What Do Companies Actually Mean by ‘AI Revenue’?

The signal

Commercial momentum is observable; a single, comparable AI revenue league table is not supported by this evidence.

One label, four different clocks

An investor needs to know when a customer commits, when the product becomes available, when the service is consumed and when the seller recognizes revenue. Those events can occur in different periods. A year-end recurring-value snapshot can rise before its full effect reaches the income statement. A multiyear backlog can grow because customers sign longer contracts, even when near-term consumption changes much less.

Our evidence panel contains four issuers and seven earnings documents available by 13 November 2026 UTC. It is a deliberately bounded comparison of disclosed measures, not a completed survey of thirty software companies. The periods also differ: Salesforce’s fiscal second quarter ended in July, Adobe’s third quarter in August, and Microsoft’s fiscal first quarter in September. They are recent observations, not synchronized measurements.

The distinction matters most when a headline presents an exceptional percentage. Before asking whether 150% is faster than 70%, establish whether the figures refer to a reported annual growth rate, a revised management target, or a change in the products counted. A percentage does not repair an incompatible denominator.

Salesforce: the series changed as the platform expanded

Salesforce’s first-quarter fiscal 2027 release identified $1.2 billion of Agentforce annual recurring revenue and nearly $3.4 billion for Agentforce and Data 360 combined. The combined figure included $1.1 billion of Informatica Cloud ARR. The company’s reporting therefore already distinguished a narrower agent product measure from a larger collection that included an acquired business. ^[1]

In its next release, Agentforce ARR exceeded $1.5 billion. Salesforce also stated that, effective that quarter, the measure included Slackbot and Headless 360. A simple comparison with the earlier $1.2 billion would mix business development with an expanded reporting perimeter. The release does not provide the component bridge needed to isolate those effects. ^[2]

This is useful disclosure: the company tells readers that the boundary moved. The analytical error would be to acknowledge that footnote and then calculate an unqualified same-product growth rate anyway. We cannot infer how much of the increase came from newly included products, existing products, pricing or customers. Equally, we cannot assume the definition change explains all of it. The correct output is a larger reported measure with an unresolved like-for-like bridge.

Investors should preserve both versions in their model. Replacing the old definition without a record makes the chart look continuous while erasing the very change that needs explaining.

ServiceNow: achieved contract value and a future target

ServiceNow reported that its AI annual contract value crossed $1 billion in the second quarter of 2026. That release also reported quarterly subscription revenue of $3.877 billion. These are distinct measures: an AI contract-value milestone and revenue recognized across the subscription business. ^[3]

The third-quarter release raised the company’s 2026 AI ACV target to $1.7 billion. It reported subscription revenue of $3.988 billion and current remaining performance obligations of $13.65 billion. The latter refers to contract revenue expected to be recognized within twelve months; it is broader than AI and remains different from revenue already earned. ^[4]

There are two legitimate conclusions. The achieved milestone supports the existence of substantial commercial demand. The higher target indicates management’s expectation of further progress. It does not establish that $1.7 billion had already been achieved, nor that the difference from $1 billion is revenue earned in the intervening quarter.

Our calculation from the reported subscription figures is a 2.86% sequential increase. That measures the whole subscription business between those two quarters. It cannot be assigned to AI, and seasonal patterns mean it should not be annualized mechanically. The calculation is modest but reproducible; an AI contribution estimate would require information this pair of releases does not provide.

Adobe: impressive growth without a complete current base

Adobe’s June release said AI-first ARR had tripled year over year and exceeded $500 million. It separately reported total ARR of $27.10 billion, including approximately $480 million from Semrush. In September, it reported AI-first ARR growth above 150% year over year, alongside company revenue of $6.76 billion. ^[5][6]

The two growth descriptions do not establish a sequential decline in AI recurring value. They compare each quarter with its own prior-year base. A business can grow between consecutive quarters while its annual percentage slows because the earlier comparison was easier. Without the absolute current AI-first ARR value and a consistent scope reconciliation, the releases cannot support a precise sequential bridge.

Nor should the Semrush contribution be casually assigned to AI-first ARR. The disclosed acquisition figure belongs to total ARR. Moving it into a narrower product category would add an assumption that the source does not make.

For valuation, this distinction determines what can enter the model as an observed input. The growth disclosure is evidence of expansion. A precise forecast of AI’s share of next year’s revenue needs additional assumptions about the starting base, recognition timing and customer retention. Those assumptions should be visible rather than disguised as company facts.

Microsoft: scale does not make a broad measure AI-specific

Microsoft reported $63.9 billion of Microsoft Cloud revenue in its first fiscal quarter of 2027 and $677 billion of commercial remaining performance obligations. Azure revenue growth was 48%. These measures establish the scale and momentum of broader cloud and commercial activity. The release does not make them a stand-alone total for AI revenue. ^[7]

A common temptation is to take a broad business whose growth is associated with AI and label its entire expansion as AI monetization. That loses the distinction between an explanatory narrative and a disclosed segment. It also risks counting the same economic chain several times: a platform’s customer commitment, a cloud provider’s recognized service revenue and an application vendor’s sales can all relate to different stages of the same customer budget.

The better use of Microsoft’s figures is as evidence of a large commercial base and contracted future activity. To turn them into an AI-specific forecast requires a separate allocation model. That model may be useful, but it is DN’s or the investor’s estimate, not a number Microsoft reported in this release.

The bullish case survives stricter definitions

Demand does not have to fit a perfect reporting taxonomy to be economically valuable. A customer may purchase a broader bundle because an agent feature makes the entire workflow more useful. Revenue can therefore benefit even if the seller does not assign a separate price to the AI component. Acquisitions can also bring complementary data or distribution that improves the legacy product.

This is the strongest countercase to an excessively skeptical reading of the disclosures. Requiring every benefit to appear as a separately invoiced AI line would miss genuine business improvement. The solution is to test that broader improvement using comparable company revenue, retention, margin and cash measures, while keeping the claimed AI attribution appropriately limited.

The skeptical case is narrower and more practical: a larger product perimeter can inflate an apparent growth comparison; a target can be mistaken for an achievement; a usage counter can expand without comparable monetization. None of those possibilities can be settled by enthusiasm or suspicion. They need bridges.

A growth model with an explicit unknown

The accompanying calculator illustrates a definition bridge. In a hypothetical example, a recurring-value measure increases from 100 to 160. Suppose 20 of the increase is a disclosed scope addition and 10 is a separately disclosed acquisition effect, with no overlap. The comparable current measure is 130, giving 30% comparable growth rather than the 60% headline increase. These inputs are illustrative and do not describe any company in the panel.

Do not use this simple bridge unless the adjustments are non-overlapping, the excluded businesses are absent from the prior-period base and both periods use the same currency basis. The calculator checks numerical bounds; it cannot verify product scope or detect overlapping adjustments. Any unexplained component must remain unresolved rather than being attributed automatically to AI or organic growth.

For the next earnings season, the decisive evidence is straightforward: a restated prior-period product definition, the absolute base behind a percentage, a clear target-versus-actual label and a reconciliation to the financial statements. A company that supplies those bridges allows investors to assign more confidence to its growth story without needing a more exciting headline.

The evidence, side by side

AI Revenue Definition Bridge

Illustrative definition bridge. Defaults are hypothetical, not company results.

Numerical checks: prior>0; non-negative inputs; exclusions<=current. User preconditions, not automatically verified: same metric/currency/period; non-overlapping adjustments; excluded business absent from prior. Otherwise reconcile both periods before using this bridge.

Scenarios to watch

Comparable acceleration

Conditional interpretation

Stable definitions, absolute values and comparable recognized revenue strengthen together. Confidence in sustained monetization rises.

Perimeter-led acceleration

Conditional interpretation

A disclosed scope bridge explains much of the larger measure. Keep the acquired or expanded business, but lower the inferred legacy growth rate.

Insufficient bridge

Observed analytical limit

Growth is reported but the necessary components are absent. Preserve a range or unknown instead of manufacturing a point estimate.

Method and limitations

DN examined seven primary earnings documents from four issuers, with reporting dates from May to October 2026 and a research cutoff of 13 November 2026 UTC. We classified each observation by metric, fiscal period, product perimeter and actual-versus-guidance status. The sample was selected to test comparability, not estimate the size or average growth of the AI market. The only company calculation in the narrative divides ServiceNow Q3 subscription revenue of $3,988m by Q2’s $3,877m, subtracting one: 2.863%. It is reported-currency, whole-business sequential growth. Source references identify the exact underlying documents.

The panel cannot establish AI market share, incremental AI profit, customer-level return or causal attribution. ARR and ACV definitions are company-specific. We did not reconstruct undisclosed product revenue or turn targets into achieved results. Approximate values and lower bounds remain approximate. A missing reconciliation describes the limits of these releases, not necessarily the limits of every disclosure the issuer has ever made.

Primary sources and evidence register

Source dates and reporting periods differ. Several documents from one issuer are one source family, not independent confirmations. The evidence download contains the observations used in this article.

Frequently asked questions

Can ARR be divided by four to estimate quarterly AI revenue?

Not reliably. ARR is a snapshot under a company’s definition; recognized revenue depends on contract timing, delivery, usage and the mix of products. A separate reconciliation is needed.

Does a definition change mean the company overstated growth?

No. It means the comparison needs an adjustment. The changed products may create real value; the unadjusted growth rate simply cannot establish same-scope performance.

Did Adobe’s slower annual AI growth mean its AI business shrank?

The two cited annual growth descriptions do not establish a sequential decline. Their prior-year comparison bases differ and the later absolute value is not supplied in the cited headline measure.

What should investors ask for next?

A stable metric definition or restated history, absolute values, an acquisition and currency bridge, and a clear distinction between achieved commercial measures and management targets.