MSFT · Research
Microsoft Is Pouring Its Software Cash Into Concrete
Microsoft still runs the best software franchise in the market, but in fiscal 2026 it spent 77% of its operating cash on data centres. After full capex, finance leases and stock pay, owner cash fell to $17.31B, and the valuation now asks for 37.86% a year of growth from that base.
Thesis in one line: Microsoft is still the best software franchise in the market, but 77% of its operating cash now goes to data centres and owner cash fell $26B in two years, from $43.20B to $17.31B; the bet is whether the concrete earns software returns.
Summary and thesis
Microsoft enters its Oct 28 print with the strongest demand signal in its history. Fiscal 2026 revenue rose 18% to $331.84B, Microsoft Cloud revenue rose 27% to $214.40B, Azure and other cloud services grew 41%, and commercial remaining performance obligation rose 84% to $678B [1].
Operating income rose 21% to $155.24B, a 46.78% margin [1].
The cash picture is different. Operating cash flow was $182.94B, but cash capex was $115.95B and new finance leases added another $24.61B.
Full capex reached $140.56B, 42.36% of revenue and 76.83% of operating cash flow, up from 25% of operating cash flow in fiscal 2017 [1]. After stock pay, interest income and the build in deferred revenue, owner cash was $17.31B, down from $43.20B in fiscal 2024.
Our thesis: Microsoft has turned from a software company that happens to own data centres into a capacity company that happens to sell the best software. Superintelligence demand is supply constrained, so the capacity is probably the right call.
But the valuation no longer rests on software economics. At the current enterprise value of $3.90T, the market is asking owner cash to compound at 37.86% a year for ten years at a 9% discount rate.
That is achievable only if capex intensity falls back as the build matures. The question for Oct 28 is whether there are signs of that.
The business model
Microsoft sells three things. Productivity and Business Processes is Microsoft 365 for businesses and consumers, LinkedIn and Dynamics 365. It earned $140.00B of revenue in fiscal 2026 at a 59.91% operating margin [1].
Growth comes from more seats (Microsoft 365 Commercial seats grew 6%) and higher revenue per user, which the company attributes to Microsoft 365 Copilot and E5 [1].
Intelligent Cloud is Azure, server products, GitHub cloud services and enterprise services. It earned $137.79B of revenue at a 41.35% margin [1].
Its cost of revenue rose 44% against revenue growth of 30%, the clearest sign in the filing of what the capacity build costs before it pays back [1].
More Personal Computing is Windows OEM, Surface, XBOX and search advertising. Revenue was $54.05B, down 1%, at a 26.62% margin.
XBOX content and services fell 5%, and the year included impairment and related costs in the XBOX business [1].
From the Oct 28 print Microsoft discloses new segments, including Agents and Infra and Devices and Consumer, recast in an 8-K filed Sep 2, 2026 [2]. Comparisons with the old segments above will need the recast history.
| Segment, fiscal 2026 | Revenue $B | Growth | Operating income $B | Margin |
|---|---|---|---|---|
| Productivity and Business Processes | 140.00 | 16% | 83.88 | 59.9% |
| Intelligent Cloud | 137.79 | 30% | 56.97 | 41.4% |
| More Personal Computing | 54.05 | -1% | 14.39 | 26.6% |
| Total | 331.84 | 18% | 155.24 | 46.8% |
Owner cash: where the money goes
Under our method, owner cash is operating cash flow minus all capex including new finance leases, minus stock pay, minus interest income and minus the build in deferred revenue. Microsoft does not disclose capitalised software separately, so that line is not shown; any such spend sits inside the capex figure.
No one-time tax item is disclosed for fiscal 2026. Cash taxes paid fell to $21.19B from $28.70B, which the filing does not explain, so we have not adjusted for it [1].
| $B, fiscal year to June | 2024 | 2025 | 2026 |
|---|---|---|---|
| Operating cash flow | 118.55 | 136.16 | 182.94 |
| Less cash capex | 44.48 | 64.55 | 115.95 |
| Less new finance leases | 11.63 | 20.51 | 24.61 |
| Less stock pay | 10.73 | 11.97 | 12.40 |
| Less interest income | 3.16 | 2.65 | 3.30 |
| Less deferred revenue build | 5.35 | 5.44 | 9.36 |
| Owner cash | 43.20 | 31.04 | 17.31 |
| Owner cash margin | 17.6% | 11% | 5.2% |
| Full capex, % of revenue | 22.9% | 30.2% | 42.4% |
The finance lease line matters. Microsoft counts finance leases in its own capex guidance, and new finance lease assets rose from $3.13B in fiscal 2023 to $24.61B in fiscal 2026 [1].
Finance lease liabilities stood at $66.59B at Jun 30, 2026, larger than the company's $40.29B of long-term debt [1]. Our enterprise value uses debt less cash and excludes investment securities and lease liabilities; adding the lease liabilities would lift it by about 1.71%.
The staircase: from EBITDA to owner cash
EBITDA for fiscal 2026 is $193.77B: operating income of $155.24B plus depreciation, amortization and other of $38.53B, both from the fiscal 2026 10-K (the D&A line is the cash flow statement figure and includes small other items) [1]. At an enterprise value of $3.90T that is 20.11x, close to what most market data sites show.
Each step below takes the same enterprise value and divides it by a narrower measure of cash.
| Step | Cash measure $B | EV multiple | What changes |
|---|---|---|---|
| EV / EBITDA | 193.77 | 20.11x | Profit before depreciation. Ignores capex entirely. |
| EV / operating cash flow | 182.94 | 21.30x | Actual cash from operations after tax and working capital. |
| After stock pay and cash capex | 54.58 | 71.39x | Subtracts $115.95B of data centre spending and $12.41B of stock pay. Capex is three times D&A, so EBITDA overstates spare cash. |
| After lease principal | 38.82 | 100.38x | Also subtracts interest income, the deferred revenue build and $3.10B of finance lease repayments. |
| Strict owner cash | 17.31 | 225.09x | Counts $24.61B of new finance leases as capex, as Microsoft does in its own capex figure. Full capex: $140.56B. |
The plain reading: EBITDA treats data centres as free. Microsoft spent $140.56B on them in fiscal 2026, more than three and a half times its $38.53B of depreciation and amortization.
Until that spending flattens, the 20.11x headline and the 225.09x bottom line describe very different companies.
What the price implies
| Cash measure, fiscal 2026 | $B | EV / measure | Implied growth at 8% | at 9% | at 10% |
|---|---|---|---|---|---|
| Owner cash (full capex incl. finance leases) | 17.31 | 225.09x | 34.5% | 37.9% | 40.9% |
| Owner cash with lease principal instead of new leases | 38.82 | 100.38x | 23.4% | 26.4% | 29.2% |
| Cash flow after cash capex and stock pay | 54.58 | 71.39x | 18.9% | 21.8% | 24.4% |
| Operating cash flow, before any capex | 182.94 | 21.30x | 3.4% | 5.8% | 8% |
Reverse discounted cash flow, ten years, 3% terminal growth. On the strict measure the market needs 37.86% annual growth in owner cash; before capex it needs only 5.82% growth in operating cash flow.
The gap between those two numbers is the whole debate. If the capacity build is temporary, the right lens is closer to operating cash flow and the stock looks reasonable.
If capex intensity stays above 40% of revenue, the strict number is the honest one. We do not publish a base value: owner cash has been positive for three years but has fallen each year, so there is no stable anchor.
Peers
| Company | Period end | EV / EBITDA | Revenue $B | Operating margin | Cash capex % revenue | Stock pay % revenue | (OCF - capex - stock pay) % revenue | EV / (OCF - capex - stock pay) |
|---|---|---|---|---|---|---|---|---|
| MSFT | 2026-06-30 | 20.11x | 331.84 | 46.8% | 34.9% | 3.7% | 16.5% | 71.39x |
| GOOG | 2026-06-30 | 23.10x | 445.87 | 33.1% | 29.7% | 6.3% | 5.6% | 158.90x |
| ORCL | 2026-08-31 | 13.72x | 71.78 | 32.1% | 105.4% | 6.7% | -46.7% | negative |
| CRM | 2026-07-31 | 18.68x | 43.94 | 19.9% | 1.4% | 8.3% | 26.2% | 20.58x |
| IBM | 2026-09-30 | 16.64x | 69.87 | 17.8% | 1.7% | 2.8% | 18.2% | 23.25x |
Peer figures are trailing twelve months from each company's latest filing and use a simpler measure than ours: operating cash flow less cash capex and stock pay, with no finance leases, interest or deferred revenue adjustments. Oracle's cash capex exceeds its revenue.
Alphabet's figures are for the whole company, using its total diluted share count across share classes. Peer EBITDA is trailing twelve-month operating income plus the depreciation and amortization lines of each cash flow statement, built from the 10-K and latest 10-Q [4].
Alphabet's cash flow statement shows depreciation of property and equipment only. IBM does not disclose operating income, so its operating income here is pre-tax income plus interest expense, and its figures run to Sep 30, 2026. Enterprise values use the same Oct 22, 2026 close and debt less cash method as Microsoft.
Amazon is not yet in our dataset.
Microsoft still turns more of each revenue dollar into cash after capex and stock pay than Alphabet does, and far more than Oracle. Salesforce and IBM, which rent rather than build capacity, trade at around 20x to 24x that measure.
Microsoft trades at 71.39x on it, a premium the market grants for the growth in its order book.
What to watch on Oct 28
- Data centre capacity. The binding variable. Any figure for capacity added, gigawatts contracted or capacity constraints easing matters more than the revenue beat.
- Capex including finance leases against guidance of over $50B for the quarter [3]. A higher number with Azure at or above the 44% to 45% constant-currency guide is the bull case; higher capex with slower Azure is the bear case.
- Accounting change. A 25-year useful life for data centre leases moves some leases from finance to operating treatment [3]. That can lower disclosed capex without changing the economics. We will keep counting both.
- Microsoft 365 Commercial revenue per user. Price increases and Copilot attach should show up here. The filing gives no Copilot seat count; that figure is not available.
- GitHub. Revenue sits inside Azure and other cloud services and is not disclosed separately; any usage or revenue figure would be new.
- XBOX. Content and services fell 5% in fiscal 2026 with impairment costs. Watch whether the new Devices and Consumer segment stabilises.
- Commercial remaining performance obligation. At $678B, about 30% is expected to be recognised in the next 12 months [1].
What would prove us wrong
We are cautious on the valuation, not the business. We would be wrong if capex intensity peaks in fiscal 2027 while Microsoft Cloud growth holds above 25%, because owner cash would then rebound sharply from a depressed base.
Our view would also be too cautious if the order book converts faster than the 30% the company expects over the next twelve months.
More Published Research
View All →Software Is Harvesting: Margins Widen, Stock Pay Shrinks, and Only 21 of 102 Names Clear 40 on Owner Cash
Across 104 software companies with more than $1B of revenue, margins widened at 69 of 97 and stock pay fell as a share of revenue at 75 of 102. Count stock pay as a cost and the median owner cash margin is 12.93%, and only 21 names clear the rule of 40.