Key takeaways
- Salesforce guided FY27 GAAP operating margin up to 20.9% and non-GAAP to 34.3%, from 20.1% and 34.1% in FY26.
- ICONIQ projects the aggregate gross margin on AI products at 52% for 2026, against 45% in 2025 and 41% in 2024. It is a cross-industry survey figure, not Salesforce's own.
- At the scaling stage, model inference is 23% of AI product cost — up from 20% pre-launch — while talent falls from 32% to 26%.
- Salesforce's Agentforce and Data 360 annual recurring revenue passed $2.9 billion, up over 200% year on year — but that includes $1.1 billion of acquired Informatica Cloud ARR.
- FY26 operating cash flow was $15.0 billion, up 15%, on revenue of $41.5 billion.
The margin cut that did not happen, and the margin squeeze that is real
Salesforce reported on 25 February 2026 under the headline "Salesforce Delivers Record Fourth Quarter Fiscal 2026 Results." Fourth-quarter revenue was $11.2 billion, up 12%. Full-year revenue was $41.5 billion, up 10%. FY26 GAAP operating margin was 20.1% and non-GAAP 34.1%. For FY27, the company initiated guidance of 20.9% GAAP and 34.3% non-GAAP operating margin — an expansion of 80 and 20 basis points respectively — alongside revenue guidance of $45.8–46.2 billion.
So the operating margin went up. The interesting question is how, because the cost pressure everybody expected is measurable and it is not imaginary.
ICONIQ projects 52% for 2026, against Salesforce's 82.7% on subscriptions
ICONIQ's January 2026 State of AI: Bi-Annual Snapshot surveyed 269 companies on gross margin for AI products. The aggregate averages: 41% in 2024, 45% in 2025, and a projected 52% in 2026. Split by strategy, application-layer innovators projected 45%, proprietary model developers 49%, and balanced differentiators 53%.
Two things that figure is not. It is not a reported outcome — 52% is the survey's 2026 projection, and the 2024 and 2025 readings are 41% and 45%. And it is not Salesforce's number: ICONIQ is surveying AI-native builders across industries, and Salesforce does not disclose a gross margin for its AI products at all.
Set it against Salesforce's own income statement rather than a rule of thumb. In FY26 the company booked $39,388 million of subscription and support revenue against $6,796 million of cost — an 82.7% gross margin on the subscription line, and 77.7% blended across everything including services. The marginal cost of the next seat is close to zero. An AI product answers each query by running a model, and the model consumes compute every single time. The chart plots the ICONIQ series — 41%, 45%, 52% — on one axis, in percent. The direction is up. The level is roughly 30 points below Salesforce's 82.7%.
Inference costs rise as products scale — the opposite of the SaaS cost curve
The same report breaks AI product cost by product stage, across 202 companies. Pre-launch, talent is 32% of cost and model inference 20%. By the scaling stage, talent has fallen to 26% and model inference has risen to 23%. Infrastructure and cloud rises from 16% to 17%.
That inversion is the whole argument. In traditional software, the dominant cost is engineering payroll, and it amortises across users: the more you sell, the less each incremental unit costs. In AI products, the dominant cost driver becomes inference, and inference scales with usage. Success does not dilute the cost line; it feeds it. NVIDIA's fiscal 2026 Data Center revenue of $193.7 billion is, viewed from the buyer's side, a cost of goods sold line spreading across the software industry.
How Salesforce expanded margin anyway
Salesforce is running two businesses at once. The legacy CRM franchise carries classic SaaS economics and is still 90%-plus of the $41.5 billion. The AI franchise — Agentforce and Data 360 — passed $2.9 billion of annual recurring revenue, up over 200% year on year.
That $2.9 billion is not all organic, and the company says so. The full sentence in the release reads: ARR "exceeds $2.9 billion, up over 200% Y/Y, including $1.1 billion Informatica Cloud ARR and $800 million Agentforce ARR, up 169% Y/Y." Informatica was acquired in November 2025. So $1.1 billion of the $2.9 billion was bought, not built, and the growth rate is flattered by the same transaction. Strip the acquired line out and organic AI-and-data ARR is roughly $1.8 billion — about 4.3% of FY26 revenue, not the 7% the headline number implies.
At that organic 4.3% mix, an AI gross margin of 52% instead of 82.7% costs the blended gross margin about 1.3 percentage points. Counting the acquired ARR too, on the full $2.9 billion, the drag is about 2.1 points. Either way it is inside the range that operating discipline elsewhere — headcount, real estate, sales efficiency — can absorb, and FY26 operating cash flow of $15.0 billion (up 15%) says it was absorbed. The squeeze is real and it is currently second-order.
The arithmetic changes if the AI line grows into the mix. At 25% of revenue on 52% gross margin, the blended drag becomes roughly 8 percentage points of gross margin — a level that operating expense cuts cannot quietly offset.
What would falsify this
The ICONIQ figures are survey data from AI-native builders, not audited financials from mature software companies, and 52% is a projection for 2026 rather than a reported outcome. If inference cost per token keeps falling faster than usage grows, the 52% converges upward and the mix argument dissolves. The report itself frames margin leadership as a function of "model selection, routing strategies, and infrastructure efficiency" — all of which can move fast.
Nor does Salesforce disclose a gross margin for Agentforce separately. The 1.3-point, 2.1-point and 8-point estimates above are constructed from a disclosed ARR figure and an external survey average; they are an order-of-magnitude check, not the company's own accounting.
The line item that will settle it
Watch cost of revenue as a share of revenue, not operating margin. Operating margin can be defended for several years by cutting sales and marketing; gross margin cannot. Salesforce's FY27 guidance implies operating margin expansion of 80 basis points — if that arrives while gross margin compresses, the company is funding AI COGS out of its go-to-market budget, and there is a floor under how long that works. and .
This content is for informational purposes only and does not constitute financial advice. Always do your own research or consult a qualified financial adviser before making investment decisions.