Datadog Signs Commitments Against a Usage Meter — Then a Top Customer's Cutback Forced It to De-Risk 2026 Guidance

Somewhere in a Datadog customer's account, a meter is running. Every log line ingested, every host monitored, every trace and metric and synthetic check clicks the counter forward, and at the end of the month a bill lands that nobody signed off on in advance. This is the native shape of cloud consumption: unpredictable, spiky, impossible to promise. Datadog's whole commercial achievement was to take that runaway meter and staple it to a contract — to get customers to promise, a year ahead, to consume a floor of usage they can't precisely forecast. The trick worked well enough that its contracted backlog reached $3.48 billion, growing 51% year-over-year while revenue grew 32%.⁶ But the meter never actually stopped. It just moved out of sight.

The official story is that Datadog turned cloud volatility into annual contracts. That framing implies the volatility got neutralized. It didn't. The volatility got relocated — off the monthly invoice and onto the guidance call, where a single customer's belt-tightening can now blow a hole in the full-year number the whole company was managed against.

Three ways to buy the same meter: Datadog didn't invent this structure to look big — it shipped with it before the IPO

Start with what the customer actually signs. Datadog's 10-K describes three ways to buy: a committed contractual amount of usage recognized ratably over the term, a committed amount delivered as it's consumed, or a plain month-to-month usage subscription — and in every case, usage above the committed amount is charged as incremental overage.¹ Read that carefully and you see the elegance. The committed floor gives finance a ratable, predictable line to recognize. The overage on top captures the upside when a customer's cloud footprint balloons. The customer gets a discount for committing; Datadog gets a revenue floor plus an uncapped ceiling. It is usage-based pricing wearing the costume of a subscription.

The popular telling casts this as a mid-life pivot — a clever monetization layer bolted on once Datadog got big enough to demand commitments. That's wrong, and the receipt is in the SEC filings. The identical three-way structure — ratable committed usage, delivered-as-used committed usage, or monthly usage-based billing, with overage on top — already appears verbatim in Datadog's 2019 draft registration statement, filed before the company went public.² This wasn't a growth hack discovered at scale. It was the architecture the company was born selling. The commitment-against-a-meter model is the product, not an afterthought.

Watch the backlog climb faster than the revenue: the commitments compound into a number that looks like accelerating demand

When customers commit ahead, the commitments pile up into Remaining Performance Obligations — the contracted revenue Datadog has booked but not yet delivered. And RPO is where the model's power shows up as a chart. At the end of the first quarter of 2020, RPO stood at $256.1 million.³ By the third quarter of 2024, it was $1.82 billion, up 26% year-over-year.⁴ A year later it reached $2.79 billion, up 53%.⁵ And in the most recent reported quarter it hit $3.48 billion, up 51% — outrunning the 32% revenue growth beneath it.⁶ A backlog growing faster than revenue reads, on its face, like demand accelerating ahead of the company's ability to bill it. That's the bull case, and it's not nothing.

But there's a mechanical footnote the bull case skips. Datadog's own management attributes part of the RPO acceleration to customers signing more multi-year deals — and longer contracts inflate RPO by construction.⁶ Sign the same annual run-rate for three years instead of one, and the backlog roughly triples without a single new dollar of demand. The backlog number is real; it just isn't only demand. It's partly the arithmetic of duration. Read it alongside the contract-length commentary or it will lie to you in the flattering direction.

The meter never stopped — it just moved to the guidance call: a commitment sets a floor and a recognition schedule, not a shield

Here is the part the 'volatility became contracts' story cannot absorb. A committed contract does two things: it sets a revenue floor and it fixes a recognition schedule. It does not stop a customer from consuming less than they committed, or from declining to renew at the same level, or from optimizing their cloud bill the moment their CFO squeezes. Datadog knows this and says so. In its Q3 2024 call, with $1.82 billion of RPO on the books, management warned that large customers who had ramped usage fast 'may optimize cloud and observability usage,' and that this 'may create volatility in our revenue growth in future quarters' — even though those customers held commitments.⁴ The commitment did not remove the risk. It changed where the risk was denominated.

Then the warning came true, in the most concentrated way possible. On the Q2 FY2026 call, CFO David Obstler confirmed that enterprise customers typically hold 'commitment and usage' contracts with 'annual plus' terms — and then disclosed that a reduction in usage from the company's single largest customer required management to 'fully derisk' that customer's contribution to full-year guidance.⁸ Sit with that. One customer trimmed usage, and a company running a $3.48 billion contracted backlog had to strip an entire account out of the number it steers by. That is not what a volatility shield does. That is what a revenue floor does when the customer chooses the floor — and a floor, by definition, is where you land after you've stopped rising.

But doesn't 120%-plus retention prove the model just wins?: the expansion engine is real, but it's cooling, and 'true recurring' was always the honest debate

The fair objection is that all of this is a rounding error against an expansion machine that has printed money for years. A former VP of Finance recalled that net dollar retention was disclosed 'north of 130%' every quarter through the mid-2010s and early public-company era — customers reliably spending far more this year than last, the closest thing SaaS has to a perpetual-motion machine.⁷ If commitments plus overage make existing customers grow 30% a year on their own, who cares whether a floor is technically a shield? The upside swamps the risk. That objection is strong, and for most of Datadog's life it was decisive.

Two things blunt it now. First, the same former finance executive flagged the debate that never went away internally: whether usage revenue drawn down against a commitment counts as 'true recurring revenue' at all — because a customer who commits and then under-consumes has recurring in name and volatile in fact.⁷ Second, the engine has slowed. By Q3 2025, trailing-twelve-month net revenue retention had cooled to 120%.⁵ Still excellent — but the '130%+' figure is history, not a present-tense hallmark, and a decelerating expansion rate is exactly the condition under which a large customer's optimization stops being absorbed by everyone else's growth. When retention was 130-plus, one account going quiet was noise. As the base matures and the rate cools, the same event becomes a guidance headline.

Datadog built something genuinely clever: a way to sell the least predictable thing in software — metered consumption you can't forecast — as a contract you can recognize ratably and stack into a multi-billion-dollar backlog. What it did not build, and never claimed in its own filings to have built, was an escape from the meter. The commitment is a floor drawn under a number that still moves. And the lesson of the fully-de-risked largest customer is the quiet one every usage-based business eventually learns: you can schedule when revenue is recognized, but you cannot schedule when a customer decides they've bought enough. The volatility didn't vanish. It grew up, put on a suit, and started showing up on the earnings call.