Who owns the Datadog bill?

Not who pays it, obviously, finance pays it, out of a cost center with a boring name. I mean who is the single person whose actual job is to make that number smaller next quarter than it was this one. Ask that in a room of ten engineering leaders and watch what happens: platform points at finance, finance points at engineering, the app teams generating most of the telemetry have genuinely never once seen what their own logs cost, and the invoice itself shows up as one lump-sum line that reads “Observability, $47k/mo” and gets approved on the nod, because the only obvious alternative is turning off monitoring, which nobody is going to say out loud in a planning meeting.

There’s the whole thing, in one line: the bill has no owner, so it grows.

Let me put real numbers on “grows,” because vague dread has never moved a budget. Observability now runs somewhere between 7 and 12 percent of total cloud infrastructure spend depending on whose 2026 survey you trust, and that’s the median, not the horror stories. Gartner says 36 percent of its enterprise clients are past a million dollars a year just to watch their own systems, and 4 percent have crossed ten million. Enterprise log volume is climbing more than 250 percent year over year. And the figure I keep in my back pocket for CFO conversations is this one: roughly 70 percent of that spend goes to storing logs that never get queried, and I mean never, not occasionally, not rarely, but never once inside their retention window. You are paying top-dollar indexing rates to store data that no engineer will read and no monitor will ever check.

The commons, with a monthly invoice

You already understand why this happens if you’ve ever shared a fridge with roommates.

A backend engineer adds a custom metric tagged by user_id because it was genuinely useful during one incident back in March, and that single tag quietly turns one metric into forty thousand, and her manager never sees a line that says “Priya’s debugging tag: $900/mo.” Now multiply that by every engineer, every service, every “let me just add a bit more logging so next time we can actually see what happened,” piled up over eighteen months, and you land on a number that grew 30 to 50 percent a year while each individual decision that drove it felt completely reasonable in the moment. Because it genuinely was reasonable. And that’s the entire trap. There’s no villain here, no runaway loop, no single dumb config to point at and fix, just a hundred defensible choices dumped onto one bill nobody reads line by line, because nobody can, because Datadog doesn’t hand it to you split by team unless you go out of your way to make it.

SpendHound’s data puts SMB Datadog spend up almost 13 percent year over year, the kind of quiet, creeping increase that never once shows up as a decision anybody actually made. Enterprise only went up 1.74 percent, which sounds a lot better right until you remember that enterprise was already enormous, and a small slice of an enormous number is still very real money.

Showback is boring and it works

The fix isn’t a tool. Please, for once in this category, don’t let anyone sell you a tool for this.

What works is that someone with actual authority decides the observability bill gets split and shown back to the teams that generate it, every month, in a place they’re forced to look. That’s it, that’s the entire intervention, and it is deeply boring. You tag telemetry by team or service, you build a dashboard that says team A spent $12k, team B spent $3k, team C spent $31k and can’t explain why, and you send it round on the same cadence as the on-call rotation. The FinOps Foundation has a whole framework for this now, treating observability as a real cost line with per-team splits and quarterly budget targets, and the framework is fine, read it if you like frameworks. But the part that actually changes behavior is much dumber than any framework: an engineer sees their own name next to a dollar amount, in a channel their skip-level can also see, and the amount goes down, every single time. I have watched a team cut its custom-metric count by two thirds in a single week, not because anyone mandated a target, but because for the first time the cost was pinned on them and sitting somewhere their boss would notice.

Visibility does the work here. The budget number is almost just for show.

What the owner actually does

Give the number to one person. A platform lead, an SRE lead, someone senior enough to walk into another team’s planning meeting and ask, out loud, why their log volume tripled in April. Not a committee, though, because a committee owns nothing, which is the entire appeal of forming one.

Their job, concretely, is short. Split the bill by team and publish it somewhere the whole org has to look, on the same boring monthly cadence as everything else. Pull the top three cost drivers each month and go ask the humans behind them what the data is actually for, and when the answer is “we might need it someday,” move it to a cheaper tier or delete it. Set a per-team budget so there’s an actual line for a team to visibly cross before anyone has to have a conversation about it. And keep one eye on the ingest-versus-index ratio, because on most stacks the cheapest win of the whole year is sitting right there untouched, logs getting indexed at $1.70 per million events that would do their job perfectly well in a flex tier at five cents.

That last one is nearly free money and I still meet teams two years into a Datadog contract who have never once opened the setting.

None of this needs a migration, a new vendor, or a quarter of platform work. It needs one person to own a number and the nerve to ask other adults why they’re spending money. That, honestly, is the hard part, the asking itself. Engineers do not enjoy being told their debug logs cost more than a junior salary. Tell them anyway.

I wrote more about where observability money actually goes here.

— Youn