What is ServiceNow gross margin trajectory through 2028?

The outcome you should expect through FY28

The honest answer to a gross margin question about a company like ServiceNow is that the number barely moves, and the reasons it barely moves are more interesting than the number itself. Non-GAAP subscription gross margin has sat in the low-to-mid 80s for years. That is the top of the enterprise software cohort, and it is not an accident — it is the arithmetic of a multi-tenant workflow platform where the marginal cost of the next seat on an existing instance is close to nothing, services revenue is deliberately kept small and run near breakeven, and pricing power is high enough that discounting rarely shows up as a cost-of-revenue problem.

What changes between now and FY28 is that the marginal cost of the next *unit of value* is no longer the next seat. It is the next inference call. That is a genuinely different cost structure, and it is the reason every SaaS CFO covering an AI-heavy product line has spent the last several earnings cycles being asked some version of this question. A seat costs you storage and a slice of compute. An agentic workflow that reads a case, retrieves from a knowledge base, drafts a resolution, and writes back to the record costs you tokens — repeatedly, per invocation, at a price set partly by a third-party model vendor.

So the expected trajectory is not a smooth line. It is a tug-of-war with a fairly narrow band of outcomes:

• Base case, ~82%. Inference cost growth stays roughly matched by data-layer and model-routing efficiency gains, AI-tier pricing attach grows steadily, and sovereign cloud expansion stays disciplined to already-announced markets. Gross margin drifts down 100–200bps over three years and nobody writes a headline about it.
• Bear case, ~80%. Inference volume outruns efficiency, the premium AI tier attaches more slowly than plan, and regulated-market infrastructure commitments expand faster than the revenue that justifies them. You still have the best gross margin in the peer group; you just gave back 300–400bps of it.
• Bull case, ~83%. In-house and smaller domain-tuned models displace enough frontier-model API traffic that per-workflow inference cost steps down rather than creeping up, AI-tier attach outperforms, and the compression essentially never materializes.

Notice the width of that band: three percentage points across three years. For a RevOps or FP&A team modeling this, the practical implication is that gross margin is *not* where the interesting variance in the model lives. Net revenue retention, AI-tier attach rate, and operating margin leverage all swing the valuation harder. Gross margin here is a signal about cost discipline and AI monetization design — it is a thermometer, not the engine.

One more framing point that gets lost. Gross margin compression driven by AI cost is only bad if the AI is not being monetized. If a vendor spends 150bps of gross margin to deliver agentic capability that lifts net revenue retention by five points and lengthens contract terms, that is one of the better trades available to a software company. The failure mode is not compression — it is compression *without* corresponding pricing capture, which is what happens when AI features ship into the base tier as a competitive checkbox rather than into a paid tier as a product.

What drives that outcome

Four forces push gross margin down and four push back. Understanding which lever a given quarter's movement came from is the whole analytical exercise.

The compressors.

*Inference passthrough.* Every AI-assisted workflow — summarization, case deflection, code generation for platform developers, agentic task execution — consumes model tokens that the vendor pays for. Some of that runs on frontier models from external providers at published per-token rates; some runs on smaller in-house or partner-tuned models at a fraction of the cost. The blended cost per workflow is the number that matters, and it is a function of routing policy more than of any single vendor's price list. This is the single most volatile line in the cost stack, because usage is elastic: give customers an agent that works and consumption climbs faster than the forecast.

*Regional and sovereign infrastructure duplication.* Regulated markets — public sector, defense-adjacent, financial services in jurisdictions with data-residency law, and several European and Middle Eastern national cloud frameworks — require infrastructure that does not enjoy full multi-tenant economics. You are running a smaller footprint in a dedicated region for a smaller customer base, sometimes with certification and audit overhead loaded into cost of revenue. Every incremental sovereign region is margin-dilutive on day one and margin-accretive only once the regional book of business scales.

*Data-layer and storage growth.* Modern platform strategies pull external data in — from data warehouses, ERP systems, HR systems — so AI agents can reason over it. That ingestion, indexing, and query volume is real cost, and it tends to scale ahead of the revenue attached to it because customers connect data before they fully deploy the workflows that monetize it. Lag of two to four quarters between cost onset and revenue realization is a normal pattern here.

*Public sector and enterprise discounting.* Large government and framework-agreement deals carry meaningful list discounts and higher delivery cost. Strategically excellent, gross-margin dilutive. The same is true of the very largest commercial renewals.

The protectors.

*Data engine efficiency.* Replacing a legacy relational layer with a purpose-built engine reduces compute per query. This is the largest cost lever a platform vendor actually owns outright — no vendor negotiation required, no customer behavior change required. When a CFO credits the data layer for offsetting AI cost on an earnings call, that is what they are describing.

*Model routing and smaller models.* Not every task needs a frontier model. Classification, extraction, routing, and template-filling can run on much smaller domain-tuned models at an order-of-magnitude lower cost per call. A mature routing policy sends the cheap 80% of traffic to cheap models and reserves expensive models for genuinely hard reasoning. Building that routing layer is the highest-ROI margin engineering available to any AI-heavy SaaS vendor right now.

*Premium AI tier pricing.* If AI capabilities sit behind a paid tier carrying a meaningful premium over the base tier, then inference cost and inference revenue rise together. The margin question then reduces to a contribution-margin question per AI workflow, which is a much more comfortable place to be. Attach rate of that tier is the number to watch.

*Scale and retention.* Very high gross renewal rates plus multi-year commitments give a vendor forward visibility to amortize infrastructure investment against contracted revenue. Growing total revenue also spreads fixed infrastructure and certification cost across a bigger base.

Benchmarks and realistic ranges across the cohort

Gross margin is only interpretable against a peer set, because the structural drivers differ enormously between business models that all get called "SaaS."

Multi-tenant workflow platforms — the category ServiceNow defines — run at the top of the range, low-to-mid 80s on a subscription basis. The economics: one codebase, one upgrade path, shared infrastructure, minimal per-customer customization at the infrastructure layer, and services deliberately kept to a small revenue share so they do not drag the blended number.

Broad application suites assembled partly through acquisition run meaningfully lower, often in the mid-to-high 70s. Acquired products arrive with their own infrastructure, their own hosting arrangements, and often their own lower-margin profile, and integration takes years. Salesforce is the canonical example of this pattern; the gap to a pure-play workflow platform is structural rather than a matter of execution.

HCM and financials platforms sit closer to the workflow platforms, around 80%, with similar multi-tenant economics but historically less premium pricing capture on AI add-ons.

Consumption-priced data platforms run lowest of the group, mid-70s, because cloud infrastructure cost is a near-direct passthrough of customer usage. Snowflake is the reference point. The tradeoff is that revenue expands automatically with usage — lower gross margin, but a growth engine that does not depend on seat expansion.

Observability and telemetry vendors sit around 80% and have faced quietly similar compression pressure to the AI cohort, for structurally identical reasons: data volume growth outpacing price-per-unit declines.

For a practitioner building a comparison model, the useful discipline is to normalize before comparing:

1. Subscription-only, not blended. Services revenue mix varies from under 3% to over 20% across this cohort and services typically run near or below breakeven. Blended gross margin mostly measures services mix, not platform efficiency.
2. Non-GAAP versus GAAP. The gap is usually 300–500bps, driven by stock-based compensation allocated to cost of revenue and amortization of acquired intangibles. Both numbers are legitimate; comparing one company's non-GAAP to another's GAAP is not.
3. Note the pricing model. Seat-based, consumption-based, and hybrid models produce structurally different margin profiles. A consumption vendor at 75% and a seat vendor at 83% may have identical operating discipline.
4. Check the international mix. Regulated-market and sovereign revenue carries higher cost of delivery. A company scaling into public sector will show gross margin drift that has nothing to do with product efficiency.

The realistic range statement, then: for a top-tier multi-tenant workflow platform through FY28, anything in the 80–84% band is normal execution. Sub-79% would indicate either a genuine AI cost problem or an acquisition that changed the mix. Above 84% would suggest AI adoption is running behind plan — which is not the good news it looks like on the margin line.

Risks, edge cases, and failure modes

Inference elasticity is the underestimated risk. Forecasts for AI cost tend to assume usage grows with seats. It does not. Usage grows with *product quality*. When an agent starts reliably resolving cases, ticket deflection workflows get invoked far more often, and per-customer inference cost can multiply within a couple of quarters while contract value stays fixed until renewal. The mismatch between usage-driven cost and annual-contract revenue is the single most common way an AI-era gross margin forecast goes wrong. Consumption caps, fair-use thresholds, or credit-based AI metering exist precisely to close this gap — watch for a vendor introducing them, because it signals the elasticity showed up.

Model vendor pricing is not under your control. Frontier model prices have trended down per unit of capability, but capability per task has also trended up, meaning teams route to more expensive models for better output. Net cost per workflow can rise even in a falling-price environment. A vendor overexposed to a single external model provider carries concentration risk on both price and availability.

Sovereign commitments are hard to reverse. Announcing a national cloud region is a multi-year infrastructure and certification commitment made against a revenue forecast. If that forecast disappoints — a procurement cycle slips, a political environment changes — the cost stays and the revenue does not arrive. This is a slow, quiet risk that shows up as unexplained international margin drift two years after the announcement.

Accounting reclassification can create phantom moves. If a company changes how it allocates internal infrastructure or capitalizes development, gross margin can move 50–100bps with no economic change. Always read the footnotes before modeling a trend break.

The FX edge case. Cost of revenue is incurred in the currencies where infrastructure and support headcount sit; revenue is billed in a different mix. A sustained currency move can shift reported gross margin by tens of basis points without any operational change.

The competitive-pressure failure mode. The scenario that actually damages gross margin is not cost — it is being forced to bundle AI into the base tier for free because a competitor did. That converts an AI cost line into pure margin destruction with no revenue offset. It is a pricing-strategy failure that arrives disguised as a cost problem, and it is worth watching pricing announcements more closely than infrastructure announcements for this reason.

The RevOps read-across. For RevOps teams inside the customer base rather than the investor base, vendor gross margin trajectory has a practical consequence: it predicts pricing behavior. A vendor absorbing AI cost inside existing gross margin will eventually push AI consumption into a metered SKU. Budget for AI-tier uplift at your next renewal rather than assuming today's per-seat price carries forward — the vendors under the most gross margin pressure are the ones most likely to restructure pricing at your renewal, and the restructure usually arrives as a new tier rather than a price increase on the old one.

Edge case worth naming: the services drag reversal. If a vendor pushes hard into a new vertical or a new geography, professional services mix rises temporarily because early deployments need more hands-on delivery. Blended gross margin falls; subscription gross margin does not. Misreading that as platform inefficiency is a common analytical error.

A practical rollout plan for tracking the trajectory

If you are actually maintaining this model — as an analyst, an FP&A partner, or a RevOps leader doing vendor due diligence — here is a workable cadence rather than a one-time estimate.

Quarter zero: build the baseline. Pull eight quarters of subscription gross margin from the filings, not from press-release summaries. Separate subscription from professional services. Note the non-GAAP reconciliation line by line so you know exactly what is excluded. Record services as a percentage of revenue in the same table — you will need it to detect mix effects later.

Every quarter: check four things in this order. (1) Sequential subscription gross margin, not year-over-year — annual comparisons hide two quarters of drift. (2) Any change in the language management uses about AI infrastructure cost; "managed within our existing corridor" and "investment phase" are meaningfully different statements. (3) Services mix, to rule out a mix explanation before reaching for a cost explanation. (4) Any new disclosure about AI-tier revenue or attach, which is the offset side of the equation and usually the last thing a vendor discloses.

Twice a year: re-underwrite the scenarios. Update the bear/base/bull spread with what you learned. The discipline is to move the *probabilities*, not the endpoints — if you find yourself moving the bear case from 80% to 76%, something structural changed and you should be able to name it.

Annually: check the peer set. If every comparable vendor compressed 150bps and your subject compressed 150bps, that is an industry cost curve, not a company-specific story. If your subject compressed 300bps while peers held flat, that is a company-specific story and worth the work of finding it.

Trigger-based reviews. Re-open the model outside the cadence when any of these happen: a new sovereign or regional cloud announcement, a change to AI packaging or pricing, a material acquisition, a change in primary model vendor relationships, or the introduction of consumption metering on AI features. Each of these moves a structural driver rather than a quarterly number.

The output of this process is not a precise FY28 number — nobody has one. It is a defensible band with named drivers, which is what a forecast of this kind is actually for.

Related questions

Does falling gross margin mean the AI strategy is failing?

No. Compression paired with rising AI-tier revenue and improving retention is a healthy trade. The warning sign is compression with flat AI monetization, which means cost is being absorbed without corresponding pricing capture — a packaging problem rather than a cost problem.

Should I model GAAP or non-GAAP gross margin?

Use non-GAAP subscription gross margin for operational trend analysis and peer comparison, since it strips stock compensation and acquisition amortization that distort cross-company comparison. Use GAAP for valuation work and cash-conversion analysis. Never mix the two in a single comparison table.

How does gross margin affect vendor pricing at my renewal?

Meaningfully. Vendors under gross margin pressure from AI cost tend to introduce metered or tiered AI SKUs rather than raise base per-seat pricing. Budget for a premium AI tier at renewal rather than assuming current pricing extends unchanged into your next term.

Which single metric best predicts the trajectory?

Cost per AI workflow, if you could see it. Since you cannot, the best public proxy is the relationship between AI-tier attach commentary and sequential subscription gross margin — rising attach with stable margin means the unit economics are working.

FAQ

Will ServiceNow gross margin fall below 80% by 2028?

Unlikely as a sustained level. The bear case reaches roughly 80%, but a durable break below that would require inference cost growth to substantially outrun efficiency gains at the same time AI-tier monetization stalls. Operating margin commitments also act as a forcing function: sustained gross margin erosion would trigger opex discipline elsewhere before it was allowed to persist.

How much of the compression is specifically AI cost?

It is not broken out publicly, and any precise split is an estimate. Directionally, AI inference is the largest of the compressors, with regional and sovereign infrastructure duplication second and data-layer growth third. The reasonable working assumption is that AI accounts for somewhere between a third and half of the total compression, with the rest split across the other drivers.

Why is ServiceNow's gross margin higher than Salesforce's?

Structural, not operational. A single multi-tenant workflow platform with a small, deliberately low-margin services business carries fundamentally better cost-of-revenue economics than a suite assembled partly through acquisition, where each acquired product brings its own infrastructure and margin profile. That gap tends to persist rather than close.

Does sovereign cloud expansion hurt margin permanently?

Not permanently, but for years rather than quarters. A dedicated regional footprint is dilutive until the regional revenue base scales enough to absorb it. The risk is expanding into more regions faster than any of them mature, which keeps a rolling investment drag on the blended number indefinitely.

What would make the bull case happen?

Routing enough workflow traffic to smaller domain-tuned models that blended cost per AI workflow declines rather than rises, combined with premium AI tier attach running ahead of plan. If inference cost per unit of delivered value falls faster than usage grows, compression never materializes.

Does gross margin matter more than operating margin here?

Operating margin matters more for valuation, because it captures the full efficiency picture including sales and marketing leverage. Gross margin matters more as an early signal, because it moves first when AI unit economics change. Watch gross margin to learn what is happening; watch operating margin to learn what it is worth.

Sources

• https://investors.servicenow.com/
• https://www.sec.gov/edgar/searchedgar/companysearch
• https://www.servicenow.com/company/media/press-room.html
• https://www.gartner.com/en/information-technology
• https://www.bvp.com/atlas/state-of-the-cloud
• https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights
• https://www.forrester.com/research/
• https://www.investor.salesforce.com/
• https://investors.snowflake.com/

Related on PULSE

• [What is Salesforce gross margin trajectory through 2028?](/knowledge/q1545)
• [What is Snowflake gross margin trajectory through 2028?](/knowledge/q1594)
• [What is Datadog gross margin trajectory through 2028?](/knowledge/q1687)
• [What is Outreach gross margin trajectory through 2028?](/knowledge/q1747)
• [What is Salesloft gross margin trajectory through 2028?](/knowledge/q1807)