1 · Business in one line
Microsoft sells software and compute to enterprises on subscription and consumption contracts — Microsoft 365 and Dynamics per seat, Azure per unit consumed, LinkedIn and Search on advertising, Windows on OEM royalty, XBOX on content. FY2026 revenue was $331.8B at a 46.8% operating margin, of which Microsoft Cloud was $214.4B (FY2026 10-K, MD&A).
It is a stalwart with a capital-intensive fast grower bolted onto it. Productivity and Business Processes still behaves like the old Microsoft — $140.0B revenue, 82% gross margin, 59.9% operating margin, +16% growth on almost no incremental capital. Intelligent Cloud no longer does: revenue grew 30% to $137.8B while its cost of revenue grew 44%, on $115.9B of property and equipment additions. The halves fail differently — PBP's risk is pricing and churn; Intelligent Cloud's is now the capital-cycle risk of a utility: sunk assets, fixed depreciation, uncertain utilization.
2 · Bull case — Peter Lynch pitch
The simple reason this stock could work is that Microsoft has already sold most of the AI revenue it needs — $678 billion of contracted commercial backlog against $331.8 billion of annual revenue — and it collects that revenue on an installed base it does not have to go out and buy.
A1The backlog is real demand and it converts›
Commercial remaining performance obligation converts to revenue on roughly its stated schedule.
Microsoft sells multi-year enterprise agreements invoiced annually at the start of each contract period, so backlog is contracted and largely invoiced — not pipeline.
Commercial RPO $678B at 30 Jun 2026, up 84% from $368B; weighted average duration 2.3 years; ~30% recognizable within 12 months (≈$203B). That is 2.0x annual revenue versus 1.3x a year earlier. (FY2026 and FY2025 10-K, Note 12)
RPO dollars AND the 12-month conversion share, read together.
A2AI is sold as price on seats Microsoft already owns›
Microsoft 365 revenue per seat keeps rising materially faster than seat count.
Copilot and E5 are upsells into an already-paying base: distribution cost is near zero and the incremental dollar lands in the 82%-gross-margin segment, not the one that needs GPUs to serve it.
M365 Commercial cloud revenue grew 17% on 6% seat growth — roughly 10–11 points from revenue per user, which management attributes to M365 Copilot and E5. PBP gross margin percentage rose in FY2026 while total company gross margin percentage fell. (FY2026 10-K, MD&A)
The spread between M365 Commercial cloud revenue growth and seat growth.
A3The infrastructure earns back its depreciation›
Microsoft Cloud gross margin stops falling as newly built capacity fills.
Microsoft can sell the same GPU-hour at three layers — raw Azure, platform services, and finished Copilot seats — so it recovers cost at whichever layer prices best. A pure infrastructure provider has only the first layer.
Microsoft Cloud gross margin 66% in FY2026 versus 69% in FY2025, which management attributes to AI infrastructure investment and growing AI usage "offset in part by efficiency gains in Azure and Microsoft 365 Commercial cloud." Azure grew 41%. (FY2026 10-K, MD&A)
Microsoft Cloud gross margin percentage; Intelligent Cloud cost-of-revenue growth against segment revenue growth.
A4The build is self-funded and does not cost shareholders their capital return›
Operating cash flow keeps covering capex plus dividends and buybacks without material new leverage.
The subscription model collects cash in advance of recognition — unearned revenue ended FY2026 at $75.7B — so the build is funded by customers as much as by the balance sheet.
$182.9B of operating cash flow (up $46.8B) funded $115.9B of capex and still left $27.0B of declared dividends and $16.7B of repurchases; long-term debt fell to $31.1B from $40.2B. (FY2026 10-K, cash flow statement and MD&A)
Additions to property and equipment as a share of operating cash flow.
Must happen. Commercial remaining performance obligation converts to revenue on roughly its stated schedule.
Why Microsoft. Microsoft sells multi-year enterprise agreements invoiced annually at the start of each contract period, so backlog is contracted and largely invoiced — not pipeline.
Commercial RPO $678B at 30 Jun 2026, up 84% from $368B; weighted average duration 2.3 years; ~30% recognizable within 12 months (≈$203B). That is 2.0x annual revenue versus 1.3x a year earlier. (FY2026 and FY2025 10-K, Note 12)
B2 The backlog's credit quality is undisclosed and partly circular
Commercial RPO nearly doubled, but the share converting within 12 months fell from 40% to 30% — the growth is in longer-dated commitments from counterparties Microsoft does not name. Separately, it discloses $24.1B of FY2026 revenue from commercial arrangements with OpenAI, inclusive of revenue-sharing payments — 7.3% of total revenue — from an entity in which it holds an approximate 25% as-converted interest and has committed $13.0B of funding, $11.9B funded. Microsoft funds the counterparty, the counterparty buys Azure, Microsoft books the revenue and the backlog.
Damage. Revenue that reverses, plus impairment of the equity-method investment and of capacity built specifically to serve it. The revenue is replaceable over time; capacity purpose-built for one counterparty is not.
Must happen. Microsoft 365 revenue per seat keeps rising materially faster than seat count.
Why Microsoft. Copilot and E5 are upsells into an already-paying base: distribution cost is near zero and the incremental dollar lands in the 82%-gross-margin segment, not the one that needs GPUs to serve it.
M365 Commercial cloud revenue grew 17% on 6% seat growth — roughly 10–11 points from revenue per user, which management attributes to M365 Copilot and E5. PBP gross margin percentage rose in FY2026 while total company gross margin percentage fell. (FY2026 10-K, MD&A)
B5 There is no second engine
More Personal Computing revenue fell 1% to $54.1B: XBOX content and services −5%, XBOX hardware −29%, plus impairment and other related expenses in FY2026. Windows OEM grew 5% but management notes channel inventory "remained elevated," which is a pull-forward rather than demand. And M365's 6% seat growth came from small and medium business and frontline worker offerings — the lowest-ARPU cohorts.
Damage. Nothing absorbs an Intelligent Cloud margin reset. The AI leg has to carry the whole result. Fixable in principle; nothing in FY2026 disclosure suggests it is being fixed.
Must happen. Microsoft Cloud gross margin stops falling as newly built capacity fills.
Why Microsoft. Microsoft can sell the same GPU-hour at three layers — raw Azure, platform services, and finished Copilot seats — so it recovers cost at whichever layer prices best. A pure infrastructure provider has only the first layer.
Microsoft Cloud gross margin 66% in FY2026 versus 69% in FY2025, which management attributes to AI infrastructure investment and growing AI usage "offset in part by efficiency gains in Azure and Microsoft 365 Commercial cloud." Azure grew 41%. (FY2026 10-K, MD&A)
B1 The depreciation staircase
Depreciation expense was $34.3B in FY2026 on gross property and equipment of $431.8B. Servers, network equipment and software at cost went from $132.8B to $215.9B in one year, depreciated over two to six years. The FY2026 charge reflects assets in service for only part of the year; a full year of the current server base at a five-year life is roughly $43B on that line alone, before FY2027 additions. Depreciation runs whether or not the racks are utilized.
Damage. Intelligent Cloud gross margin was 58.0% in FY2026 versus 62.2% in FY2025. At current segment revenue each further point costs about $1.4B of gross profit, against a cost base that cannot be cut in the year the demand disappoints. Largely permanent. The assets are sunk, and Microsoft's own risk factors state that overestimating demand "may result in underutilization of infrastructure and may lead to impairment of assets on our balance sheet."
Must happen. Operating cash flow keeps covering capex plus dividends and buybacks without material new leverage.
Why Microsoft. The subscription model collects cash in advance of recognition — unearned revenue ended FY2026 at $75.7B — so the build is funded by customers as much as by the balance sheet.
$182.9B of operating cash flow (up $46.8B) funded $115.9B of capex and still left $27.0B of declared dividends and $16.7B of repurchases; long-term debt fell to $31.1B from $40.2B. (FY2026 10-K, cash flow statement and MD&A)
B1 The depreciation staircase
Depreciation expense was $34.3B in FY2026 on gross property and equipment of $431.8B. Servers, network equipment and software at cost went from $132.8B to $215.9B in one year, depreciated over two to six years. The FY2026 charge reflects assets in service for only part of the year; a full year of the current server base at a five-year life is roughly $43B on that line alone, before FY2027 additions. Depreciation runs whether or not the racks are utilized.
Damage. Intelligent Cloud gross margin was 58.0% in FY2026 versus 62.2% in FY2025. At current segment revenue each further point costs about $1.4B of gross profit, against a cost base that cannot be cut in the year the demand disappoints. Largely permanent. The assets are sunk, and Microsoft's own risk factors state that overestimating demand "may result in underutilization of infrastructure and may lead to impairment of assets on our balance sheet."
B3 The obligations now exceed the balance sheet and are not cancellable
Total contractual obligations were $743.8B at 30 Jun 2026 against total assets of $758.4B, with $241.9B due in FY2027 alone. Leases not yet commenced — primarily datacenters — rose from $92.7B to $329.1B; purchase commitments, described as datacenter open purchase orders and take-or-pay contracts, are $194.1B. Cash and short-term investments fell from $94.6B to $76.8B while finance lease liabilities rose to $66.6B from $46.2B.
Damage. It removes the option to simply stop spending. If demand disappoints the payments continue, and the flexibility comes out of the buyback first — which is already token: 36 million shares repurchased in FY2026 moved the diluted count only from 7,465 million to 7,453 million. Fixable only over years, at the pace the commitments roll off.
B4 Headline earnings are flattered by a non-cash mark
FY2026 net income rose 31% to $133.7B, but $5.0B of that — $0.67 of the $17.95 diluted EPS — was net gains on the OpenAI investment, primarily a dilution gain from the recapitalization: an accounting consequence of Microsoft's ownership shrinking, not cash earned. On the company's own adjusted basis net income rose 22%, not 31%.
Damage. No effect on cash, but the growth rate taken off the income statement overstates the operating result by nine percentage points — which matters if the position is underwritten on reported earnings momentum. A measurement issue that will keep swinging both ways, not a value issue.
Why the market might be missing it: reported free cash flow has fallen three years running — roughly $74.1B in FY2024, $71.6B in FY2025, $67.0B in FY2026 — while net income rose from $88.1B to $133.7B. A reader anchored on free cash flow sees deterioration. A reader who checks the backlog sees the opposite: the 84% jump in RPO landed in the same year as the capex, which is the sequence you want if the spending is demand-led rather than speculative.
3 · Bear case — Munger invert
The most likely way I lose money is that the $115.9 billion Microsoft spent in FY2026 shows up as depreciation before it shows up as revenue, and Intelligent Cloud's margin structure resets permanently lower.
Ordered by how permanent the damage is, not by the order of the bull assumptions.
B1The depreciation staircaseattacks A3, A4›
Depreciation expense was $34.3B in FY2026 on gross property and equipment of $431.8B. Servers, network equipment and software at cost went from $132.8B to $215.9B in one year, depreciated over two to six years. The FY2026 charge reflects assets in service for only part of the year; a full year of the current server base at a five-year life is roughly $43B on that line alone, before FY2027 additions. Depreciation runs whether or not the racks are utilized.
Intelligent Cloud cost of revenue continuing to grow faster than segment revenue — it grew 44% against 30% in FY2026 — while Azure decelerates from 41%.
Intelligent Cloud gross margin was 58.0% in FY2026 versus 62.2% in FY2025. At current segment revenue each further point costs about $1.4B of gross profit, against a cost base that cannot be cut in the year the demand disappoints.
Largely permanent. The assets are sunk, and Microsoft's own risk factors state that overestimating demand "may result in underutilization of infrastructure and may lead to impairment of assets on our balance sheet."
B2The backlog's credit quality is undisclosed and partly circularattacks A1›
Commercial RPO nearly doubled, but the share converting within 12 months fell from 40% to 30% — the growth is in longer-dated commitments from counterparties Microsoft does not name. Separately, it discloses $24.1B of FY2026 revenue from commercial arrangements with OpenAI, inclusive of revenue-sharing payments — 7.3% of total revenue — from an entity in which it holds an approximate 25% as-converted interest and has committed $13.0B of funding, $11.9B funded. Microsoft funds the counterparty, the counterparty buys Azure, Microsoft books the revenue and the backlog.
Accounts receivable from OpenAI growing faster than revenue from OpenAI — it stood at $6.0B against $24.1B at year end, roughly 91 days; RPO growth stalling; any disclosed renegotiation.
Revenue that reverses, plus impairment of the equity-method investment and of capacity built specifically to serve it.
The revenue is replaceable over time; capacity purpose-built for one counterparty is not.
B3The obligations now exceed the balance sheet and are not cancellableattacks A4›
Total contractual obligations were $743.8B at 30 Jun 2026 against total assets of $758.4B, with $241.9B due in FY2027 alone. Leases not yet commenced — primarily datacenters — rose from $92.7B to $329.1B; purchase commitments, described as datacenter open purchase orders and take-or-pay contracts, are $194.1B. Cash and short-term investments fell from $94.6B to $76.8B while finance lease liabilities rose to $66.6B from $46.2B.
Another step-up in leases not yet commenced without a matching step-up in RPO.
It removes the option to simply stop spending. If demand disappoints the payments continue, and the flexibility comes out of the buyback first — which is already token: 36 million shares repurchased in FY2026 moved the diluted count only from 7,465 million to 7,453 million.
Fixable only over years, at the pace the commitments roll off.
B4Headline earnings are flattered by a non-cash markattacks A4›
FY2026 net income rose 31% to $133.7B, but $5.0B of that — $0.67 of the $17.95 diluted EPS — was net gains on the OpenAI investment, primarily a dilution gain from the recapitalization: an accounting consequence of Microsoft's ownership shrinking, not cash earned. On the company's own adjusted basis net income rose 22%, not 31%.
The same line reversing, as it did in FY2025 (−$3.6B) and FY2024 (−$1.1B).
No effect on cash, but the growth rate taken off the income statement overstates the operating result by nine percentage points — which matters if the position is underwritten on reported earnings momentum.
A measurement issue that will keep swinging both ways, not a value issue.
B5There is no second engineattacks A2›
More Personal Computing revenue fell 1% to $54.1B: XBOX content and services −5%, XBOX hardware −29%, plus impairment and other related expenses in FY2026. Windows OEM grew 5% but management notes channel inventory "remained elevated," which is a pull-forward rather than demand. And M365's 6% seat growth came from small and medium business and frontline worker offerings — the lowest-ARPU cohorts.
M365 seat growth slowing further while the revenue-per-seat spread narrows, with MPC still declining.
Nothing absorbs an Intelligent Cloud margin reset. The AI leg has to carry the whole result.
Fixable in principle; nothing in FY2026 disclosure suggests it is being fixed.
4 · Signals to monitor
Each signal carries the tag of the assumption or risk it tracks.
5 · Bottom line
- 1
The stock could work because Microsoft has $678B of contracted commercial backlog against $331.8B of revenue and monetises AI as a price increase on a seat base it already owns, where the incremental dollar carries an 82% gross margin.
- 2
What must go right is that Microsoft Cloud gross margin troughs near 66% and recovers as the FY2026 capacity fills, so that the revenue from the backlog arrives faster than the depreciation on the $431.8B gross asset base that serves it.
- 3
The thesis most likely breaks on the depreciation staircase: $115.9B of capex and a server base that went from $132.8B to $215.9B at cost create a fixed two-to-six-year charge that lands whether or not utilisation follows, against $743.8B of contractual obligations that cannot be cancelled if it does not.
- 4
The evidence that would change my mind is two or three consecutive quarters in which Intelligent Cloud cost of revenue grows faster than its revenue while Azure decelerates and the 12-month share of commercial RPO keeps falling — that combination would mean the capacity was built ahead of demand rather than behind it.
6 · Sources
Company filings used (all from the MSFT research folder)
- FY2026 Form 10-K, year ended 30 June 2026, filed 29 July 2026 — primary source for all FY2026 and FY2025 comparatives
- FY2025 Form 10-K, year ended 30 June 2025 — RPO, leases not yet commenced, Microsoft Cloud gross margin comparatives
- FY2024 Form 10-K, year ended 30 June 2024 — FY2023/FY2024 cash flow and capex
- FY2026 Q1 Form 10-Q, filed 29 October 2025 — intra-year RPO trajectory ($392B, +51%)
- FY2026 Q3 Form 10-Q, filed 29 April 2026 — intra-year RPO trajectory ($627B, +99%); nine-month capex
Not available in sources — read the memo knowing this
- No earnings call transcripts, earnings presentations or investor day materials — management's forward capex commentary is not in evidence
- Azure revenue is disclosed only as a growth rate, never as a dollar figure
- No customer concentration disclosure inside the $678B commercial RPO
- No Copilot seat or attach-rate disclosure
- "Microsoft 365 Consumer subscribers" was removed as a disclosed metric in Q1 FY2026
- Several DEFA14A proxy supplements failed to download during collection
Figures are as reported by Microsoft in the filings above. Interpretation, classification and the ordering of risks are the analyst's. This is a thinking anchor — not a valuation, a target or a recommendation.