What is ServiceNow gross margin trajectory through 2028?

Direct Answer

ServiceNow exits FY25 with non-GAAP subscription gross margin running ~83-84%, the high-water mark of the modern enterprise SaaS cohort, and the trajectory through FY28 is best modeled as a controlled compression of 100-300bps to a 80-83% range — not a cliff, but a deliberate trade where the company spends GM points to buy AI differentiation, sovereign cloud access, and international footprint.

Three forces compress: (1) AI inference cost passthrough from Now Assist, AI Agent Studio, and the Anthropic/OpenAI model layer, (2) sovereign cloud regional duplication in Germany, France, Saudi, and India, and (3) hyperscaler infrastructure scaling with Workflow Data Fabric / RaptorDB volume.

Two forces protect the floor: RaptorDB's 50%+ efficiency gain over the legacy data layer (the single largest internal cost lever ServiceNow controls) and the Pro Plus AI bundle pricing uplift that monetizes the same inference ServiceNow is paying for. McDermott's publicly committed 76% non-GAAP operating-margin guard-rail is the discipline forcing function — it caps how much GM erosion can be absorbed before opex has to give.

Bear case ~80%, base case ~82%, bull case ~83% by FY28 — and the swing factor is RaptorDB inference-per-query trend more than any single line on the income statement.

What ServiceNow's GM Looked Like FY24-FY26

• Non-GAAP subscription GM FY24: ~84.5% reported, the structural anchor for the modern run-rate; total non-GAAP GM ~83% blended with services drag.
• Non-GAAP subscription GM FY25: held in the 83-84% band per quarterly disclosures despite first full year of Now Assist at scale — Q1 FY26 commentary credits RaptorDB efficiency for offsetting Now Assist inference cost growth.
• GAAP vs non-GAAP gap: ~3-4 pts, driven primarily by stock-based comp allocated to cost of subscription revenue and amortization of acquired intangibles; gap widens slightly as headcount-heavy AI engineering scales.
• CFO commentary cadence: Gina Mastantuono has consistently telegraphed margin discipline as non-negotiable on earnings calls, framing AI inference cost as "managed within the existing GM corridor" rather than a separate breakout — signaling internal absorption strategy.
• Q1 FY26 specific signal: RaptorDB efficiency gains called out as protecting subscription GM despite Now Assist consumption running ahead of plan — read this as inference cost pressure being real but offset, not absent.

What Compresses GM Through FY28

• Now Assist + AI Agent Studio inference passthrough: Anthropic Claude and OpenAI GPT model calls billed per-token to ServiceNow then bundled into Pro Plus pricing — the gap between marginal inference cost and incremental Pro Plus revenue is the central GM equation. Every 100M monthly Now Assist queries at current model pricing represents an estimated 30-60bps GM exposure.
• Sovereign cloud regional duplication cost: Germany (Frankfurt), France (SecNumCloud partner), Saudi (Riyadh), India (Mumbai/Hyderabad) require dedicated infrastructure that does not benefit from multi-tenant economics — estimated 75-125bps cumulative drag on blended international GM, growing as regulated-market revenue mix grows.
• Workflow Data Fabric / RaptorDB scaling cost: The data layer ingestion volumes from Fabric (Snowflake, Databricks, SAP, Workday, ServiceNow zero-copy connectors) scale with AI agent activity — storage and query cost growth lags revenue growth by 6-12 months but is real.
• AWS / Azure / GCP infrastructure cost: As ServiceNow follows hyperscaler region expansion (free spoke addition for international growth) the absolute infrastructure spend grows even as per-region efficiency improves; Bessemer State of the Cloud 2026 benchmarks suggest hyperscaler cost-of-revenue compounds 25-35% YoY for AI-heavy SaaS.
• Expanding international support headcount: Multilingual L2/L3 support for German, French, Japanese, Mandarin, Arabic regulated customers is loaded into cost of subscription revenue — incremental headcount per Tier-1 international market estimated at 50-100 FTE.
• Public sector and sovereign discounting: UK G-Cloud, US FedRAMP High, German BSI C5 deals carry 15-25% list discounts plus higher delivery cost — accretive to revenue and strategic posture, dilutive to GM line.

What Protects The GM Floor

• RaptorDB efficiency vs. Competitor data layer: ServiceNow's owned data engine (RaptorDB Pro) reportedly delivers 50%+ throughput improvement vs. The legacy MariaDB layer per company commentary, which translates to lower compute-per-query and is the single largest internal cost lever; the cost curve runs against the inference-cost curve and that race is what determines GM trajectory.
• NVIDIA partnership for AI cost optimization: ServiceNow + NVIDIA collaboration on domain-specific small language models (Now LLM family) reduces dependency on premium frontier-model API calls — the Now-LLM-tuned-on-Now-Platform-data approach can swap a $0.01-0.03 frontier API call for a $0.001-0.005 in-house inference call on customer-specific workflows.
• Customer commits + multi-year contracts: ~98% renewal rate plus growing multi-year commit structure (3-year+ becoming standard at Fortune 500) gives ServiceNow forward visibility to spread sovereign and AI infrastructure capex across known revenue.
• Pro Plus margin uplift on AI bundling: Pro Plus skus carry 30-50% pricing premium over Pro for AI-enabled workflow modules; if Pro Plus penetration reaches 30%+ of installed base by FY28, the bundle uplift offsets inference cost passthrough at the contribution-margin level.
• Scale leverage on fixed cost base: Total revenue growing toward $20B+ by FY27 lets ServiceNow amortize the absolute capex of sovereign clouds, data center expansion, and AI infrastructure across a larger denominator — operating leverage masks some GM line pressure.

The 3 Scenarios FY27-FY28

• Bear case (~80% subscription GM by FY28): Inference cost outruns RaptorDB efficiency curve, Pro Plus penetration stalls below 25%, sovereign cloud commitments expand to Indonesia/Vietnam/South Africa adding 50bps drag, and FX headwinds compound; named pull-throughs failing — Walmart, JPMorgan, Toyota delaying Pro Plus expansion.
• Base case (~82% subscription GM by FY28): Inference cost growth managed within RaptorDB + Now-LLM efficiency curve, Pro Plus penetration reaches 28-32%, sovereign cloud expansion stays disciplined to current named markets, FX neutral; named pull-throughs delivering — Bundeswehr, HMRC, Aramco, MUFG ramping on schedule.
• Bull case (~83% subscription GM by FY28): RaptorDB Pro v2 + Now-LLM family deliver step-function efficiency gain, Pro Plus exceeds 35% penetration, NVIDIA partnership produces inference cost step-down, and sovereign-cloud-eligible customers absorb 20-30% premium pricing — fully neutralizing the drag; named pull-throughs accelerating — Microsoft loses head-on AI workflow deals to ServiceNow.

What Investors Should Watch

• Subscription GM line each quarter: The single most important number in the model — watch for Q-over-Q sequential pressure, especially Q2 and Q3 when Now Assist consumption typically ramps.
• Pro Plus revenue mix disclosure: ServiceNow has been opaque on exact Pro Plus penetration; any new disclosure on AI bundle revenue contribution is a leading indicator of GM offset capacity.
• Sovereign cloud cost trend in CFO commentary: Listen for Mastantuono's framing — "managed within corridor" is bullish, "investment phase" is neutral, "transitory pressure" is bearish.
• Now Assist infrastructure cost line: If ServiceNow ever breaks out AI infrastructure as a separate cost-of-revenue line, that's a structural shift — currently buried in cost of subscription revenue.
• S&M leverage as offset: Operating margin guidance holds at 30-31% non-GAAP; if S&M productivity improvements deliver opex leverage, GM compression matters less to the operating-margin guard-rail.
• RaptorDB Pro efficiency callouts: Each earnings call mention of RaptorDB efficiency improvements is a direct cost-side data point — count the mentions and track the language intensity.

Comparable GM Trajectory

• Salesforce (~76% subscription GM): Lower structural GM driven by acquisition-heavy growth (Slack, Tableau, MuleSoft each carry lower GM profiles) and Data Cloud / Agentforce inference cost — ServiceNow sits ~700-800bps higher and the gap is structural, not closing.
• Workday (~80% subscription GM): HCM + Financials platform with similar multi-tenant SaaS economics; the closest pure-play comp to ServiceNow on infrastructure profile but lacks the AI bundling premium ServiceNow gets from Pro Plus.
• Snowflake (~74-77% subscription GM): Consumption-based model with direct AWS/Azure/GCP cost passthrough makes their GM structurally lower; ServiceNow's seat-based + bundled-AI model captures more value per inference dollar.
• Datadog (~80% subscription GM): Observability with similar consumption pressure from log/metric/trace volume growth — Datadog's GM trajectory has been slowly compressing for similar reasons (AI-driven query volume), useful read-across for ServiceNow.
• Why ServiceNow sits higher: Workflow platform with sticky CMDB anchoring, low services drag (services run near breakeven by design), and the highest enterprise pricing power in the cohort — that combination supports the 80%+ floor that competitors structurally cannot reach.

GM Trajectory Table

GM Drivers and Floor Outcome

Bottom Line

ServiceNow's GM trajectory through FY28 is a controlled compression story, not a structural break — base case ~82% non-GAAP subscription GM with a bear floor at ~80% and a bull ceiling at ~83%. The race that determines the outcome is RaptorDB efficiency vs. AI inference cost, and ServiceNow controls both sides of that equation more than any peer in the SaaS cohort.

McDermott's 76% non-GAAP operating-margin guard-rail is the discipline forcing function — it caps the absorption envelope and forces hard trade-offs on sovereign cloud expansion pace, Pro Plus pricing strategy, and S&M productivity. The investable read is: ServiceNow stays the highest-GM enterprise SaaS at scale through FY28, and the comp-set gap to Salesforce, Workday, Snowflake, and Datadog widens before it narrows.

(see also: q1610, q1612, q1617, q1626)