BLOG · 2026-05-19
My AI bill spiked this month
You build something with AI, it works, you let it run. At the end of the month the bill arrives and you get a shock: double, triple what you expected. The AI didn't warn you because, as far as it's concerned, it did its job. The problem is that most AI services are paid by usage, and usage can spike in silence.
How AI is paid for (and why it surprises)
Unlike a fixed subscription, most AI services charge by amount of work: every question and every answer costs a little, depending on length. A conversation costs cents. The problem is that "a little" multiplied by many times is a fat bill. And nobody adds up the cents until the total arrives.
It's like water: each glass is insignificant, but a tap left dripping for a month shows up on the bill.
The usual causes of the shock
- More usage than expected. Your chatbot was a hit and far more people used it than planned. Good news for the business, bad for the bill if you didn't anticipate it.
- Conversations that drag everything along. Some ways of setting up AI make each message resend the entire previous conversation. The longer it is, the more expensive each turn. It grows like a snowball.
- A runaway loop. An automation that calls the AI in a cycle and, through a bug, doesn't stop. The bill climbs by itself all night.
- Someone abusing it. If your AI is open to the public with no limits, someone can use it heavily — for their own purposes, at your expense.
How to avoid the shock
- Set a ceiling. Almost all services let you set a maximum monthly spend. Set it. It's the most basic safety net and almost nobody turns it on.
- Alerts on the way up. Get alerted at half or 80% of what's planned, not once you've already overshot.
- A per-user limit. One person (or one abuser) shouldn't be able to consume without brakes. A cap per user, per hour.
- Use the right-sized AI. You don't need the most expensive, most powerful model for simple tasks. Using a lighter one where it suffices saves a lot.
- Watch the meter. Keep how much you've spent this month in view, instead of discovering it on the bill.
The difference between a shock and control
AI by usage isn't expensive — it's unpredictable if you don't watch it. The same tool can cost you little or spike, and the difference isn't in the tool, but in whether you set limits and alerts. It's the difference between a tap with a meter and an alert, and one you only find open when the bill arrives.
That's why, when we build AI automations, the caps, alerts and right-sized model come as standard, not as an extra. In our own product we use AI sensibly: the right task on the right model, with clear limits. Let the AI work for you, not work through your budget.