What is Datadog gross margin trajectory through 2028?

The outcome you should expect

The honest answer to "what is Datadog gross margin trajectory through 2028" is *boring*, and that is the point. A company already operating at roughly 81% non-GAAP gross margin and roughly 78% GAAP has very little headroom left in the numerator. The remaining 19 points of cost of revenue are not fat waiting to be trimmed — they are cloud infrastructure, data storage, bandwidth, customer support delivery, amortization of acquired intangibles, and stock-based compensation allocated into COGS. Each of those has its own inflation curve, and none of them goes to zero. So the realistic expectation for the trajectory through 2028 is a band, not a line: low-80s non-GAAP, high-70s GAAP, with quarter-to-quarter wobble of a point or so in either direction depending on contract renewal timing and product launch cadence.

That framing matters more than the specific number, because the way most people model this question is wrong. They take the current margin, apply a growth-improves-margin assumption borrowed from early-stage SaaS, and extrapolate 84–86% by 2028. That assumption is valid when a company is climbing from 60% to 75% — the fixed portion of COGS amortizes across a growing revenue base and the margin genuinely expands. Datadog finished that climb years ago. The fixed-cost amortization story is largely exhausted; what remains is a variable cost structure that scales roughly linearly with data ingested, retained, queried, and processed. Doubling revenue roughly doubles telemetry volume, which roughly doubles the underlying infrastructure bill absent efficiency gains. The efficiency gains are real, but they are fighting volume growth, not riding on top of it.

There is a second reason to expect flatness rather than expansion: the company's own stated priority is not gross margin maximization. Management commentary across recent reporting periods has consistently framed the trade as investing in new product surface area — AI observability, cloud cost management, security products — at the expense of near-term margin optimization. That is almost certainly the right call for enterprise value, because a point of gross margin is worth far less than a durable new revenue line. But it means anyone modeling gross margin expansion through 2028 is modeling against the company's revealed preference. If Datadog wanted 85% gross margins, it could get closer by cutting free-tier generosity, throttling data retention defaults, and shelving compute-intensive AI features. It is doing the opposite.

For a RevOps or finance operator consuming this analysis, the practical translation is: model gross margin as a constant in your Datadog forecast, then spend your modeling energy on operating margin, where the actual leverage lives. Sales and marketing as a percentage of revenue, R&D as a percentage of revenue, and G&A leverage are where the 2028 story gets interesting. Gross margin is the stable input; operating margin is the variable output. Treating them the other way around produces forecasts that miss badly in both directions.

The one scenario that would break flatness in a meaningful way is a structural change in the cost of AI inference. If the per-token cost of running the model-backed features collapses — which is plausible given the trajectory of inference pricing — then the drag from AI product mix largely disappears and the offsets win outright. If inference costs stay sticky while AI feature adoption accelerates, the drag compounds. That single variable has more explanatory power over the 2028 number than everything else combined, which is why any forecast that does not state an inference-cost assumption is incomplete.

What drives that outcome

Cost of revenue for an observability platform decomposes into a small number of large buckets, and understanding their relative weight is what makes the trajectory forecast tractable rather than hand-wavy.

Cloud infrastructure is the dominant line. Datadog runs predominantly on public cloud infrastructure, with AWS as the primary provider. Compute for ingestion pipelines and query execution, object storage for retained logs and traces, and data transfer between availability zones and out to customers together constitute the large majority of COGS. The pricing on these is negotiated through committed-spend agreements, which trade volume commitments for discounted rates. This structure creates a specific and predictable pattern: margin improves during the middle of a commitment term as actual usage catches up to committed volume, then compresses at renewal when the new commitment is priced against current-generation rates and a larger base.

Data volume growth outpaces revenue growth in most enterprise accounts. This is the structural headwind that distinguishes observability from ordinary application software. Customers instrument more services, emit more custom metrics, retain traces longer, and generate more logs every year — often growing telemetry volume faster than they grow their contract value, because pricing tiers include volume allowances. Every unit of that excess volume is pure COGS with no matching revenue. Compression improvements, tiered storage, adaptive sampling, and index-on-demand architectures exist precisely to counteract this, and they work — but they are running to stay in place.

New product mix is dilutive at launch and accretive at maturity. A newly launched product carries proportionally more infrastructure cost per dollar of revenue than a mature one, because it has not yet been optimized, has not reached scale on its dedicated infrastructure, and is often priced promotionally to drive adoption. Products built on large language model inference are the extreme version of this: the marginal cost of serving a query is meaningfully higher than the marginal cost of returning a pre-computed dashboard. As those products mature — better caching, smaller distilled models for routine paths, larger models reserved for genuinely hard queries — their unit economics improve toward the platform average.

Support and customer success delivery scale sub-linearly. This is a genuine offset. Headcount-based support cost per customer declines as documentation improves, self-service diagnostics mature, and automated triage handles the routine tier of tickets. In a company adding thousands of customers per year, holding support cost roughly flat while the customer count grows is worth real basis points.

Stock-based compensation drives the GAAP-to-non-GAAP spread. The roughly three-point gap between the two figures is primarily SBC allocated to COGS-bearing employees plus amortization of intangibles from acquisitions. This gap tends to narrow slowly over time as revenue grows faster than equity grant expense, which means the GAAP margin trajectory through 2028 is slightly *better* than the non-GAAP trajectory even when the non-GAAP number is flat. Analysts who quote only the non-GAAP figure miss this convergence.

The interaction between these buckets is what produces a flat line rather than a trend. Volume growth and renewal pricing push down; compression, automation, and SBC convergence push up. Neither side dominates by more than a point or two in any given year, and the timing of the pushes is uncorrelated — a renewal-heavy year with light new-product launches looks different from the reverse. That decorrelation is exactly why the trajectory reads as a band with noise rather than a clean slope.

Benchmarks and realistic ranges

Context makes the number interpretable. Placing Datadog against comparable public software companies clarifies both how good the current margin is and how little room remains above it.

Pure application software sits highest. Companies whose cost of revenue is essentially support headcount plus modest hosting — CRM platforms, ITSM platforms, marketing automation — cluster in the low-to-high 80s. They have almost no data-gravity cost. A record in a database costs essentially nothing to store relative to its revenue contribution. This is the ceiling Datadog structurally cannot reach, because Datadog's product *is* data volume.

Consumption-priced data platforms sit lowest among quality names. Cloud data warehouses run materially below Datadog because storage and compute are the product, and the pricing model passes through a large fraction of infrastructure cost by design. The roughly ten-point gap between a consumption data warehouse and Datadog is the clearest illustration of how pricing model determines gross margin: per-host and per-seat pricing decouples revenue from infrastructure cost in a way consumption pricing deliberately does not.

Security and endpoint platforms sit in between. Agent-based security companies typically land in the mid-to-high 70s — they carry real data ingestion and retention cost, but less than a full observability platform, and their detection processing is more amenable to edge computation on the agent itself.

Datadog sits at the top of the data-heavy cohort. Roughly 81% non-GAAP is an unusually strong result for a company ingesting the volume it does, and it reflects genuine architectural advantage: efficient storage formats, aggressive compression, tiered retention, and a pricing model that charges by host and by feature rather than purely by byte. The relevant benchmark question through 2028 is not "can it reach 85%" but "can it defend 80% against volume growth and AI cost." The realistic answer is yes, with effort.

Realistic ranges by scenario. A base case holds non-GAAP in the 80–82% band with GAAP converging from high-70s toward the 79–80% area as SBC leverage improves. An upside case — where inference costs fall materially, compression gains land ahead of schedule, and a renewal cycle prices favorably — reaches 82–84% non-GAAP by the end of the period. A downside case — where cloud pricing inflates faster than efficiency gains, AI feature adoption scales ahead of unit-cost optimization, and competitive discounting deepens — settles in the 76–79% range. The probability mass sits heavily on the base case because the offsetting forces are structurally self-balancing: a year of heavy AI adoption is also a year of heavy optimization investment.

What to actually watch each quarter. Cost of revenue as a percentage of revenue is the single cleanest indicator, and it is directly readable from the income statement without any modeling. Sustained readings above 20% signal that the downward pressures are winning; sustained readings below 18% signal the offsets are. Second, watch the disclosed mix of revenue from newer products — rising new-product share correlates with near-term margin drag and longer-term margin recovery. Third, watch commentary about infrastructure commitments; renewal language in the risk factors and MD&A sections often telegraphs a margin step before it appears in the numbers.

Do not benchmark gross margin against private or open-source competitors. Open-core observability vendors operate at structurally lower gross margins because they monetize a fraction of usage while bearing hosting cost for their cloud offering. Comparing their margin to Datadog's is comparing different business models, not different levels of operational efficiency. The competitive threat from those vendors is real, but it shows up as pricing pressure in Datadog's revenue line and net retention, not as a gross margin benchmark.

Risks, edge cases, and failure modes

The base case is robust, but several specific mechanisms could break it, and each has a distinguishable signature in the reported numbers.

Cloud commitment renewal at unfavorable terms. The largest single-point risk. Committed spend agreements are negotiated in multi-year blocks; a renewal priced against a much larger base without proportional discount improvement compresses margin immediately and durably. The mitigation Datadog has available is credible multi-cloud capability — running meaningful workload on a second provider changes the negotiating dynamic even if the second provider never becomes primary. Watch for infrastructure diversification language as a leading indicator of negotiating posture. The failure mode is a company that publicly commits to single-cloud architecture and then negotiates from weakness.

AI feature adoption outrunning unit-cost optimization. If model-backed features become the primary interaction surface faster than the cost per interaction falls, margin compresses in a way that is hard to reverse without degrading the product. The specific edge case that hurts most: heavy usage by customers on flat-rate or bundled pricing, where incremental inference cost carries zero incremental revenue. The mitigation is architectural — route routine queries to small distilled models, reserve frontier-model calls for genuinely hard reasoning, cache aggressively at the query-shape level, and meter the truly expensive paths as a separately priced SKU rather than a bundled feature.

Telemetry volume growth outpacing compression gains. Compression and tiered storage have improved dramatically, but they face diminishing returns — you cannot compress the same data twice. If customer telemetry volume grows 35–40% annually while storage efficiency improves 15–20%, the delta lands in COGS. The edge case that makes this acute is a shift in customer instrumentation patterns toward high-cardinality metrics and full-fidelity tracing, both of which are dramatically more expensive per unit of business value than sampled traces and low-cardinality metrics. This is partly within the vendor's control through defaults and guardrails, and partly not.

Competitive discounting at the enterprise tier. Hyperscaler-native monitoring tools are bundled with cloud spend and price at effectively subsidized levels. They are less capable, but for a meaningful segment of workloads they are good enough. Every large renewal now happens with that alternative on the table. Deep discounting to defend those renewals compresses revenue without compressing cost, which mechanically compresses gross margin. The signature is net revenue retention softening alongside stable customer counts — revenue per customer falling while logo count holds.

Pricing model change at the low end. Flattening or simplifying pricing for smaller customers improves win rates and reduces sales friction, but typically trades away revenue per unit of infrastructure consumed. This is a deliberate margin sacrifice for growth, and it is usually the right trade — but it should be modeled explicitly rather than discovered in the results.

The failure mode of the analysis itself. The most common modeling error is treating gross margin as the headline metric for a company at this stage. A one-point gross margin move is worth a fraction of what a one-point operating margin move is worth, and operating margin has an order of magnitude more room to move through 2028. An analyst who spends their effort forecasting gross margin to the tenth of a point and hand-waves sales efficiency has optimized the wrong variable. Gross margin here is a stability check — confirmation that the unit economics are not eroding — not a source of forecast alpha.

Edge case worth naming: acquisitions. An acquisition of a company with materially different unit economics changes reported gross margin mechanically, independent of any operational change. Acquired intangible amortization flows into GAAP COGS and widens the GAAP-to-non-GAAP spread for the amortization period. A model that does not separate organic margin from acquisition effects will misread a deal as operational deterioration.

A practical rollout plan

If you are the operator responsible for producing and maintaining this forecast — whether in a RevOps function modeling a vendor, an investment team modeling a position, or a competitive intelligence function modeling a rival — here is the sequence that produces a defensible number rather than a plausible-sounding one.

Step one: pull the actual reported figures, not summaries. Go to the primary filings and the quarterly earnings materials. Record cost of revenue in dollars and as a percentage of revenue for the last twelve quarters. Record the GAAP and non-GAAP gross margin separately, and record the reconciling items — SBC in COGS and intangible amortization — as separate lines. Most published analyses collapse these, which destroys the ability to distinguish operational change from accounting change. This step takes an hour and eliminates most of the error in downstream work.

Step two: build the cost-of-revenue decomposition. You will not get an exact breakdown from public filings, but you can bound it. Infrastructure cost scales with data volume; support cost scales with customer count; SBC and amortization are disclosed. Estimate the infrastructure share as the residual, sanity-check it against disclosed commitment obligations in the contractual obligations table, and carry a range rather than a point estimate. Label every number as disclosed, derived, or estimated — this discipline is what makes the model auditable six months later when someone challenges it.

Step three: pick your inference-cost assumption explicitly and write it down. As established, this is the highest-leverage single variable. State whether you assume per-unit inference cost falls, holds, or rises through 2028, state the magnitude, and state why. Any forecast that leaves this implicit is not falsifiable and therefore not useful.

Step four: run the three scenarios, not one. Base, upside, downside — with the specific mechanism named for each, drawn from the risk section above. Assign rough probability weights. The output of this step is a distribution, and a distribution is what you should actually communicate to whoever consumes the forecast. A single point estimate for a five-year margin trajectory conveys false precision.

Step five: define your revision triggers before you need them. Write down, in advance, what observation would cause you to move off the base case. Cost of revenue crossing 20% for two consecutive quarters. Explicit management commentary on infrastructure cost pressure. A disclosed change in commitment structure. New-product revenue share crossing a threshold you specify. Pre-committing to triggers prevents the two failure modes of forecast maintenance: anchoring on your original number past the point of evidence, and revising on every quarter's noise.

Step six: review quarterly, revise annually. Check the triggers each quarter — that is a fifteen-minute exercise against the earnings release. Rebuild the model annually against the fresh filing. Anything more frequent is churn; anything less frequent means you are carrying a stale view into decisions.

A note on how this plugs into RevOps practice. If you are using this forecast to inform vendor negotiation rather than investment, the useful output is different. Knowing that the vendor operates at roughly 81% gross margin tells you there is meaningful room in a negotiation on price — but also that the vendor's cost of serving a high-volume, high-cardinality account is genuinely material, so volume-based concessions are harder to win than seat-based ones. Structure your ask around commitment length and predictable volume rather than raw discount percentage; that is the concession that actually costs the vendor least and is therefore most winnable.

Related questions

Why doesn't Datadog's gross margin expand as revenue grows?

Because most remaining cost of revenue is variable, not fixed. Infrastructure and storage scale with telemetry volume, which grows roughly in line with — sometimes faster than — revenue. The fixed-cost amortization that drives early-stage SaaS margin expansion was largely exhausted before the company reached its current scale.

How does the GAAP and non-GAAP gross margin gap behave through 2028?

The roughly three-point gap should narrow modestly. It consists mainly of stock-based compensation allocated to COGS plus acquired intangible amortization. Revenue growing faster than equity grant expense compresses the gap slowly, meaning GAAP margin trajectory is marginally better than non-GAAP even when non-GAAP is flat.

Which single variable most affects the 2028 number?

Per-unit inference cost for model-backed features. If it falls materially, the AI product mix drag disappears and the offsets win outright. If it stays sticky while adoption accelerates, the drag compounds. No other single input has comparable explanatory power over the endpoint.

Should I model gross margin or operating margin for Datadog?

Operating margin. Gross margin is near its structural ceiling with roughly one to two points of realistic movement in either direction. Operating margin has far more room, driven by sales efficiency, R&D leverage, and G&A scaling. That is where forecast effort produces actual differentiation.

What indicator signals the trajectory is breaking down?

Cost of revenue as a percentage of revenue sustained above 20% for two or more consecutive quarters, especially if paired with softening net revenue retention. That combination indicates cost pressure and pricing pressure arriving together, which is the specific pattern that pushes margin into the downside range.

FAQ

What is Datadog's gross margin right now?

Non-GAAP gross margin runs around 81%, with GAAP around 78%. The difference is driven by stock-based compensation allocated into cost of revenue plus amortization of intangibles from acquisitions. Both figures are disclosed in the quarterly earnings materials and the annual report, and you should pull them from the primary filing rather than a secondary summary — the reconciliation table is where the useful detail lives.

Is the trajectory through 2028 up, down, or flat?

Flat, in a band. The base case holds non-GAAP in the 80–82% range across the period with normal quarterly variance of about a point in either direction. Upward and downward forces are roughly balanced in magnitude, and their timing is uncorrelated, which produces noise around a level rather than a discernible slope.

What would push it above 82%?

A material fall in per-unit AI inference cost, compression and tiered-storage gains landing ahead of telemetry volume growth, a favorable infrastructure commitment renewal, and product mix shifting back toward mature high-margin SKUs. Several of those need to happen together — any single one is worth well under a point on its own.

What would push it below 79%?

Cloud pricing inflating faster than efficiency gains, AI feature adoption scaling ahead of unit-cost optimization, deep competitive discounting on enterprise renewals, and an unfavorable commitment renewal. As with the upside, this requires multiple factors compounding; isolated pressure gets absorbed within the band.

How does this compare to peers?

Datadog sits at the top of the data-heavy software cohort. Pure application software companies run higher because they carry almost no data-gravity cost. Consumption-priced data warehouses run roughly ten points lower because their pricing model passes infrastructure cost through by design. Agent-based security platforms land in between. Datadog's position reflects genuine architectural efficiency, not accounting choice.

Does gross margin matter much for valuation here?

Less than most people assume. At this level, gross margin functions as a stability check confirming unit economics are not eroding. Operating margin carries the valuation story through 2028, because that is where the leverage remains. A forecast that models gross margin to the tenth of a point while hand-waving sales efficiency has optimized the wrong variable.

Sources

• Datadog investor relations and SEC filings: https://investors.datadoghq.com/
• Datadog quarterly results and press releases: https://investors.datadoghq.com/news-releases
• SEC EDGAR full-text filing search: https://www.sec.gov/edgar/search/
• AWS EC2 pricing: https://aws.amazon.com/ec2/pricing/
• AWS S3 pricing: https://aws.amazon.com/s3/pricing/
• Snowflake investor relations: https://investors.snowflake.com/
• CrowdStrike investor relations: https://ir.crowdstrike.com/
• MongoDB investor relations: https://investors.mongodb.com/
• Bessemer Cloud Index: https://cloudindex.bvp.com/
• SaaS Capital benchmark research: https://www.saas-capital.com/research/

Related on PULSE

• [What is Snowflake gross margin trajectory through 2028?](/knowledge/q1594)
• [What is Salesforce gross margin trajectory through 2028?](/knowledge/q1545)
• [What is ServiceNow gross margin trajectory through 2028?](/knowledge/q1627)
• [What is Outreach gross margin trajectory through 2028?](/knowledge/q1747)
• [What is Datadog's data-center and multi-cloud strategy?](/knowledge/q1696)
• [Is Datadog's pricing model broken at the bottom of the market?](/knowledge/q1707)