Summary

  • Datadog grants 100 custom metrics for every Pro host and then meters custom metrics as an account-wide monthly average, which the same page states outright: not on a per-host basis.
  • Hosts and custom metrics leave one hourly reading and are settled by opposite rules, hosts on the maximum of the lower 99 percent of hours and metrics on the plain mean of all of them, so five hundred hosts that exist for one percent of a month cost nothing on the host line and still leave 500 billable custom metrics behind.

The price list prints the unit in seven words. Per 100 custom metrics, per month, with $5 beside it in each of three columns for annual, month-to-month and on-demand billing (price list). A separate row sells Per 100 ingested custom metrics, per month at $0.10. Neither unit names a host. The asterisk on the row sends existing customers to their latest billing agreement for included allotments, which qualifies the allowance and leaves the unit alone.

Now read the grant. The pricing page answers it in one sentence: Datadog offers 100 custom metrics for every host monitored with Infrastructure Pro, or 200 custom metrics for every host monitored with Infrastructure Enterprise. The feature matrix repeats it in a cell, 100 per host allotted. So the allowance is a multiplication, and its multiplier is the billable host count (pricing).

The next answer on the same page forbids that multiplication. Custom metrics are counted hourly and averaged monthly across your entire account, not on a per-host basis. Two sentences, one page, two different operations: one multiplies by hosts, the other says the meter never does.

The counted object tightens the knot. On the custom metrics billing page, A custom metric is uniquely identified by a combination of a metric name and tag values (including the host tag). The product documentation carries the same identification rule (custom metrics). So the host appears in the price twice: as the multiplier in the allowance, and as a term inside the identity of every counted series. Adding a host does not simply add 100 to an allowance. It also renames every series that host reports.

Then come the two aggregations, printed in adjacent sentences on one page. The billing overview opens with Datadog meters the count of hosts and custom metrics hourly. and then splits:

The billable count of hosts is calculated at the end of the month using the maximum count (high-water mark) of the lower 99 percent of usage for those hours. Datadog excludes the top 1 percent to reduce the impact of spikes in usage on your bill. The billable count of custom metrics is based on the average number of custom metric hours for the month.

(billing overview)

Same meter, opposite settlements. The host rule discards the top one percent of hours and takes the maximum of what remains. The metric rule discards nothing and takes the mean of everything. One figure is a trimmed peak, the other an unweighted average. The custom metrics billing page prints the averaging method in full: the total of all distinct custom metrics for each hour, summed over the month, divided by the number of hours in the month (custom metrics billing).

Set the two side by side and the allowance stops being a simple product. Its size is fixed by a trimmed peak times a constant. Its consumption is a mean of every hour measured against that size. Both inputs are hourly, and that is all they share.

Here is what the two published figures say to each other, in my arithmetic and not Datadog's. One custom metric costs $5 divided by 100, five cents a month. A Pro host's included 100 are therefore worth $5.00 at list against the $15 per host per month displayed for Infrastructure Pro on annual billing, one third of the host price. An Enterprise host's included 200 are worth $10.00 against $23, or 43.5 percent. And the included hundred carries exactly the same $5 as one overage block of a hundred, so the allowance is not a discount. It is the same quantity at the same rate, folded into the host line.

Now the shape those two sentences were written for. Take a 730-hour month, ten Pro hosts carrying 100 distinct series each for the whole month, and a burst of 500 hosts, each reporting the same 100 series against its own host tag, present for exactly 7.3 hours, which is one percent of the month. My figures, on those stated assumptions: the estate logs 10,950 host-hours and a mean host count of 15.0, but the billable host count is 10, because 99 percent of the hours sit at ten hosts. The series-hours are 1,095,000, so the billable custom metric count is 1,500. The allowance is 10 hosts times 100, or 1,000.

The overage is 500 metrics, five blocks, $25.00 at list.

Read what the burst did. Five hundred hosts for seven hours is five host-months of infrastructure. It contributed 500 billable metrics, exactly five blocks, $25.00 of metric usage, and no billable hosts at all, because the trimming rule was built to absorb precisely this shape of event. Without the burst the same month would show 1,000 billable metrics against a 1,000 allowance, nothing owed on metrics and $150 on the host line. Had those 500 hosts existed all month, the host line alone would be $7,500, with the metric allowance of 50,000 meeting usage of 50,000 exactly.

So the same estate costs $7,500 flat and $175 as a one-percent visit: one percent of the duration carries 2.33 percent of the money, and the $25 the burst adds lands on the line the host meter has just declined to see.

This is not a rounding difference. A series present for one hour of 730 contributes one seven-hundred-and-thirtieth of a billable metric. It is amortised, not forgiven. Hosts get forgiveness and metrics get amortisation, and the two halves of one product are settled by rules that disagree about what a spike is.

Pooling sharpens it. The custom metrics billing page works the allowance through: a three-host Pro account is allocated 300 indexed custom metrics that can be concentrated on a single host. So the allowance is an account-wide pool whose size a host count sets. A metric-heavy fleet with few hosts reaches the ceiling first; a host-heavy fleet with modest series counts accrues allowance it cannot spend. That second reading is mine, not the page's.

One more thing sits under the same word. A second pricing model is printed for custom metrics. Metric Name pricing bills custom metrics based on the number of unique metric names you submit and the volume of datapoints those names produce. It replaces the cardinality-based Custom Metrics billing model for organizations that opt in. It bills each unique name with more than 100 indexed datapoints in a month, includes 10 million indexed datapoints per billed name, and allows ingested volume up to five times indexed volume. Its rates and tier boundaries are contract-defined, month-to-month 20 percent above an annual baseline and on-demand 40 percent above it. The exclusivity is stated too: These SKUs are mutually exclusive with the Timeseries (cardinality) pricing SKUs. (metric name pricing)

The optimisation lever points the other way under the two regimes, and that is printed. Under cardinality billing, only configured metrics contribute to your Ingested custom metrics volume. If a metric is not configured with Metrics without Limits™, you're only charged for its indexed custom metrics volume. (Metrics without Limits) Under the name-based model, ingested volume is part of the billed base for every name, with an allowance expressed as a multiple of indexed volume. A buyer who trims configuration to save on one model is preparing a different bill on the other. The custom metrics billing page does open by naming which model it describes and sending name-priced contracts elsewhere, so a precedence sentence exists between the models. What none of the seven documents read here contains is a sentence relating the two aggregations inside one model.

The reading limits belong beside those numbers. The price list and the pricing page's questions were read from search-retrieved excerpts of those exact addresses rather than full rendered pages, and the documentation pages from the markdown mirror of the same documents. The Usage Overview address, which the custom metrics billing page names as the place billable usage is reflected, returned HTTP 404 on 21 November 2026. The indexed overage cell could not be captured verbatim, so how the contract-defined indexed rate relates to the $5 list row is unresolved rather than inferred. No effective date for the list prices was visible.

So what can a buyer do with this? Both lines are recomputable on their own terms: from hourly host counts you can rebuild the trimmed peak, and from hourly distinct series counts you can rebuild the mean. What no buyer can do from these pages is check the multiplication the unit implies, because the unit's two printed factors are not the two quantities that reach the invoice. The metric charge is not hosts times metrics. It is a mean measured against a trimmed peak multiplied by a hundred, and the two are free to move apart every time the estate changes shape.

That is the whole mechanism. The identity this piece is bound to is the directory entry for Datadog, Inc.. The grant is a peak. The bill is a mean. A short, wide estate is nearly free on one line and not free on the other, and the one surface that would let a buyer see both at once, a worked example printing hosts and custom metrics as numbers together, appears in none of the seven documents.