Microsoft.
What the price requires.

Purpose. This is a quality sample of company research. It states findings and their evidence.

Fair value. Our central estimate is $326.86 per share. The 21 Oct 2026 close of $532.95 is 63.1% above this value.^S10S11 Practitioner checks give $84.49 to $609.26 (cross-check).

Risk measure. Scenarios use real company years. The FY2023 trough regime gives $221.35. The FY2026 run-rate regime gives $521.69; the price is 2.2% above it. The range is 92% of the central value. A stress with trough operations and capex held at 25% of revenue gives $157.25; it is outside the weights.

The business. Microsoft is an exceptional franchise in its heaviest investment phase. FY2026 revenue rose 17.8% to $331.8bn, with a 46.8% operating margin. Capex including finance leases reached $140.6bn, 42% of revenue; free cash flow after all of it was $42.4bn.^S1S7

What changes the view. Three inputs move the value most: the pre-depreciation margin at scale, the capex intensity after FY2030, and the discount rate. One point off the WACC gives $397.87. Holding the FY2026 margin of 57.1% gives $375.98. Terminal capex at 15% of revenue gives $360.16. Two thesis breakers sit outside the weights (thesis breakers). Watch Azure growth, commercial RPO excluding the partner and capex each quarter.

Review status: provisional. One blind round with four separate reviewers ran. It closed three facts (cost of debt, depreciation rate, a label) and moved the central value by less than 2%. The stability rule needs two quiet rounds; Q1 FY2027 results on 28 October 2026 come first.

No results filing followed the FY2026 10-K of 29 July. The capex guide changed basis, segment reporting changed, and the discount rate stayed high. The main open item is the concentration of the backlog in one AI-model partner. Q1 FY2027 results on 28 October will test it.

Each event is sorted by its channel into value: estimate, multiple, discount rate or flows. Window: 2026-07-29 to 2026-10-22.

Thesis challenged. The franchise evidence is intact. The thesis that the AI build-out earns its cost of capital is challenged by the concentration of the backlog: commercial RPO grew 84%, but 25% without one partner.^S1S13

The P/E method gives the highest figure. It assumes the market multiple of FY2023–26 (median 35.3×) holds while capex quadruples; that multiple was set when free cash flow was close to net income.^S10S3 The owner-earnings method treats depreciation as maintenance capex. The free-cash-flow method uses FY2026 cash after all capex, the trough of the build-out. The DCF sits between them because it lets capex intensity fall from 42% to 20% of revenue.

Micro forces. Azure-type revenue grew 41% in FY2026 and 43% in Q4.^S1S3 Seats for Microsoft 365 Commercial grew 6%; revenue per user rose with Copilot and E5.^S1 Windows OEM and Devices fell, and management guides a high-teens fall in FY2027.^S13

Macro forces. The 10-year Treasury yield is 5.13%.^S8 Cloud infrastructure spending grew 43% in Q2 2026, the fastest in eight years.^S15 Power and component supply limit capacity across the industry.

Validated data. Each figure carries its source, period and label: reported by the company, adjusted by the company, or calculated by us. Filing figures were read twice: from the document and from the regulator's structured data.^S1S7 Six source levels apply; conclusions rest on levels 1–3.

Deterministic model. One script computes every number in this note from the stated inputs. Prose numbers come from the model outputs. Route: a goodwill-heavy franchise, screened by owner earnings against market value and checked by earning power against reproduction cost.

Blind multi-angle review. Separate reviewers received the facts without the conclusion: a bull, a bear, a number check and a gap finder. Each worked alone. Their findings changed three inputs (changes).

Signal strength. Evidence is graded from regulator data down to platform signals. Press-only facts are marked Unverified and do not move the central case.

Compared with typical automated research. Typical outputs give one target price. This note shows what the price requires, three cases from real company years, stress cases outside the weights, and the evidence for and against each.

Weights. Scenario weights are the reader's judgement. The calculator in the summary computes a weighted value from any weights.

Reviewers also confirmed 14 reported figures against the 10-K and found no error in them. The Bull bridge includes the partner stake at its reported value; a reviewer read this as an error, but it is a stated case input.

We solve the central model for the growth that makes value equal the $532.95 price, holding margins, capex rules and the discount rate.^S25

Implied revenue: $1,496bn in FY2035 (fiscal year to June 2035), nominal US dollars, of which $1,002bn from Azure-type cloud. FY2026 revenue was $331.8bn.^S1 Our central path reaches $934bn.

Implied growth: 18.2% a year, FY2026 to FY2035, against 12.2% in the central path and 16.1% a year in FY2023–26 (calculated by us).

Implied discount rate: 7.33% at central operations, against our dated WACC of 9.57% (10-year Treasury 5.13% on 21 Oct 2026).^S8

Implied margin: a pre-depreciation margin of 73.4% from FY2030, against 57.1% in FY2026.

Implied market share: Azure-type revenue of $1,002bn is 60% of a FY2035 cloud infrastructure market of $1,670bn, against 20% today; total revenue is 46% of a $3,221bn total cloud market. Both market figures extend the dated estimates below at 20% a year to 2030 and 10% to 2035 (calculated by us).

Implied 9-year revenue CAGR by WACC and terminal pre-depreciation margin

Central value by WACC and terminal growth

Every cell of the first table asks for faster growth than the central path, unless the WACC falls by a point and the margin holds at the FY2026 level.

Each breaker starts from the Base path and changes one mechanism. Loss sizes are our assumptions.

Segment figures are reported by the company.^S1S2 Costs of AI infrastructure are allocated across segments by gross margin, so segment margins are not standalone economics.^S1 Depreciation by segment is not disclosed.

The company states that Azure revenue "surpassed $75 billion" in FY2025 and "$100 billion" in FY2026.^S4S3 Our FY2026 figure of $105.8bn fits the second statement. Earlier years are back-solved and are Unverified as dollar figures.

Totals add finance-lease additions to cash capex, both reported.^S1S7 The FY2027 guide of about $175bn includes finance leases.^S13S14 The life extension from 15 to 25 years moves more new leases to operating leases; their cost lands in operating expense, not capex.^S29

Maintenance versus growth. The company does not split capex. We use depreciation as the maintenance proxy: $34.3bn of $140.6bn in FY2026, so about 76% was growth capex (calculated by us). Management said about two-thirds of Q4 capex went to short-lived assets, mainly CPUs and GPUs.^S13 Servers and network equipment rose from $132.8bn to $215.9bn at cost in one year.^S1

Consolidated ROIC was 28.7% in FY2026 (42.0% without goodwill and intangibles); the incremental return on capital added since FY2023 was 18.6%, close to twice the WACC.

If AI demand fell short, the next-best use of the server base is wholesale rental of GPU capacity to other buyers. We value that route.

• GPU rental index, 7 Oct 2026: median H100 rent $4.20 per GPU-hour; spot fell from $6.62 (late 2023) to $2.83 (Feb 2026). ^S26
• Capacity contract, 8 Sep 2025: Microsoft agreed to buy $17.4bn of GPU capacity over five years from a neocloud. ^S27
• Peer filing, Mar 2025: one buyer was 62% of a GPU cloud's 2024 revenue, so rental demand itself is concentrated. ^S28

Assumptions (ours): servers at cost of $215.9bn, net $140.3bn after a 35% depreciation share; first-year rental revenue 30% of that net base; prices fall 15% a year; cash margin 55%; six-year life; taxed and discounted at the WACC.

Fallback value per share: $8.49, against $326.86 central. The fallback recovers a small part of the value. The $329.1bn of leases not yet commenced would still run.^S1

The central path puts Azure-type revenue at $541bn in FY2035, 32% of our extended infrastructure market. The price needs 60%.

• Counter-argument: the market grows faster. Q2 2026 growth was 43%, the 11th quarter of acceleration.^S15 At 30% a year to 2030 the FY2035 market is larger and the implied share falls. Our 20% then 10% path is an assumption.
• Counter-argument: scope. Microsoft sells software on its own cloud; part of Copilot and agent revenue sits outside infrastructure trackers, so a share of an infrastructure market understates reach.
• Counter-argument: concentration of buyers. A share of a market where a few AI labs buy most new capacity carries counterparty risk that a share figure hides.^S28

The 10-K names the main partner in its related-party note and states the ownership as about 25% on an as-converted basis.^S1 The partner's legal form changed to a public benefit corporation in the October 2025 recapitalisation.^S1S18

Operating lease liabilities ($21.9bn) are not deducted: their cost sits inside the operating margin. Finance leases are deducted because their assets sit in PP&E and future additions sit in capex.

No country premium is added; the company sells worldwide and the US mature-market premium is a stated simplification. Tax moves from the FY2026 effective 19.4% to 21% by FY2030.

Bear uses FY2023: revenue +7%, Azure about +30%, pre-depreciation margin 47.0%, capex 14.7% of revenue. Base uses FY2024–25: revenue +16% and +15%, margin 50.8%–53.4%. Bull uses FY2026: revenue +18%, Azure +41%, margin 57.1%.^S1S2 All cases anchor FY2027 on the operating-margin guide.^S13

Base path, fiscal years, $bn

FY2027 counts only the 252 days after the valuation date. Free cash flow stays low to FY2029 because capex follows the plan; it rises when capex falls towards 20% of revenue.

The company is goodwill-heavy (goodwill and intangibles $138.3bn), so a price-to-net-assets screen does not apply; at 13.0× tangible book it says nothing useful. The screen is owner earnings against market value; the check is earning power against reproduction cost.^S24

The NOPAT yield of 3.16% sits below the 5.13% Treasury yield. No-growth earning power is $172.92 per share; growth must supply the rest of the price. Earning power ($1,307bn) far exceeds a crude reproduction base of net PP&E, leases and acquired intangibles ($473bn): the franchise is real.

• Azure revenue before FY2025 and after the FY2027 segment change is calculated by us from growth rates (Unverified as dollar figures).^S5
• The partner's share of the backlog is a press estimate (Unverified).^S21 The 10-K does not split RPO by customer.
• Capex guidance comes from third-party call reports; the issuer transcript was not opened.^S13S14
• The partner stake value of about $135bn is a press figure (Unverified); Base keeps book value.^S18
• The registration status of the partner rests on EDGAR search and press (Unverified beyond 22 Oct 2026).^S19S6
• Depreciation by segment and the dollar effect of the life extension are not disclosed (Unverified).

A. Sensitivities on the central value

B. Annual forecasts, $bn

C. Inputs and provenance