools earn their keep without any surprises on the statement. If you are about to turn on an AI tool that bills by usage and you are not sure where the cap lives, that is a quick thing to check, and a good conversation to have.