Microsoft: where is AI’s profit?

Microsoft reported 48% Azure revenue growth on 28 October for the quarter ended 30 September. Across the group, operating cash flow less cash purchases of property and equipment fell 27.39% year on year. Cloud sales growth and funding the infrastructure raise different questions, each requiring its own business scope when assessing the industrial bet on AI. R04

For the year ended 30 June 2026, Microsoft reported 41% growth in “Azure and other cloud services”. In the presentation of its new metrics, Azure growth for that same year comes to 40%. Microsoft 365 commercial cloud moves from 17% to 18%. The periods are identical; the composition of the businesses being measured has changed. That one-point shift opens a much larger question: what can we actually learn about AI profitability when the categories in the accounts evolve? M02

The change makes new information available. Microsoft now publishes quarterly Azure revenue histories in dollars and sets out the migration of several businesses. At the same time, operating results are brought together in two broad segments. Visibility becomes more detailed for certain sales and more concentrated for profits. The reader needs to follow both movements together. M01 M02

The investigation reveals a second shift, less visible in the table headings. The estimated useful life of certain buildings can change the classification of future leases, and then the perimeter of reported CapEx. An announced investment figure therefore depends on the industrial programme, the timing of assets and the metric’s definition. The same infrastructure passes through several accounting representations. F03

Finally, the OpenAI partnership reaches several layers of the financial statements: commercial revenue, receivables at the reporting date, and investment gains or losses. These relationships shed light on AI’s economic role at Microsoft, provided their units, dates and counterparties remain attached. M03

Our guiding principle is to connect each figure to the decision it can inform. Sales reveal monetisation; operating income describes a portfolio and its allocated costs; cash flows show how capacity is financed; commitments establish future horizons. Measuring AI’s own profitability would require connecting these pieces within a common economic perimeter.

This investigation has a cutoff date of 28 October 2026 and includes the FY2027 Q1 results released this evening, for the quarter ended 30 September. Annual series cover fiscal years ended 30 June. Recast historical figures, management statements, l0g calculations and the two hypothetical examples retain their respective status. July commentary retains its original date; tonight’s quarterly release supplies the latest reported results. R04

A new map for an integrated company

The formal announcement dates from 2 September 2026. Microsoft publishes a regulatory document accompanied by a presentation of its FY2027 segments and metrics. The shareholder letter dated 14 October then sets out the strategic rationale. FY2027 is the fiscal year that began on 1 July 2026; its first quarter therefore covers July to September. These dates distinguish the start of the economic period, the reporting announcement and its presentation to shareholders. M01 M02 M04

The three previous segments were Productivity and Business Processes, Intelligent Cloud and More Personal Computing. The new ones are Agents and Infra and Devices and Consumer. The first brings together a wide range of applications, agents, cloud services and licences; the second includes Windows OEM and devices, Xbox, search and advertising. Management presents this organisation as reflecting increasingly integrated businesses and resource allocation across several technological layers. M02

That logic can be seen in the paths products take. GitHub cloud, other developer cloud services and Security Copilot move into Microsoft 365 commercial cloud, having previously sat within Azure and other cloud services. Healthcare and life-sciences cloud services move into Industry solutions cloud. The new Azure is more closely focused on consumption-based infrastructure and platform services, together with virtual desktop. M02

LinkedIn offers a particularly clear example of why businesses need to be followed beyond their brand. Talent Solutions and Sales Solutions move into Industry solutions cloud, within Agents and Infra. Marketing Solutions and Premium Subscriptions move into Search and advertising, within Devices and Consumer. A single acquired company is thus split between the two segments, according to the activities grouped under the new organisation. M02

Another instructive detail: Microsoft 365 Consumer remains in Agents and Infra. The second segment’s name is therefore insufficient to place every consumer-facing business within it. The published definitions are more precise than the intuition suggested by the labels. They explain which customers, products and sales models enter each total. M02

Management explains this integration through changes in usage and billing. In its July remarks, it describes the development of plans combining seats and consumption, particularly around Copilot and agents. Infrastructure, models and applications then contribute to the same service sold to a customer. This strategic explanation helps make sense of the grouping; the profitability of each layer remains a separate measurement question. F03 M04

The nature of the documents also matters. The annual financial statements in the 10-K use the former three-segment presentation. The September presentation supplies recast historical figures as part of a regulatory communication to investors. It is a precise source for reconstructing comparisons, with the status and limitations described in the source guide. M01 M02 M03

The same year, two allocations of earnings

The first check is particularly straightforward: add the columns. In both FY2026 presentations, Microsoft retains $331.839 billion in revenue and $155.237 billion in operating income. Moving the businesses redistributes those totals between segments. This invariance provides a firm anchor for following the change in the map. M02

Under the former structure, Intelligent Cloud reported $137.791 billion in revenue and $56.972 billion in operating income. Under the new one, Agents and Infra reports $268.127 billion and $136.365 billion. Their calculated margins are 41.35% and 50.86%, respectively. They describe two different portfolios in the same year. Measuring an improvement requires comparing each portfolio with its own recast history. M02

Microsoft supplies that basis. Recast FY2025 revenue for Agents and Infra is $218.783 billion, with operating income of $110.814 billion. We can therefore examine the new grouping’s annual development while preserving its definition. This bridge provides useful information: it allows readers to distinguish portfolio growth from the effect of changing its composition. M02

The weight of the new grouping naturally attracts attention. Calculated for FY2026, Agents and Infra accounts for approximately 80.80% of consolidated revenue and 87.84% of operating income. Those shares measure the concentration of the accounts in a grouping that includes mature and new businesses, subscriptions, licences and infrastructure. Any interpretation needs to preserve that breadth. M02

Growth rates make the effect of the perimeter more tangible. For FY2026, the shift from the former Azure and other cloud services to the new Azure changes reported growth from 41% to 40%. Microsoft 365 commercial cloud growth moves from 17% to 18%. Commercial seat growth moves from 6% to 7%, with paid GitHub seats included in the metric. All these rates cover the same historical periods; the populations being compared have been redefined. M02

The new Azure also has dollar figures: $72.610 billion in FY2025 revenue, followed by $101.938 billion in FY2026. Their ratio gives calculated growth of 40.39%, consistent with the reported 40% after rounding. Readers can now examine dollars and percentages under this precise definition. The improvement in transparency is real and directly usable. M02

The release published on 28 October 2026 provides an initial concrete example. For the quarter ended 30 September, Azure reports $33.165 billion in revenue, against $22.384 billion in the recast comparative period, representing reported growth of 48%. The available operating income figure remains that of Agents and Infra: $40.558 billion, on $76.795 billion in revenue. The difference in granularity is directly visible in the tables: a sales line specific to Azure, and earnings covering the entire portfolio. R04

The same discipline applies to guidance. The September presentation mechanically adjusts the first-quarter outlook to the new categories while preserving the company total. Guidance rewritten in the new nomenclature should be compared with results prepared in that nomenclature. Interpreting a surprise therefore requires keeping the guidance date, its currency basis and the line’s definition. M02

The presentation gives the aggregate figures for both systems, while leaving some detailed allocations outside the table. Subtracting More Personal Computing from Devices and Consumer produces a net difference; attributing that entire difference to the transferred LinkedIn businesses would require an additional bridge for revenue and costs. The arrows in our map therefore describe changes in perimeter, using a uniform width. M02

A segment margin answers a management question

Why are the columns shaped this way? The 10-K explains that the chief operating decision maker, who is also the chief executive officer, uses operating income to assess segments and allocate resources. Reporting thus follows the organisation of internal management. US segment-reporting rules require, among other things, disclosure about the profit measures used and significant expenses provided to that decision maker. M03 R01

Understanding the reported result then means looking at how costs reach it. Note 18 of the 10-K for FY2026 describes direct charges and allocations. Some costs of revenue are allocated according to relative revenue. Shared operating expenses for AI infrastructure and training, along with certain marketing expenses, are generally allocated according to the segments’ relative gross margin amounts. Other central functions use their own allocation bases, often gross margin or headcount. M03

These rules serve a management purpose: distributing costs that benefit several businesses and making managers’ results comparable under a defined convention. Their economic interpretation nevertheless depends on the question being asked. A cost charged to a segment tells us about the allocation chosen by the group. Measuring the incremental cost of a request, a customer or a training run would require a different observation.

A hypothetical example makes the distinction concrete. Three businesses generate 60, 30 and 10 units of gross profit, respectively. The group allocates a shared expense of 12 units in proportion to those amounts. The allocated charges are 7.2, 3.6 and 1.2. The total remains 12. The allocation rule explains why a business receives a share of the cost; usage data would be needed to connect that share to its actual consumption of infrastructure.

The problem becomes more demanding when calculating a return on invested capital. Note 18 states that assets are not allocated to segments in internal reporting. It also explains that some depreciation is included in overhead allocations, with separate identification by segment considered impracticable. The published document therefore provides segment results without all the denominators needed for a return on capital specific to Azure or AI. These are the policies described for FY2026. M03

Analysis can still make progress. It can measure a portfolio’s margin within the announced perimeter, observe its costs and track the group’s investment. Isolating incremental AI profit would require knowing the additional revenue, associated costs, share of common capacity used and usage that would otherwise have occurred. A Copilot sale may supplement a subscription, shift spending or change consumption of a product already sold. Each situation calls for a different calculation.

This distinction also protects comparisons between companies. Two groups may build different segments around a cloud business or an assistant. Their margins then reflect different portfolios, assets and cost conventions. Comparison becomes more useful when it starts with those definitions and then examines trends within each stable perimeter.

Commercial indicators measure different scopes

Microsoft publishes several signs of AI monetisation. One of the most widely repeated appears on 29 April 2026: Satya Nadella announces that the AI business has exceeded a $37 billion annual revenue run rate, up 123% year over year. The release uses the wording annual revenue run rate. The documents reviewed do not detail the business’s full composition, the annualisation formula or its reconciliation to revenue for an accounting quarter. E04

The metric describes a commercial pace announced at a particular date. Recognised annual revenue, by contrast, adds together services attributable to a fiscal year. The distinction is particularly important for a rapidly growing business. The pace reached at the end of a period may be well above the average pace of the preceding months. In this dossier, the $37 billion therefore remains attached to its April announcement and published definition.

Paid seats provide another measure. In July, management says Microsoft 365 Copilot exceeds 30 million paid seats. This stock measures commercial adoption. Deriving revenue from it would require the actual negotiated prices, plans sold, effective dates, discounts and activation periods. Multiplying it by a list price would produce a billing scenario whose assumptions would need to be stated. F03

The Microsoft Cloud perimeter offers a third perspective. For FY2026, under the definitions used in the annual publications, it accounts for $214.4 billion in revenue and a 66% gross margin. This aggregate brings together several commercial businesses, including Microsoft 365 Commercial cloud, Azure and other cloud services, commercial LinkedIn and Dynamics 365. It is a portfolio spanning several businesses, with different costs of revenue, products and billing models. M03

In its fourth-quarter remarks, management attributes pressure on the Cloud margin to the mix shift towards Azure, investment in AI infrastructure and increased product usage, partly offset by efficiency gains. It also describes improvements in capacity and delivery that allowed more available resources to be monetised. These explanations identify mechanisms to track; they remain management attributions, without a complete public breakdown of the cost of each use. F03

The transition from software largely billed per seat to services combining seats and consumption makes the question more interesting. A customer’s revenue may increase with usage; the resources needed to serve that customer may also increase. Margin then depends on pricing, model efficiency, machine utilisation and the mix of tasks. Commercial data illuminate part of this chain. Reaching earnings requires following it through to costs and assets.

Investment absorbs more operating cash

The consolidated accounts provide a consistent series across three fiscal years. Cash flow from operations rises from $118.548 billion in FY2024 to $136.162 billion in FY2025, then $182.935 billion in FY2026. Over the same period, cash purchases of property and equipment ; buildings, equipment and other tangible fixed assets ; rise from $44.477 billion to $64.551 billion, then $115.948 billion. The periods are annual and the amounts cover Microsoft as a whole. M03

Subtracting the second line from the first produces a different movement from operating cash flow: $74.071 billion, then $71.611 billion, then $66.987 billion. Between FY2025 and FY2026, operating cash flow rises 34.35%, cash purchases of fixed assets rise 79.62%, and the difference falls 6.46%. These percentages are our calculations from the published statements. They show that strong growth in net operating cash flow can accompany investment that absorbs an increasing share of those flows. M03

Data and calculations

Difference = operating cash flow − cash PPE purchases. Finance-lease principal and other investing flows remain outside this formula.

Microsoft consolidated · US$ billions · years ended 30 June

Fiscal year	Operating cash flow	Cash PPE purchases	Difference
FY2024	118.548	44.477	74.071
FY2025	136.162	64.551	71.611
FY2026	182.935	115.948	66.987

This evening’s release extends the observation. For July to September 2026, operating cash flow reaches $58.000 billion and cash PPE purchases $39.367 billion. The difference is $18.633 billion, against $25.663 billion a year earlier, a calculated decline of 27.39%. These quarterly amounts enter the dossier with their own period; the three fiscal years in the figure remain a consistent annual series. R04

This definition matches the numerical basis of the free cash flow management discusses for the fourth quarter: $55.441 billion in operating cash flow, less $35.802 billion in cash purchases of fixed assets, gives $19.639 billion, rounded to $19.6 billion on the call. The reconciliation allows readers to reproduce the figure and understand its construction. F05 F03

The SEC notes that free cash flow has several possible definitions and calls for an explicit calculation reconciled to the accounts. It also points out that certain mandatory cash payments may remain outside this measure. The word “free” therefore calls for examination of the formula and other commitments. R02

At Microsoft, two lines deserve this attention. Principal payments on finance leases reach $3.101 billion in FY2026 and appear within financing cash flows. The Other, net line in investing cash flows represents a $19.861 billion net outflow. The annual commentary primarily links the increase in outflows in this latter category to facilitating component purchases; it provides this explanation without isolating a consistently defined amount of GPU prepayments. These lines should be read alongside operating cash flow less cash PPE purchases. M03

Operating cash flow itself has several drivers. The report mentions customer collections and lower cash taxes, among others. Cash taxes net of refunds fall from a rounded $28.7 billion in FY2025 to $21.188 billion in FY2026. Changes in receivables, payables, unearned revenue and taxes contribute to the conversion of earnings into cash. Attributing the entire increase in operating cash flow to AI profits would therefore go beyond what the accounts measure. M03

The useful finding concerns the industrial expansion: Microsoft generates substantial operating cash flow and devotes an increasing share of it to assets. The return on this construction will then depend on the services sold, their prices, utilisation and the spending required to operate them. The income statement and cash-flow statement describe different stages of that trajectory.

Purchases, service and expenses follow different timetables

Infrastructure spending can produce several entries at different times. A commitment establishes a contractual obligation. Delivery or the start of a lease results in recognition of an asset under the applicable rules. Payment converts a liability or commitment into a cash outflow. Depreciation then allocates the asset’s cost over its estimated useful life. Microsoft’s policies and cash-flow tables allow these differences to be followed. M03

The following example is entirely hypothetical. A new piece of equipment costs 120 units, paid upfront. Its residual value is zero and it is used for full years. Under straight-line depreciation over four years, the annual expense is 30. Over six years, it is 20. The initial payment remains 120 in both cases; the expense follows a different timetable. At the end of the fourth year, the net carrying amount is zero on the first path and 40 on the second.

Data and calculations

Cost 120 paid upfront, zero residual value, complete service years. Two schedules chosen at acquisition. Annual charge = 120 / useful life.

Fictional example · illustrative monetary units

Year	Expense / 4 years	Expense / 6 years	Net value / 4 years	Net value / 6 years
0	0	0	120	120
1	30	20	90	100
2	30	20	60	80
3	30	20	30	60
4	30	20	0	40
5	0	20	0	20
6	0	20	0	0

This mechanism has a precisely documented precedent at Microsoft. In July 2022, the group reassessed the useful life of servers and network equipment, from four to six years, effective from FY2023. For that fiscal year, it disclosed an increase of $3.7 billion in operating income and $3.0 billion in net income, based on the fleet’s net carrying amount at 30 June 2022. Our example compares two theoretical schedules at acquisition; Microsoft’s actual change applied prospectively to an existing fleet. F04

Economic interpretation then requires a second question: how long will the asset remain economically productive? Accounting useful life is an estimate of that use. The ability to serve customers also depends on reliability, energy consumption, required performance and the possibility of redeploying machines to other tasks. These parameters must be observed to assess the estimate. A longer life spreads the cost; the economic return depends on the services actually produced during that life.

The FY2026 report gives a range of two to six years for servers and network equipment. Depreciation expense on property and equipment reaches $34.3 billion that year. The cash-flow statement shows a broader line, Depreciation, amortization, and other, at $38.534 billion. For tracking physical assets, the PPE note and that reconciliation line therefore have different perimeters. M03

This separation becomes important when assets arrive rapidly. A fleet may receive substantial capacity during the year while its annual depreciation expense has yet to reach a full year of service. Conversely, revenue may depend on the speed of commissioning and customer adoption. Evaluating an investment cohort requires connecting these timetables: payment, availability, utilisation, revenue and cost. An annual CapEx figure alone leaves those stages unresolved.

Leases can change columns

On 29 July 2026, Amy Hood announces a change affecting another asset category: datacenters and office buildings. Their estimated useful life moves from 15 to 25 years, effective at the start of FY2027. She expects a small positive effect on operating income for that year and describes a larger effect on the classification of future leases. F03

Under this explanation, more future datacenter leases will be classified as operating leases. Microsoft includes finance leases in the CapEx it communicates and excludes operating leases. In July’s guidance, this change leads to approximately $175 billion in CapEx for calendar 2026, with other investment expectations unchanged in that commentary. The amount retains its announcement date, prospective status and calendar-year basis. F03

The mechanism deserves to be followed through the accounts. At a lease’s commencement date, the contract results in recognition of a right-of-use asset and a liability under the applicable rules. Lease classification then determines several aspects of presentation, including expenses and payments. For Microsoft’s communicated CapEx measure, it also determines inclusion in the metric. Management speaks of more future contracts; the document leaves the conditions for classification to the analysis of each contract. F03 M03

Operating leases do appear on the balance sheet. At 30 June 2026, Microsoft reports $24.177 billion in related right-of-use assets and $21.925 billion in liabilities. Excluding this category from communicated CapEx therefore concerns an investment measure, while the financial statements continue to present the associated rights and obligations. M03

The cash-flow statement provides another distinction. Operating lease payments and interest on finance leases enter operating cash flows; finance lease principal appears within financing. A reading of cash therefore needs to consider the categories together. Communicated CapEx, cash PPE purchases, recognised right-of-use assets and lease payments answer four different questions. M03

The useful-life change announced in July concerns the buildings identified by management. The earlier four-to-six-year change concerns servers and network equipment from FY2023. Bringing them together helps explain the role of estimates while preserving the identity of the assets and the timing of each decision. F03 F04

The analytical task is therefore precise: follow committed physical capacity and its future cost through successive definitions. A revision to announced CapEx may reflect asset quantities, prices, timing or classification. The July documents explicitly identify this last mechanism. Measuring an acceleration or slowdown in the industrial effort requires reconstructing a comparison that preserves it.

OpenAI reaches several layers of the accounts

The same partner can be a customer, a revenue-sharing counterparty and an investee. The 10-K for FY2026 makes these relationships particularly visible for OpenAI. Commercial arrangements, including revenue-sharing payments, generate $24.1 billion in revenue during the fiscal year. The document provides this category of arrangements; a complete breakdown among the different services remains to be established. M03

At the 30 June reporting date, OpenAI-related receivables reach $6.0 billion. This amount measures what remains receivable at a date, under the applicable recognition rules. Reconstructing the year’s cash collections would also require the opening balance and movements during the period. Adding annual revenue to the closing receivable would combine two measurement bases. M03

The investment forms a third circuit. Microsoft applies the equity method using an approach that reflects the economic rights and priorities attached to its interest. For FY2026, OpenAI investments produce an effect of $6.530 billion before tax, or $4.963 billion after tax. The report primarily links the annual gains to dilution from the recapitalisation. The effect appears in Other income (expense), net, below operating income. M03

The movement between fiscal years is revealing. The after-tax OpenAI effect was negative $3.620 billion in FY2025; it becomes positive $4.963 billion in FY2026. The swing therefore amounts to $8.583 billion. Over the same period, GAAP net income rises $31.917 billion. Microsoft’s reconciliation allows another component of $23.334 billion to be isolated, corresponding to the change in net income excluding this OpenAI effect. This is an arithmetic decomposition, with each component retaining its own perimeter. M03

The word “adjusted” also needs its definition alongside it. Microsoft’s adjustment removes the effects of its OpenAI investments. The $3.2 billion gain on the Anthropic investment, reported in the fourth quarter, remains included in adjusted other income and expense before tax. Adjusted net income therefore retains the effect of that gain after tax. A reader seeking an operating measure therefore needs to continue reconciling the components beyond the label. F03 F05

The commercial relationship is also evolving. The October 2025 announcement included an additional $250 billion in Azure services purchases by OpenAI. The amendment of 27 April 2026 provides greater scope for distribution on other clouds and a non-exclusive Microsoft licence to models and products through 2032. Revenue-sharing payments from OpenAI to Microsoft are to continue through 2030, subject to a cap whose amount is not disclosed in that announcement. E07 E06

These relationships create a common object of analysis: the customer’s financial strength, the investment’s value, the services provided and the terms of payment. Their simultaneous presence calls for reading both contracts and financial statements. Quantifying their effect on a future margin would require, among other things, actual delivered volumes, prices, associated costs and the agreements’ detailed economic terms.

Commitments establish a horizon for analysis

At the end of FY2026, commercial remaining performance obligations, or RPO, reach $678 billion. Management gives a weighted-average duration of approximately 2.3 years and expects approximately 30% to be recognised as revenue over the following twelve months. These measures describe a backlog of contractual revenue yet to be recognised, with an expected schedule. The service’s price, cost and cash collection remain separate dimensions. F03

Backlog concentration must remain attached to its date. At the second-quarter release, management had said OpenAI represented approximately 45% of commercial RPO at the end of December 2025. July’s commentary provides other information: 25% growth in RPO excluding OpenAI, and sequential backlog growth coming from customers outside frontier-model companies. The December percentage and June commentary describe two different snapshots. E05 F03

Infrastructure has its own horizons. At 30 June 2026, leases not yet commenced total $329.1 billion, primarily for datacenters, with expected starts between FY2027 and FY2033 and certain contractual conditions. The lease commitments table, including imputed interest, reaches $443.506 billion and already includes those future contracts. The amount for leases not yet commenced is a component of this perimeter. M03

These commitments differ from discounted liabilities already recognised on the balance sheet and from a single year’s payments. They indicate the period over which capacity will need to find uses and customers. Assessing that trajectory requires connecting consistent schedules: availability of buildings, arrival of equipment, customer commitments, revenue recognition and payments. A crude comparison between a commercial backlog, annual spending and a multiyear lease total would erase those distinctions.

The balance sheet contains another useful reference point. PPE purchases remaining in accounts payable reach $26.7 billion at the reporting date. This stock reflects part of the gap between acquisition and payment. Reconciling it with annual cash purchases would require the opening balance, settlements and other relevant movements. Capacity growth can thus be followed across several complementary financial statements. M03

Connect the accounts to the economic question

The new tables make the revenue of several businesses easier to read. Recast historical figures provide a basis for tracking growth in the portfolio that brings them together. The annual statements describe allocated costs, cash flows and commitments. Combining them supports substantial analysis of Microsoft’s economics while leaving the measurement of AI’s own profit open.

The first requirement is to preserve comparable series. A growth rate must connect the same product, population and exchange-rate convention. A margin must connect revenue with expenses within the same perimeter. The September presentation provides tools for doing so; a subsequent quarterly release should be connected to those historical figures, with its own definitions and commentary.

The second is to connect investment with entry into service. Orders and future lease payments indicate contractual intent; available assets indicate capacity; billed usage indicates monetisation. For assessing returns, the most useful information concerns how a capacity cohort moves from one stage to the next, together with the corresponding costs and revenue.

The third is to track the conventions that change how the accounts are read: allocation of shared expenses, useful-life estimates, lease classification and the definition of adjusted earnings. Each serves an accounting or management purpose. Making them explicit allows us to identify which part of a change describes the business and which part describes the measure chosen to present it.

Finally, readers need to be able to examine relationships that cross several tables. An important partner may generate revenue, a receivable and an investment result. A contract may enter a backlog before becoming a service and then a cash receipt. These links are precisely where a hurried reading loses the economic mechanism.

The dossier therefore allows a more demanding question to be put to every new column: which service produces this revenue, which costs and assets are allocated to it, and when does the whole arrangement turn into cash? The available answers already describe a highly profitable group engaged in a rapid build-out of capacity. The lasting profitability of that construction will be judged by the continuity between contracts, usage, expenses and payments.

Continue reading: circular AI financing, Citi’s GPU facility for Nebius and control rights at Anthropic and OpenAI.

Key terms: capex, free cash flow, RPO, finance lease under US GAAP and operating lease under US GAAP.

Sources

The 15 documents retain their dates, status and scope. Fictional examples and l0g calculations are identified in the text and figures.

1. M01 · Microsoft / SEC EDGAR · Published : 2026-09-02 · Document dated : 2026-09-02 · Accessed: 2026-10-28Microsoft Corporation : Form 8-K, September 2, 2026
   Passages and scope

   Item 7.01, “Regulation FD Disclosure”; Item 9.01, Exhibit 99.1.

   The filing dates the announcement of the two new segments and links to the recast presentation. It establishes the regulatory starting point for the reporting change.

   The presentation is furnished under Regulation FD, with its legal status specified in the 8-K. The announcement does not report a new realised quarter.

2. M02 · Microsoft / SEC EDGAR · Published : 2026-09-02 · Document dated : 2026-09 · Accessed: 2026-10-28FY27 Segments and Investor Metrics : September 2026
   Passages and scope

   Slides 7–9: business migrations; 11–13: definitions; 15–16: segments, revenue, costs and operating income; 17: offerings; 18: growth rates; 20–21: adjusted outlook.

   The central document for tracing businesses between the old and new columns. It supplies two years of recast history, metric definitions and the mechanical adaptation of Q1 guidance.

   Historical accounts are reclassified; guidance remains prospective. The document supplies neither a complete dollar bridge by migrated business nor standalone Azure or AI operating income. Rounding may cause totals not to add.

3. M03 · Microsoft / SEC EDGAR · Published : 2026-07-29 · Document dated : 2026-07-29 · Accessed: 2026-10-28Microsoft Corporation : Form 10-K for the fiscal year ended June 30, 2026
   Passages and scope

   Part II, Items 7–8: Financial Summary, non-GAAP reconciliation, Liquidity and Capital Resources and financial statements. Notes 1: revenue, PPE and Equity Method Investments; 3: other income/expense; 6: PPE; 11: taxes; 12: RPO/unearned revenue; 13: leases; 18: segments and allocations, printed pp. 83–85.

   The foundation for earnings, cash, assets, depreciation, leases, contracts and OpenAI exposure. Note 18 describes cost allocations and limits to asset disaggregation in the FY2026 accounts.

   Consolidated and segment measures do not isolate standalone AI economics. Note 18 allocation rules describe FY2026; unchanged application in FY2027 requires confirmation in the new accounts. Lease additions, cash payments and future commitments have different bases.

4. M04 · Microsoft · Document dated : 2026-10-14 · Accessed: 2026-10-28Microsoft Annual Report 2026 : Shareholder letter
   Passages and scope

   Shareholder letter: “Agents and Infra” and “Devices and Consumer” sections; signature dated 14 October 2026. Financial Review: FY2026 accounts retaining the historical three-segment presentation.

   The letter presents the “Agents and Infra” / “Devices and Consumer” strategic narrative. It reiterates adoption indicators and allows management’s plan to be compared with the annual accounts already published.

   14 October is the letter date; its upload date was not independently established. Strategic statements are attributed to management. The site reproduces the M03 accounts and is not an independent second dataset.

5. F03 · Microsoft Investor Relations · Document dated : 2026-07-29 · Accessed: 2026-10-28Microsoft Fiscal Year 2026 Fourth Quarter Earnings Conference Call
   Passages and scope

   Amy Hood’s remarks: results and other income/expense (Anthropic); commercial results (bookings, RPO, Cloud margin); transition before outlook (building useful life); end of outlook (CapEx). Search anchors: “$3.2 billion”, “25 years”, “operating leases”, “over $50 billion”.

   The transcript connects results with capacity, leases and customers. Amy Hood announces the longer building useful life, explains CapEx, and discusses RPO, Cloud margin and the Anthropic gain.

   The 29 July event mixes realised results and guidance. The 15-to-25-year change concerns datacenters and office buildings from FY2027; it does not give servers a 25-year life. More future contracts are expected to qualify as operating leases. Preserve published rounding.

6. F04 · Microsoft · Document dated : 2023-10-16 · Accessed: 2026-10-28Microsoft 2023 Annual Report
   Passages and scope

   MD&A, “Change in Accounting Estimate”; Note 1, “Estimates and Assumptions”. Search “four years to six years”; change based on net carrying amount at 30 June 2022.

   The documented precedent: Microsoft extended the useful life of server and network equipment from four to six years and disclosed the effect of that estimate on FY2023 accounts.

   16 October is the annual-letter date. The disclosed effect concerns FY2023; it is not a constant annual saving and cannot be attributed entirely to today’s AI equipment. This precedent concerns a different asset class from buildings.

7. F05 · Microsoft Investor Relations · Published : 2026-07-29 · Document dated : 2026-07-29 · Accessed: 2026-10-28FY26 Q4 Earnings Release: Microsoft Cloud and AI Strength Fuels Fourth Quarter Results
   Passages and scope

   Opening CEO quotation; Business Highlights; Fiscal Year 2026 Results; Income Statements; Cash Flows Statements; Non-GAAP Definition and reconciliation tables. Search “Anthropic”, “30 million” and “investments in OpenAI”.

   The release brings together Q4 and annual results, Cloud revenue, the Azure threshold, Copilot seats and cash-flow tables. Its non-GAAP definition specifies exactly the exclusion of OpenAI investment effects.

   Quarterly tables are labelled unaudited; audited annual accounts appear in M03. Some comparative cash-flow amounts were recast. The OpenAI adjustment retains other items, including the Anthropic gain. Copilot seats measure adoption.

8. E04 · Microsoft Investor Relations · Published : 2026-04-29 · Document dated : 2026-04-29 · Accessed: 2026-10-28Microsoft FY2026 Q3 : Press Release and Webcast
   Passages and scope

   Opening Satya Nadella quotation: search “annual revenue run rate” and “$37 billion”.

   The source for an AI annual revenue run rate above $37bn, announced by Satya Nadella with 123% year-over-year growth.

   The release describes an annualised rate announced in April. It does not detail the annualisation formula, full product perimeter or reconciliation to recognised annual or quarterly AI revenue. ARR should not be expanded to “annual recurring revenue”.

9. E05 · Microsoft Investor Relations · Document dated : 2026-01-28 · Accessed: 2026-10-28Microsoft Fiscal Year 2026 Second Quarter Earnings Conference Call
   Passages and scope

   Amy Hood, discussion of commercial remaining performance obligations; search “45%”, “$625 billion” and “2.5 years”.

   Management attributes approximately 45% of the $625bn commercial RPO balance to OpenAI. This dated reference supports analysis of backlog concentration without transferring the percentage to another quarter.

   The balance and concentration relate to 31 December 2025. The 45% share cannot be applied to the June 2026 RPO balance. Rounded figures in the commentary are not an exact dollar reconciliation.

10. E06 · Microsoft Official Blog · Published : 2026-04-27 · Document dated : 2026-04-27 · Accessed: 2026-10-28The next phase of the Microsoft–OpenAI partnership
    Passages and scope

    Five provisions under “Amended Agreement Provides Long-Term Clarity”: cloud/Azure-first; intellectual property through 2032; end of Microsoft payments; capped OpenAI payments through 2030; shareholding.

    The amendment specifies the primary-cloud role, distribution across clouds, the non-exclusive licence and the two directions of revenue sharing.

    The Azure-first rule includes the exception stated in the announcement. The revenue-share percentage and aggregate cap are not published. These terms do not support a separate quantified margin effect.

11. E07 · Microsoft Official Blog · Published : 2025-10-28 · Document dated : 2025-10-28 · Accessed: 2026-10-28The next chapter of the Microsoft–OpenAI partnership
    Passages and scope

    Paragraphs on the investment following recapitalisation and incremental Azure purchases; search “27%” and “$250 billion”.

    The historical agreement announces an additional $250bn of Azure purchases by OpenAI and describes Microsoft’s interest at recapitalisation.

    Terms were subsequently amended by E06. Approximately 27% describes recapitalisation; M03 reports approximately 25% at 30 June 2026. The Azure commitment is contractual and multiyear; it is not immediately recognised revenue or cash already collected.

12. R01 · Financial Accounting Standards Board · Published : 2023-11-27 · Document dated : 2023-11 · Accessed: 2026-10-28Accounting Standards Update 2023-07 : Segment Reporting (Topic 280): Improvements to Reportable Segment Disclosures
    Passages and scope

    Summary, printed pp. 1–4 (PDF pp. 4–7); amendments to ASC 280-10-50-21, 50-26A through 50-26C and 50-29.

    The update explains segment reporting based on internal management, the performance measures used and significant expenses supplied to the chief operating decision maker.

    Authoritative requirements reside in the Codification; the ASU communicates amendments. This framework does not prescribe an AI-specific profit measure. The publication date is corroborated by the FASB completed-project page.

13. R02 · U.S. Securities and Exchange Commission, Division of Corporation Finance · Published : 2022-12-13 · Document dated : 2016-05-17 · Accessed: 2026-10-28Non-GAAP Financial Measures : Compliance and Disclosure Interpretations, Question 102.07
    Passages and scope

    Question 102.07, addressing free cash flow; answer-specific date: 17 May 2016.

    The answer explains that free cash flow has multiple definitions and requires an explicit calculation reconciled to the accounts. It highlights mandatory payments that may remain outside the measure.

    SEC staff interpretation, without approval of any company’s calculation. Answer 102.07 is dated 17 May 2016; the page states a last update of 13 December 2022.

14. R03 · Microsoft Source · Published : 2026-10-07 · Document dated : 2026-10-07 · Accessed: 2026-10-28Microsoft announces quarterly earnings release date
    Passages and scope

    First paragraph: results after market close on 28 October; conference call at 2:30 p.m. Pacific Time.

    The announcement schedules FY2027 Q1 results for 28 October after market close and gives the call time. It establishes the need for an intraday evidence check.

    This announcement establishes a timetable. The FY2027 Q1 release is now available and was checked during this edition; its figures are cited separately under R04.

15. R04 · Microsoft Investor Relations · Published : 2026-10-28 · Document dated : 2026-10-28 · Accessed: 2026-10-28Microsoft Announces First Quarter Results : FY2027 Q1
    Passages and scope

    Business Highlights; Non-GAAP Definition and OpenAI reconciliation; Income Statements; Balance Sheets; Cash Flows Statements; Segment Results; Product and Service Offerings Revenue. Read the recast notes below the segment, offering and cash-flow tables.

    The release reports FY2027 Q1 results under the two new segments, including revenue, costs and operating income. It supplies Azure revenue in dollars, cash flows and the reconciliation of OpenAI investment effects.

    The statements are unaudited and cover three months. Certain FY2026 comparatives have been recast, notably for the segments announced in September. Quarterly data retain their period and definition. The release does not provide a standalone AI income statement or the full notes of a 10-Q.