Datadog now bills its Bits AI features in monthly credits, a use-it-or-lose-it line that resets to zero each month

Datadog is known for billing a stack of separate meters: infrastructure per host, APM per host again, logs billed twice. Its AI features add one more, and it does not work like the others. Datadog meters its Bits AI products in AI Credits, a monthly allowance that resets to zero on the first of each month. If you have committed to a bundle, whatever you did not use is gone, which is a different shape of bill from the consumption meters most Datadog planning is built around.

What an AI Credit is

In Datadog's own words, one AI Credit represents a unit of intelligent work performed by a Datadog AI product. So instead of pricing these features per host or per GB, Datadog prices the work the AI does. Four Bits AI products draw down credits:

• Bits Chat: natural-language search, analysis, and dashboard or notebook creation.
• Bits Investigation: automated alert investigation and root-cause identification.
• Bits Code: AI-assisted code generation, review, and debugging.
• Bits Agent Builder: custom AI agents for workflow automation.

How the credits are bought

There are three ways to pay for them, and the difference between them is the whole story of this line:

• On-Demand: no commitment, billed monthly on your actual usage. You pay for the work you did and nothing more.
• Monthly Commit: month-to-month, purchased in 500-credit monthly bundles.
• Annual Commit: a 12-month commitment, also in 500-credit monthly bundles.

The committed options are cheaper per credit, the pattern Datadog uses across its products, but they carry the catch below. If your usage in a month runs past your committed bundle, the extra credits are charged automatically at the On-Demand rate, so a busy month does not stop working, it just spills onto the more expensive rate.

The catch: committed credits do not roll over

Here is the line that makes this meter behave unlike the rest of your Datadog bill. Datadog states that AI Credits reset on the first day of each calendar month, and unused Commit credits do not roll over to the next month. So a committed bundle is use-it-or-lose-it: buy a 500-credit monthly bundle, use 200, and the other 300 expire on the first. Your host and log meters bill on what you consume, so under-use just lowers the bill; a committed credit bundle bills whether you use it or not.

That flips the usual forecasting advice. On hosts, over-committing slightly is often fine because the annual rate is cheaper. On AI Credits, over-committing is pure waste every month it happens, while under-committing spills to On-Demand. The honest default for a new or spiky workload is to start On-Demand, watch a few months of real usage, and only commit once you can see a stable floor of credits you burn every single month.

Where it sits on the bill

AI Credits are a new dimension on top of the meters most cost models already track, and they scale with how much AI-assisted work your team runs, not with your host count or log volume. That makes them easy to miss when you forecast: they do not show up in a host-based estimate at all, and a team that leans on Bits Chat and Bits Investigation daily can run a real monthly line that a spreadsheet built last year has no row for.

Our free Datadog cost estimator adds up the infrastructure, APM and log meters that make up the bulk of a typical bill, so you can see which of those runs away first. It does not price the Bits AI credit line, because that depends on how your team uses the AI features, so if you are turning Bits AI on, treat the credits as a separate line and start On-Demand until you can measure it. Our piece on why your Datadog bill keeps climbing walks the log line that usually runs away first, and the Datadog vs New Relic vs Grafana Cloud comparison sets the platform cost next to two alternatives.