ServiceNow (NOW) Q3 2026: AI Contract Targets Rise, but Shareholder Returns Must Still Clear Margin and Acquisition Costs
ServiceNow's Q3 2026 subscription revenue was US$3.988 billion, up 21%, and the year-end AI annual contract value target was raised for the second time this year to US$1.7 billion. But non-GAAP subscription gross margin fell from 83% to 80%, and in nine months the balance sheet swung from about US$8.6 billion of net cash to about US$600 million of net debt. Following our pricing-power thesis, we separate AI contracts, recognized revenue, usage and cash: pricing power held at the renewal layer, AI is already earning money at the plan layer, and this year's growth still has to prove itself next year.
📌 Key Conclusions
- Q3 subscription revenue was US$3.988 billion, up 21% year over year; cRPO growth at constant currency remained between 21% and 21.5% for three consecutive quarters, and the year-end target for AI annual contract value was raised for the second time this year to US$1.7 billion, and customers are still spending more on ServiceNow.
- Non-GAAP (adjusted) subscription gross margin fell from 83% to 80%; of the US$388 million increase in subscription costs, US$119 million in amortization accounted for approximately 30%, with most of the rest attributable to personnel, third-party cloud, and infrastructure costs. Full-year guidance of 81% implies that Q4 must recover to approximately 81% or above.
- Operating cash flow for the first three quarters was US$3.096 billion, down approximately 3.4% year over year, and grew only approximately 7% even after adding back acquisition-related cash payments; the improvement in FCF came mainly from lower capital expenditures and expense add-backs.
- In nine months, approximately US$8.6 billion in net cash became approximately US$600 million in net debt; goodwill and intangible assets totaled approximately US$13.6 billion, exceeding shareholders' equity; there were no share repurchases in Q2 or Q3.
- Pricing power held at renewal, AI is already generating revenue at the plan level, and usage-level charges have not been disclosed; next, watch whether Q4 non-GAAP (adjusted) subscription gross margin is at least 80.5%, actual cRPO growth holds at 20% year over year, and operating cash flow growth catches up with revenue.
Demand is not the problem. The cost is. cRPO growth at constant currency remained between 21% and 21.5% year over year for three consecutive quarters, and the year-end target for AI annual contract value was raised twice, from US$1 billion at the start of the year to US$1.7 billion, and customers are still spending more on ServiceNow. But in 2026, the company traded three things for that growth: non-GAAP (adjusted) subscription gross margin fell from 83% to 80%; the balance sheet moved from approximately US$8.6 billion in net cash to approximately US$600 million in net debt in nine months; operating cash flow for the first three quarters fell approximately 3% from a year earlier and grew only approximately 7% even after adding back acquisition-related cash payments, far behind revenue. My judgment is: Pricing power has held at renewal, and AI has begun generating revenue; but a considerable part of this year's growth was bought, and the next two quarters will show whether it can earn that cost back on its own.
From the battle over SaaS pricing power to today
PVL's research on ServiceNow began with the April 12 v2.0, “AI Is Putting SaaS Pricing Power on Trial”: after agentic AI reduces manual work, will it dismantle the basis of ServiceNow's per-seat pricing? The May v3.0 gave a more optimistic answer, arguing that NOW is a coordination layer across systems, charging for workflows and decoupling from the number of seats; the AI billing research published the same month concluded that ServiceNow would pass inference costs on to customers through a pool of credits and overage charges. Original NOW research v3.0; AI billing research.
The verdict in April's Q1 earnings research was that AI had not dismantled NOW's subscription base, while hybrid billing had yet to produce verifiable numbers. Six months later, the first part still holds, and the second has progressed; but several judgments in v3.0 and the AI billing research also need to be corrected this time.
First, a correction to our own comparison error. Reported cRPO growth in Q1, Q2, and Q3 2026 was 22.5%, 21%, and 20.5% year over year, respectively; growth at constant currency was 21%, 21.5%, and 21%, respectively. The Q1 article stated that cRPO growth of +22.5% “exceeded Q4 FY25's +21% cc,” comparing reported growth with constant-currency growth; on a consistent basis, both quarters were 21%. The v3.0 article stated that Veza was acquired in December 2025 for approximately US$1 billion; the 10-Q states that the acquisition closed on March 2, 2026, for approximately US$1.2 billion, and we correct that here as well. The correct reading is: Demand is on a plateau of approximately 20%, neither collapsing nor accelerating.Q1 announcement; FY2025 Q4 announcement.
The progress is in AI contracts. In Q2, the company announced that ServiceNow AI annual contract value had surpassed US$1 billion, reaching the year-end target set at the start of the year; in Q3 it raised the year-end target again to US$1.7 billion, 70% above the original target. The original research tracked Now Assist ACV (approximately US$600 million); the ServiceNow AI measure the company now reports is broader, so the three figures cannot be joined into a consistent growth curve. But the difference in scope does not change one fact: Management raised its AI contract target for the second time this year, which is direct evidence that customers are willing to pay for AI, and the strongest support to date for the bullish case in the original research.Q2 announcement; Q3 announcement.
So this time the question moves one step further: Once AI starts generating revenue, can the additional contracts become more cash per share after cloud costs, acquisition amortization, and interest? The original research treated integration costs as a short-term issue, and Q3 gross margin and cash flow tell us that the cost is greater than expected and will continue at least through year-end.
Industry: AI value will concentrate in the layer that can execute and take responsibility for outcomes
ServiceNow sits between AI models and enterprise systems. Upstream model and cloud providers supply inference and computing resources, while enterprise ERP, CRM, and identity systems hold existing records, and ServiceNow turns a request into a workflow across systems: who approves, who executes, who is responsible, and what records are retained. Models can answer questions, but enterprises need work completed safely, and This is the layer where ServiceNow stands.
I believe this position will become more valuable, not less, as competition among models increases. The more model providers there are and the more intense the price competition, the lower ServiceNow's cost of purchasing inference, and the more customers need an execution layer that is not tied to a single model and can manage permissions and audits. But it is not yet time to reap the rewards: the 10-Q shows that third-party cloud contract expenses rose US$79 million year over year this quarter, and model and cloud providers are still taking a share of the additional value.
Microsoft approaches from productivity tools and Copilot, Salesforce from CRM and customer service, and enterprises can also connect models themselves. The real threat is not wholesale replacement, but being relegated to the back end. If employees become accustomed to issuing instructions through someone else's AI interface, ServiceNow will be left with only the value of execution and governance.
The company's response is to bet on both sides. On one side, it connects to others' entry points: ServiceNow launched a ChatGPT plugin that connects enterprises from OpenAI inference to governed execution; it also works with Google Cloud to let employees launch ServiceNow workflows directly from Gemini Enterprise, with NVIDIA and others as additional partners. On the other side, it is building its own entry point: in Q2 it launched ServiceNow Otto, integrating Now Assist, Moveworks, and AI Experience, while a customer example in the Q3 announcement presented EmployeeWorks as employees' “AI front door.” Q3 announcement. My reading is: ServiceNow is not betting that it will necessarily own the entry point, but is ensuring that, whichever entry point customers use to issue instructions, the work ultimately runs on its platform. Its bargaining power therefore depends on how indispensable the execution layer is. The best evidence is that large customers are still increasing their spending: in Q3, 680 customers had annual contract value above US$5 million, up approximately 23% year over year, and there were 19 new contracts individually exceeding US$5 million, up more than 135% year over year.
Business model: Tickets create switching costs, while seats and allowances determine the pace of revenue collection
ServiceNow started in enterprise IT service management: when employees encounter account, device, or system problems, enterprises use the platform to open tickets, set permissions, and run workflows. The same workflows later expanded to employee services, customer service, security, risk, and application development. It sells not a model, but a way to connect people, data, permissions, and existing systems so that each task can be assigned, approved, executed, and recorded.
Switching costs are real. Enterprises build tickets, service catalogs, permissions, system integrations, and internal rules on the platform; replacing it requires migration, retesting, and retraining users. The renewal rate reported by the company was 98% in each year from 2023 through 2025, and also 98% in Q3 2026. Note that this 98% is calculated based on ACV, so a customer reducing some modules does not count as churn, and it is not a net revenue retention rate; therefore, 98% proves that “customers stay,” not that “customers buy more.” Q3 10-Q definition of renewal rate.
Two factors determine the pace of revenue collection. First, subscription fees are generally billed annually in advance at signing or renewal, cloud subscriptions are recognized evenly over the service period, and the annual report also notes that collections are usually highest in the first quarter. Cash is received before revenue is recognized, so quarterly operating cash flow must be read across quarters and cannot be used directly to compare operating performance in a single quarter. Q3 10-Q revenue recognition and billing explanation
Second, AI is charged on top of seats. Official documentation says add-on plans such as Pro Plus, Enterprise Plus, and Creator Plus are purchased per seat, with each seat carrying a set allowance of assists and more complex interactions consuming more; it also states that the initial allowance is generally sufficient, and assist packs are purchased only when more capacity is needed. An editor's note on that page says Now Assist has become ServiceNow Otto; according to the Q2 announcement, Otto is a unified experience integrating Now Assist, Moveworks, and AI Experience. Actual allowances and credit deductions remain subject to each customer's contract. Official billing explanation.
The intent of this design is clear: Get customers using it within their allowance first, then charge at renewal once usage grows. It protects short-term renewal and seat revenue, at the cost of ServiceNow absorbing inference costs first.
Two earlier articles need correction here. First, v3.0 said NOW was “decoupled from the number of seats”; the official billing page shows that AI add-on plans are still purchased per seat, and AI pricing remains tied to seats today; what has decoupled is the value of workflows, not the way they are charged. Second, the AI billing research concluded that the credit pool was designed to pass on costs and that metered billing was a “defense of gross margin”; by Q3, that defense had not appeared in the financial results, and non-GAAP (adjusted) subscription gross margin had instead declined. At least so far, ServiceNow has chosen to absorb costs first in exchange for adoption; the cost pass-through expected in the earlier article has not yet happened.
Three years of growth: Operating leverage is real, and gross margin pressure predates the acquisitions
The following covers the three full years from 2023 to 2025, in US$ million. Free cash flow uses a simplified measure, operating cash flow less purchases of property and equipment, and is not equivalent to the company's Non-GAAP definition, which adds back acquisition or legal expenses.
Metric | 2023 | 2024 | 2025 | Comparison |
|---|---|---|---|---|
Revenue | 8,971 | 10,984 | 13,278 | Growth of approximately 22.4% and 20.9% in 2024 and 2025, respectively |
Subscription revenue | 8,680 | 10,646 | 12,883 | Approximately 97% of total revenue in 2025; subscriptions are the primary revenue source |
GAAP subscription gross margin | 81.5% | 81.8% | 80.1% | Fell approximately 1.7 percentage points in 2025; cost pressure had already emerged before the 2026 quarters |
GAAP operating income | 762 | 1,364 | 1,824 | As revenue grew, marketing and other expenses fell as a share of revenue, and operating income expanded |
GAAP operating margin | 8.5% | 12.4% | 13.7% | Improved approximately 5.2 percentage points over two years |
GAAP net income | 1,731 | 1,425 | 1,748 | 2023 includes a tax impact and cannot be directly equated with the operating income trend |
Cash flow from operating activities | 3,398 | 4,267 | 5,444 | Grew approximately 60% over two years, but includes advance payments and stock-based compensation added back |
Purchases of property and equipment | 694 | 852 | 868 | Capital expenditures increased but remained below operating cash flow in each year |
Operating cash flow less purchases of property and equipment | 2,704 | 3,415 | 4,576 | Substantial growth, before deducting cash paid for acquisitions |
Stock-based compensation expense | 1,604 | 1,746 | 1,955 | Does not require cash salary payments in the current period, but carries economic costs for shareholders, including dilution |
Source: Three years of income statements and cash flows in the 2025 annual report. Gross margins are self-calculated from subscription revenue and subscription costs and may differ from the company's rounded figures.
The three years of operating leverage are real: revenue increased by approximately US$4.31 billion and GAAP operating income by approximately US$1.06 billion, while sales and R&D expenses did not consume the additional revenue. But subscription gross margin had already fallen 1.7 percentage points in 2025, when neither Armis nor Veza had yet joined the company. The acquisitions did not cause the gross margin pressure; they amplified it. This is the main reason I judge that part of the 2026 gross margin problem is structural.
A large part of the gap between operating cash flow and net income comes from adding back stock-based compensation. Stock-based compensation does not require a cash payment, but shareholders bear dilution, or the company spends money on buybacks to offset it, so I track cash flow per share rather than the total. At its May analyst day, the company committed to reducing stock-based compensation to below 10% of revenue by 2029; it was approximately 14.7% in 2025 (self-calculated). But this year's trend runs counter to that commitment: stock-based compensation was 15.5% of revenue in the first three quarters, and full-year guidance is 15%, both above 2025. To get below 10% by 2029, the share must first fall noticeably in 2027, or the commitment will be nothing more than an analyst-day slogan. I regard it as a more important indicator of per-share value than FCF margin. Q2 announcement
Contracts and AI monetization: Signing, recognition, usage, and cash collection are four separate things
Q3 metric | Company-reported figure | What it represents |
|---|---|---|
Subscription revenue | US$3,988 million, up 21% year over year, 20.5% in constant currency | Recognized revenue |
Total revenue | US$4,106 million, up 20.5% year over year | Includes professional services |
cRPO | US$13.65 billion, up 20.5% year over year, 21% in constant currency | Contract revenue expected to be recognized over the next 12 months |
RPO | US$29.7 billion, up 22% year over year, 22.5% in constant currency | Remaining performance obligations over a longer period |
Annual AI ACV target | US$1.7 billion by the end of 2026 | Management target, not revenue recognized this quarter |
Source: Company Q3 announcement and definitions. The AI ACV target is an annualized contract value and cannot be added to RPO or a single quarter's revenue.
First, contracts. The original research identified cRPO falling below 20% as a warning sign and inferred that growth needed to accelerate to 22% to 23% in the second half. Neither happened in Q3: it did not fall below the warning line, and it did not accelerate. The company's Q4 cRPO guidance is 18.5%, apparently below the warning line. But the company's cRPO guidance has consistently been conservative: in Q1 it guided Q2 to 19%, and the actual result was 21% (see this site's Q1 earnings research); in Q2 it guided Q3 to 19.5%, and the actual result was 20.5%, exceeding guidance by 1 to 2 percentage points each quarter. Q2 announcement. Applying the same gap, the actual Q4 figure will land between 19.5% and 20.5%, right on the original research's 20% warning line (the warning line uses the reported figure). The Moveworks acquisition closed on December 15, 2025, so it enters the comparison base in Q4 and the acquisition's contribution to growth begins to disappear, one mechanical reason for the lower guidance. v3.0 separately sets "cRPO falling below 18%" as a downgrade condition, and guidance is only 0.5 percentage points above it. My judgment is: Q4 will not be a demand collapse, but the company itself already expects cRPO to slide from its 21% plateau toward 20%, so the original research's growth assumptions must be lowered.
Q3 subscription revenue growth fell from 24.5% in Q2 to 21%, primarily for reasons other than a slowdown. The Q2 announcement explained that strong U.S. federal demand pulled some on-premises subscription revenue originally scheduled for recognition in Q3 into Q2.
Acquisitions also contribute to growth. In its Q1 guidance, the company estimated that Armis contributed approximately 1.25 percentage points to Q2 subscription revenue and cRPO growth; Veza (March 2026) and Moveworks (December 2025) were also absent from the prior-year Q3 base. Subtracting Armis alone reduces the 21% constant-currency figure to approximately 19.75%, and after subtracting the other two, organic cRPO growth in the existing business is already below 20% (self-calculated; Veza's and Moveworks' contributions were not disclosed). v3.0 said organic growth was "still above +20%," and that judgment must be lowered. Organic growth of around 19% remains healthy, but this is the figure to use when assessing ServiceNow's core business, rather than the reported 21%. Q1 announcement
Next, AI monetization. Breaking it into four layers makes clear what has been demonstrated and what remains unknown:
Layer | What can be confirmed now | What has not been disclosed |
|---|---|---|
AI contracts (ACV) | Annual contract value exceeded US$1 billion in Q2; year-end target of US$1.7 billion | How much represents new payments and how much comes from allowances included in existing plans |
Recognized revenue | Q3 subscription revenue of US$3.988 billion and first-three-quarter revenue of US$11.536 billion, across all products | Recognized AI revenue |
Usage | Production-environment agentic AI usage in the first three quarters increased by more than 13 times year over year | The measurement definition, absolute base, and proportion paid for beyond included allowances |
Cash collection | Subscription fees are mostly invoiced annually in advance; operating cash flow for the first three quarters was US$3.096 billion | AI-related cash collections and gross margin |
ACV is an annualized contract value, best compared with annualized subscription revenue. On that basis, Q2's US$1 billion was approximately 6% of annualized subscription revenue for the quarter (US$3.877 billion multiplied by 4); the US$1.7 billion year-end target is approximately 10% of annualized Q4 subscription revenue guidance (approximately US$4.25 billion multiplied by 4) (both are my estimates and compare contract value, not recognized revenue). If the year-end target is met, AI will grow from approximately one-sixteenth to approximately one-tenth of the subscription base in half a year, enough to prove that AI is already one of ServiceNow's growth engines, not marketing rhetoric. At its May analyst day, the company set a target for AI to account for 30% of ACV in 2030; 10% is only one-third of the way there.
But ACV does not turn into revenue all at once. Take a hypothetical example: a three-year cloud subscription totaling US$3 million, with equal amounts in each year, has an ACV of US$1 million; if service starts on October 1 and revenue is recognized evenly over time, only approximately US$250,000 is recognized that year. Therefore, the amount of the US$1.7 billion year-end target reflected in 2026 revenue will be far below US$1.7 billion; the company also has not separately disclosed recognized AI revenue. Q3 10-Q revenue recognition policy.
The most important comparison is between usage and contracts. The company said that from January 1 to September 30, 2026, cumulative production-environment agentic AI usage increased by more than 13 times from the same period last year; in the same announcement, the CFO said new annual contract value for AI exceeded expectations, prompting a second increase in the year-end target. Official Q3 announcement. The 13-fold figure is year-over-year usage growth, while 70% is the increase in the target; they measure different things and cannot be converted into each other.
This leads to my central judgment on AI monetization: ServiceNow sells plans, not usage. AI revenue follows ACV, and the company begins recognizing it over time when a customer signs a plan that includes AI; the company is already collecting money at this layer. The 13-fold increase in usage brings stickiness and costs: deployment deepens, switching costs rise, and each use also incurs inference and cloud costs. For example, if a customer whose contract includes an allowance of 1 million points increases usage from 200,000 to 800,000, ServiceNow does not collect more money; usage turns into revenue only when the customer exhausts the allowance and purchases more, or upgrades the plan at renewal (the points are a hypothetical example). The company has not disclosed either additional purchases beyond allowances or upgrades at renewal. What can be confirmed now is that AI is already generating payments at the plan level, usage-based charges are not yet visible, and costs are already showing up in gross margin.
Returning to the original research's question about pricing power: has AI offset seat risk? It has at the contract level, but not yet at the revenue and profit level. The company has not provided the breakdown needed to calculate directly whether customers' additional payments for AI exceed the combined effect of fewer seats and higher delivery costs. Going forward, the most telling data will be purchases of additional capacity, total contract value at renewal, and whether the non-GAAP (adjusted) subscription gross margin can stop falling.
Profit: Acquisition Accounting Accounts for 30% of Gross Margin Pressure; Most of the Rest Is Real Delivery Costs
Figures are in US$ million.
Metric | 2025 Q3 | 2026 Q3 |
|---|---|---|
Total revenue | 3,407 | 4,106 |
Gross profit | 2,633 | 2,895 |
Operating expenses | 2,061 | 2,510 |
GAAP operating income | 572 | 385 |
GAAP net income | 502 | 277 |
GAAP diluted EPS | 0.48 | 0.27 |
Non-GAAP operating income | 1,140 | 1,338 |
The company has retrospectively adjusted last year's per-share figures for the stock split. Gross profit increased by US$262 million, while operating expenses increased by US$449 million, causing GAAP operating income to fall by US$187 million, or approximately 30%. Revenue grew 20%, yet GAAP operating income fell. There was no operating leverage at the GAAP level this quarter, and readers who look only at the 17% growth in Non-GAAP operating income will miss that.
GAAP subscription gross margin was 73.5%, versus 80% last year; the non-GAAP (adjusted) figure was 80%, versus 83% last year. Even after excluding acquisition amortization and stock-based compensation, it still fell 3 percentage points. This proves that gross margin pressure is not just an accounting issue.GAAP and Non-GAAP reconciliation
The 10-Q provides a clear breakdown of the increase in subscription costs: in 2026 Q3, subscription costs increased by US$388 million from a year earlier, including increases of US$119 million in intangible asset amortization, US$112 million in personnel and related costs, US$79 million in third-party cloud contract costs, and US$56 million in infrastructure depreciation, software, maintenance and other costs. The four items total US$366 million, leaving approximately US$22 million unspecified. On that basis, approximately 30% came from acquisition accounting, while approximately 65% came from real delivery costs such as personnel, cloud and infrastructure. The 10-Q also says the increase in personnel costs included expenses to support customers in regulated markets. The increase in third-party cloud costs had two sources: the Q2 announcement said gross margin guidance reflected more customers moving to hyperscaler hosting and accelerating customer AI adoption; the proportion attributable to each was not disclosed. The company also states directly in the 10-Q that it expects full-year 2026 subscription gross margin to be lower than in 2025, mainly because of growth in third-party cloud usage and additional amortization from acquisitions; based on existing intangible assets, amortization will still be US$783 million in 2027 and approximately US$3.594 billion in total thereafter. In other words, the acquisition drag on GAAP gross profit will continue for several years.
Subscription costs increased approximately 58% year over year, far above the approximately 21% revenue growth rate; however, the US$689 million increase in revenue still exceeded the US$388 million increase in costs, and subscription gross profit increased by approximately US$301 million. Each additional dollar of revenue is still profitable, but the gross profit retained has narrowed.2026 Q3 10-Q.
So is this a case of “invest first, harvest later,” or “the more it is used, the more money it loses”? I judge it to be the former, for two reasons. First, adjusted subscription gross profit for the first three quarters was US$9.318 billion and subscription revenue was US$11.536 billion, giving a gross margin of approximately 80.8%; management's full-year guidance is 81%, and using the US$15.785 billion midpoint of full-year subscription revenue guidance and the US$4.25 billion Q4 midpoint, Q4 must recover to above approximately 81%, meaning management is signaling that gross margin will bottom and rebound (self-calculated; allowing for the company's rounding, the lower bound is approximately 80.7%). Second, on an unrounded basis, the margin fell only approximately 0.4 percentage points from Q2 to Q3; I infer that part of this came from on-premises revenue being recognized earlier in Q2, boosting Q2 gross margin, and that the underlying margin was essentially flat. The condition that would overturn this judgment is also clear: if Q4 adjusted subscription gross margin is below 80.5%, full-year guidance will be missed, and I will change my judgment to the latter.Q3 announcement and full-year guidance
The difference between GAAP and adjusted operating income this quarter was US$953 million, comprising US$633 million in stock-based compensation, US$198 million in intangible asset amortization, US$35 million in acquisition-related costs, and US$87 million in severance. Stock-based compensation was US$630 million on the cash flow statement, a different basis from the operating income reconciliation. These adjustments help readers observe recurring operations, but stock-based compensation dilutes shareholders, and the acquisitions were paid for with real money; adjusted profit cannot be treated as profit without costs.
A large part of the EPS decline was also unrelated to operations. Pretax income fell from US$694 million to US$370 million, a decline of US$324 million. Of that, interest expense rose from US$6 million to US$81 million because of acquisition borrowing; interest income fell from US$115 million to US$69 million, which the 10-Q attributed to declines in both investment portfolio balances and interest rates, and the decline in balances was itself the result of cash used for acquisitions and repurchases. Net interest swung from positive to negative, falling by US$121 million in total, or approximately 37% of the pretax decline; interest on new borrowing accounted for approximately 23% of the decline (self-calculated). EPS fell from 0.48 to 0.27, and nearly 40% of the decline came from changes to the balance sheet, not a loss of competitiveness in AI products.
A closer look at the timeline makes the shape of the pressure clearer. The table below retains the company's rounded subscription gross margins, while operating margins are self-calculated by dividing each quarter's income by revenue.
2026 quarter | Q1 | Q2 | Q3 |
|---|---|---|---|
GAAP subscription gross margin, company reported | 77.5% | 73.5% | 73.5% |
Adjusted subscription gross margin, company reported | 81.5% | 80.5% | 80.0% |
GAAP operating margin, self-calculated | 13.3% | 4.1% | 9.4% |
Adjusted operating margin, self-calculated | 31.8% | 29.4% | 32.6% |
Sources: Q1 announcement, Q2 announcement, Q3 announcement. Veza's acquisition closed in March and Armis's in April, so the consolidation scope and cost mix differed across the three quarters.
Q2 was the bottom for operating margin, and the worst quarter is behind us; gross margin has yet to stop falling. Armis was consolidated in April, GAAP operating margin fell to 4.1% in Q2 and recovered to 9.4% in Q3, and Q4 guidance is 12% on a GAAP basis and 32% adjusted. Adjusted operating margin recovered to 32.6% in Q3, above Q1 but still below 33.5% in the same quarter last year. Integration is not finished either: Q3 severance of US$87 million exceeded the US$80 million total for the first half, and Q4 guidance still includes severance and acquisition-related costs of approximately 1% of revenue each. v3.0 had expected adjusted operating margin to recover to 33% to 34% in the second half and gross margin to have room for structural improvement; neither expectation materialized. More revenue from on-premises software is recognized upon delivery, which can cause quarterly gross margin fluctuations, so I use whether full-year guidance is achieved in Q4 as the criterion for my judgment rather than chasing decimal points quarter by quarter. Q3 10-Q revenue recognition policy.
Free Cash Flow: An Attractive Number of Low Quality
Cash flow, US$ million | 2025 Q3 | 2026 Q3 |
|---|---|---|
CFO | 813 | 839 |
Cash paid for PP&E | 235 | 91 |
CFO less PP&E | 578 | 748 |
Add back acquisition-related cash payments | 14 | 74 |
Company Non-GAAP FCF | 592 | 822 |
Company FCF increased by US$230 million, of which CFO contributed only US$26 million (approximately 11%); the rest came from US$144 million less in equipment spending and US$60 million more in acquisition expense add-backs. FCF grew nearly 40%, while operating cash inflow itself barely grew.
The figures for the first three quarters are more direct. CFO was US$3,096 million, down approximately 3.4% from US$3,206 million last year; CFO less PP&E was US$2,750 million, above US$2,576 million last year. Company adjusted FCF was US$3,121 million, including US$371 million in acquisition-related cash payments added back. The 10-Q primarily attributed the decline in operating cash flow to acquisition-related expenses; after adding back US$371 million for the first three quarters (US$28 million last year), operating cash flow was approximately US$3,467 million, up approximately 7% from US$3,234 million last year (self-calculated). The company's own adjusted FCF margin also fell from 27% in the same period last year to 26.5%. After excluding acquisition expenses, operating cash flow grew only 7% while revenue grew 20%; this is the gap that most deserves attention this quarter. Full-year guidance puts operating cash flow at 37% of revenue, below 41% in 2025.
Management describes the decline in capital expenditures as a more capital-efficient operating model. I infer that these are two sides of the same coin: lower equipment spending and higher third-party cloud costs amount to moving capital expenditure on company-owned data centers into subscription costs. FCF looks better as a result, and gross margin looks worse; the 10-Q attributes the decline in gross margin to growth in third-party cloud usage, which fits. This is not a bad strategy, as cloud flexibility is better suited to uncertain AI usage, but readers cannot cheer FCF and worry about gross margin without recognizing that they are looking at the same thing.
On a per-share basis, basic weighted-average shares for the first three quarters fell from approximately 1,035.96 million shares in 2025 to 1,033.81 million shares in 2026; on that basis, self-calculated operating cash flow less purchases of fixed assets per share rose from approximately US$2.49 to US$2.66, an increase of approximately 7%, while operating cash flow per share fell from approximately US$3.09 to US$2.99. The share counts use the company's retrospective stock-split adjustment; this is a calculation for research purposes, not company-reported FCF per share, and it does not deduct acquisition cash payments.
Shares outstanding after deducting treasury shares were approximately 1,047.28 million shares at the end of 2025 and approximately 1,038 million shares at the end of September 2026, so repurchases did reduce the share count. But repurchases did not increase operating cash inflow per share for shareholders; the improvement in FCF per share came from spending less on capital expenditures, and that improvement will disappear when equipment spending rises again.Q3 10-Q cash flow, equity and per-share notes.
Capital Allocation: From Repurchases to Offset Dilution to Debt-Funded Acquisitions
Net cash paid for business combinations in the first three quarters was US$9,167 million, 3.3 times CFO less PP&E for the same period (US$2,750 million). The Armis acquisition, completed in April 2026, had provisional consideration of approximately US$7.6 billion, paid entirely in cash; Veza, completed in March, was approximately US$1.2 billion, also mainly in cash. Armis strengthens security for connected devices and cyber-physical assets, while Veza strengthens identity and access governance. Moveworks, acquired at the end of 2025, focuses on enterprise search and AI assistants; the actual acquisition consideration listed in the 10-Q was approximately US$2.407 billion, comprising stock and cash, on a different basis from the initially announced transaction value. 10-Q business combinations note.
The balance sheet makes clear where the money came from. At the end of 2025, ServiceNow had approximately US$10.06 billion in cash, short-term and long-term marketable securities combined and US$1.49 billion in debt, for approximately US$8.6 billion in net cash; by the end of September 2026, the three items totaled approximately US$6.96 billion, while short-term debt of US$2.086 billion plus long-term debt of US$5.438 billion totaled US$7.52 billion, turning into approximately US$600 million in net debt (self-calculated). Goodwill increased from US$3.578 billion to US$10.012 billion, while intangible assets increased from US$1.121 billion to US$3.594 billion; together they totaled approximately US$13.6 billion, already exceeding shareholders' equity of US$13.1 billion. In nine months, ServiceNow went from a software company with ample net cash to a company whose balance sheet is bet on acquisition synergies.
I believe the strategic direction is sound. If a device vulnerability or excessive permissions are detected, the system can automatically open a ticket, assign an owner and track remediation, upgrading ServiceNow from a ticketing tool to a security governance platform; the more AI agents there are, the more enterprises need to know which assets each agent can access and with what permissions, and Armis and Veza fill those two gaps. But the price is high. Armis's provisional goodwill was approximately US$5.323 billion, or approximately 70% of the consideration, putting the vast majority of the purchase price on future integration and synergies. In Q1, management expected Armis to reduce 2026 subscription gross margin by approximately 0.25 percentage points, operating margin by approximately 0.75 percentage points, and FCF margin by approximately 2 percentage points, and expected margins to return to expansion in 2027. Q1 announcement Whether margins return to expansion in 2027 as management says is the scorecard for this acquisition. Public information is insufficient to calculate a standalone payback period for Armis, so I will judge the acquisition by cross-selling, gross margin and cash per share rather than by growth in consolidated revenue.
There is no near-term debt repayment concern. The company borrowed US$4 billion in a term loan for Armis in April 2026 and repaid it in May mainly with proceeds from issuing notes; the two principal amounts should not be counted twice. At the end of September, cash and cash equivalents were approximately US$2.393 billion and short-term marketable securities were approximately US$2.214 billion, totaling approximately US$4.607 billion; together with approximately US$7 billion in long-term marketable securities, there was also an undrawn US$3 billion revolving credit facility maturing in 2031. Long-term notes had US$5.5 billion in principal, with maturities from 2028 to 2056; commercial paper was approximately US$2.1 billion, with a weighted-average remaining term of approximately 62 days and an interest rate of approximately 4.05%, and requires rolling refinancing. Q3 10-Q liquidity and borrowings notes. Debt repayment is not a problem, but interest has become a fixed cost: Q3 interest expense was US$81 million, or approximately US$320 million annualized (self-calculated), an expense that did not exist before.
The pace of shareholder returns has also changed. Cash paid for repurchases in the first three quarters was US$2,225 million, all in Q1; no common shares were repurchased in Q2 or Q3, and approximately US$4.2 billion remained under the repurchase authorization. Liquidity on the balance sheet and the credit facility are still available, so zero repurchases for two consecutive quarters were management's choice. My interpretation is that management is prioritizing cash for integration and deleveraging, and the scale of future repurchases will follow FCF for the period rather than cash on the balance sheet. For shareholders who rely on repurchases to offset dilution from stock-based compensation, this makes the previously mentioned goal of “stock-based compensation below 10% by 2029” more important.
My Judgment: Pricing Power Has Held, but This Year's Growth Must Be Proven Next Year
Returning to the central thread of this research series, my conclusion has three layers.
First, pricing power has held at renewal. A 98% renewal rate, cRPO growth of 21% to 21.5% in constant currency for three consecutive quarters, and approximately 23% year-over-year growth in customers above US$5 million show that the Q1 bear thesis that “AI will dismantle the subscription base” had still not appeared in the numbers by Q3.
Second, AI is beginning to generate revenue at the plan level, but not yet visibly at the usage level. AI contract scale was approximately 6% of annualized subscriptions in Q2 and would be approximately 10% at year-end if the target is met; the target raised twice this year proves that customers are willing to pay for AI. But AI pricing remains tied to seats, revenue from excess usage has not been disclosed, and ServiceNow absorbs inference cost first. The bull thesis in the original research has been confirmed at the contract level, but not yet at the revenue and profit levels; neither v3.0's “decoupling from seats” nor the AI pricing research's “cost pass-through” has happened yet.
Third, 2026 is a year in which ServiceNow traded gross margin and its balance sheet for scale. Adjusted subscription gross margin fell 3 percentage points, approximately US$8.6 billion in net cash became approximately US$600 million in net debt, and operating cash flow for the first three quarters fell approximately 3%. I withdraw the original research judgments that “integration costs are only short term” and that “gross margin has room for structural improvement”: amortization will continue for several years, interest has become a fixed expense, and organic growth in the existing business has already fallen below 20%.
Going forward I will watch only three numbers. Q4 adjusted subscription gross margin of at least 80.5% (below this level, full-year guidance of 81% will be missed); actual year-over-year cRPO growth does not fall below 20% (the reported figure; Q4 guidance of 18.5% does not count); operating cash flow growth catches up with revenue growth (after adding back acquisition expenses, it was only 7% for the first three quarters). If all three are achieved, the original research's bull thesis also holds at the profit level, and Armis is an expensive but worthwhile acquisition; if gross margin guidance is missed, I will change my judgment to say that AI delivery costs have already exceeded the pricing power it brings. Full-year adjusted FCF margin guidance of 36% is not on this list because part of it comes from lower capital spending and expense add-backs, and it cannot represent returns on capital after the acquisition.
This article is updated through November 13, 2026. Financial data covers Q3 and the preceding three full fiscal years; share price valuation is not included. This article is based on public information for research and investment education purposes and does not constitute investment advice.
ProfitVision LAB|Independent Research
