SYNTHOS RESEARCH

Microsoft MSFT

Technology · Software - Infrastructure · Synthos Deep Dive · 2026-08-04

$513.53
Buy — Tactical
—
—

The Overview

Microsoft sells the software that businesses run on — the office suite, the email, the databases, the professional network, the video-game business, and above all the rented computing power that companies use instead of owning their own machines.

For most of the past year, investors did not like it. The shares fell 34% in the first three months of 2026 and are still down about 6% over twelve months in which the wider market rose 24%. The reason was simple and rational: Microsoft has been spending staggering sums building data centres. It spent $116 billion last year, which is 35 cents of every dollar of revenue. Historically Microsoft was a business that took your money and barely had to spend any to earn more of it. Now it has to build factories.

The worry is that all that concrete and equipment has to be written off over time as an expense, and that the write-offs will arrive before the revenue does — turning a wonderfully profitable software company into an ordinary infrastructure company.

Then the annual report arrived on 29 July and gave the strongest available answer to that worry. Microsoft disclosed that it has $678 billion of business already signed and contracted but not yet delivered — up 84% in a single year, against annual sales of $332 billion. In other words, more than two years of business is already booked. About 30% of it turns into revenue within the next twelve months.

That does not prove the spending will pay off. But it is the closest thing to proof that anyone has offered, because it is a contractual number, not a forecast.

Two things you should hold against it. First, the cash Microsoft actually keeps has been falling for three years — $74 billion, then $72 billion, then $67 billion — even as accounting profits rose from $88 billion to $134 billion. All the difference went into buildings and equipment. Second, the report reveals $329 billion of data-centre leases that have been signed but have not started yet. That is a real obligation, running out to 2033, and it appears in none of the debt figures anyone quotes.

The shares cost 25 times what analysts expect the company to earn in the year to June 2027. For a business earning a 33% return on its equity, that is reasonable. Our estimate of fair value is $550 against a price of $492.81.

The catch is timing. The shares have jumped 19% in three months, and the standard measure of "overbought" reads 80.6 out of 100 — the most stretched in this batch. Good business, fair price, poor moment. Buy it over several weeks rather than in one go.

Putting a number on it: our fair-value estimate is $550 against a current price of $513.53 — real upside if our numbers are right.

Target entry zone $431 – $514 accumulate in this band; ideal adds on a dip toward the 200-day average near $431, keeping roughly a 7% margin below our $550 base-case fair value

Our summary metrics

Downside Risk (lower = safer)5/10Moderate

"Rated 5 — moderate, with an unusually well-documented and unusually large off-balance-sheet component. The supports are exceptional: trailing return on equity of 33.2%, return on invested capital of 20.6%, gross margin of 67.9% and operating margin of 46.8% (both near the top of the entire S&P 500); interest coverage of 50.9x; $76.843B of cash and short-term investments; operating cash flow of $182.935B in FY2026, up $46.8B; a $678B contracted commercial backlog with a 2.3-year weighted duration; accrued legal liabilities of only $553M with a further ~$400M reasonably possible; and the vendor's own quality model rating it A− with an overall score of 4 — the highest in this batch. Against that, four specific items. (a) Free cash flow has DECLINED three years running — $74.071B (FY2024), $71.611B (FY2025), $66.987B (FY2026) — while net income rose from $88.136B to $133.749B; the free-cash-flow yield is 1.83% and the price-to-free-cash-flow ratio is 54.6x. (b) Capital expenditure reached $115.948B, 34.9% of revenue, up from $44.477B two years ago. (c) The 10-K discloses $329.1 BILLION of data-centre leases that have not yet commenced, running from fiscal 2027 to fiscal 2033 with terms of one to twenty years — an obligation 2.6 times the entire on-balance-sheet debt load and absent from every leverage ratio in the vendor file. (d) Of the $128.808B of reported total debt, $88.514B (68.7%) is finance-lease obligation, not bonds; the true bond debt is $40.294B. Microsoft's own risk factors now state plainly that these investments 'are in advance of fully developed revenue streams' and that 'the associated revenue may not be realized in the expected timeframes or at expected levels.' This is a returns-on-capital risk, not a solvency risk — but it is the reason the score is 5 rather than 3."

Growth Quality8/10Very High

"Rated 8 — accelerating, high-quality, and better documented than any other name in this batch. Revenue: $211.915B (FY2023), $245.122B (FY2024, +15.7%), $281.724B (FY2025, +14.9%), $331.839B (FY2026, +17.8%) — the growth rate has accelerated at a $330B base. The FY2026 10-K supplies the composition: 'Microsoft Cloud revenue increased 27% to $214.4 billion'; 'Azure and other cloud services revenue increased 41%'; Microsoft 365 commercial cloud +17% and consumer cloud +28%; Dynamics 365 +18%; LinkedIn +11%; search advertising excluding traffic-acquisition costs +12%. At the product-line level, Server Products and Cloud Services grew 31.5% to $129.425B and Microsoft 365 Commercial grew 16.2% to $101.997B — together 69.8% of revenue. Segment operating income rose $26.7B or 21%, with Intelligent Cloud contributing +$12.4B (+28%) and Productivity and Business Processes +$14.1B (+20%). Adjusted diluted EPS grew 22.3% ($14.13 to $17.28) once the equity-method investment swing is removed. And the forward visibility is the strongest number on this page: commercial remaining performance obligation rose 84% to $678B against $331.839B of annual revenue, with roughly 30% expected to be recognised within twelve months — approximately $203B of the next year's revenue already contracted. What holds it at 8 rather than 9: two segments are shrinking (Windows −1.3%, gaming −7.1% on the vendor's product line and −5% on the filing's content-and-services measure), and consensus models adjusted EPS growth decelerating to +13.6% in FY2027 before re-accelerating."

Exponential Potential7/10High

"Rated 7 — genuinely high, located in cloud infrastructure and in the contracted backlog rather than in the mature software franchises. The case for the 7: an infrastructure line growing 41% at scale; a commercial backlog that grew 84% in a single year to more than two years of total company revenue; the operating leverage of a business whose gross margin is 67.9% applied to a revenue base compounding 17-20%; and a distribution position across essentially every large enterprise on earth, which the tracked lane describes as 'every enterprise runs on it.' The case against a 9: Microsoft is executing a deliberate transformation from an asset-light licensing model into an asset-heavy infrastructure operator, and asset-heavy businesses do not compound exponentially — property, plant and equipment has gone from $154.552B (FY2024) to $229.789B to $337.253B in two years, capital expenditure is 34.9% of revenue, and depreciation from that base is a permanent new claim on the income statement. Consensus reflects it: adjusted EPS growing ~13.6% next year against revenue growing 17.4%. Additionally, at $3.66 trillion the arithmetic of size applies. Exponential in backlog and in the cloud line; increasingly industrial in structure."

Fair value$550 $393–$667
What does “fair value” mean?

Fair value is Synthos’s estimate of what one share is worth today, based on our model of the company’s future cash generation and the risks to it. It is not a price target or a prediction of where the stock trades next quarter — it’s the price where we think risk and reward are balanced. Above it, you’re paying for outcomes better than our base case; below it, the market is offering a margin of safety.

The model’s inputs and weightings are proprietary — they’re the product. What we publish is the output, plus an independent, public-math cross-check further down the page (“Check our number”) so you can judge it for yourself.

What do the 5 tiers mean? (Core · Tactical · Watch · Hold · Avoid)
Buy — CoreOwn it as a foundation — start or add now, size it for years, let dips be gifts.
Buy — TacticalGood price + confirmed trend + a defined exit — buy the setup, not a marriage.
WatchWe want the business, just not at this price/setup — act only when the listed trigger hits.
HoldFine to keep if you own it — no reason to buy more; new money does better elsewhere.
AvoidDon't own it — the problem is the business or the expectations, so a cheaper price won't fix it.

The Road Ahead

What we expect to matter in each window, and the evidence that would prove us wrong.

Short term 0-6 months

No differentiated view
Driver
"The near-term set-up is the weakest part of an otherwise strong case, and it is a momentum problem rather than a fundamental one. The stock closed 2026-08-04 at $492.81, up 1.06% on the day, having risen 19.1% over three months against SPY's 7.6% and 16.4% over six months against SPY's 11.1%. RSI is 80.6 — strongly overbought, the highest reading in this batch — and MACD is +19.45. Spot sits 22.5% above the 50-DMA of $402.21 and 13.7% above the 200-DMA of $433.35, a gap that historically does not persist. It is still 9.1% below the 52-week closing high of $542.07 and 39.7% above the low of $352.83. The twelve-month return remains NEGATIVE at −5.97% against SPY's +24.3%, a 30-point deficit, so the recovery has undone a drawdown rather than produced a gain. There is no print for 85 days. A pullback toward the $430-450 area would be entirely ordinary and would not damage the thesis."
What we’re watching
"The 200-DMA at $433.35 as the first meaningful support on any retracement; the 50-DMA at $402.21 as the second, and roughly the level at which a tracked source made a dated call that the stock was worth $550-600. Watch whether RSI unwinds through consolidation rather than decline. Watch also for any capital-markets activity: Microsoft reduced debt by $3.0B net in FY2026, but the scale of the lease commitments makes issuance plausible, and note that the analyst target range is unusually wide ($400 low against $680 high) for a company of this size — that dispersion is itself a signal that the capex question is unresolved."
Confidence
Medium

Medium term 6-24 months

Tailwind
Driver
"The medium-term case rests on the single strongest disclosure in this batch: commercial remaining performance obligation rose 84% to $678 billion, with a weighted average duration of approximately 2.3 years and roughly 30% expected to convert to revenue within twelve months. Against annual revenue of $331.839B, that is more than two years of contracted commercial business already signed. Consensus models revenue of $389.567B in FY2027 (+17.4%, 32 analysts) and $465.859B in FY2028 (+19.6%, 34 analysts), with adjusted EPS of $19.631 and $23.254. At $492.81 that is 25.1x and 21.2x — not a demanding multiple for a business earning a 33.2% return on equity with 67.9% gross margins. The mechanism that converts backlog into share price is straightforward: revenue arrives on contract, the depreciation from the 2025-2026 capital programme is already in the cost base by then, and the operating leverage that consensus does not currently model becomes visible."
What we’re watching
"The Intelligent Cloud operating margin as the FY2026 capital programme depreciates — the direct test of whether $115.948B of capex was investment or expense, and the exact fear the knowledge base articulates. Watch the free-cash-flow line, which has now fallen for three consecutive years; a fourth decline would materially change the assessment. Watch the commercial backlog at each print: +84% is not repeatable, but the level and the duration are what matter. Watch the pace at which the $329.1B of not-yet-commenced leases begin, since commencement converts an off-balance-sheet commitment into on-balance-sheet finance-lease liability and depreciation. And watch the share count: diluted shares fell only 0.16% in FY2026 (7.465B to 7.453B), so the buyback is barely more than offsetting dilution."
Confidence
Medium

Long term 2+ years

Tailwind
Driver
"Over a multi-year horizon Microsoft's position is close to unassailable in its core: an enterprise productivity franchise generating $101.997B (+16.2%), a server and cloud franchise generating $129.425B (+31.5%), a professional network at $19.817B, and a search-advertising business at $15.176B — with 51.5% of revenue in the United States and 48.5% outside it. Consensus models revenue of $558.098B by FY2029 and adjusted EPS of $28.174, which at a 20-22x multiple implies $563-620 in three years. The structural argument the tracked lane makes best is that the company sells to the buyer rather than the user: enterprise software purchased through an existing commercial relationship is far harder to displace than software purchased by individuals, and the backlog quantifies exactly that stickiness. The long-horizon risk is not competitive; it is that the capital intensity of the new model permanently lowers the multiple the market pays."
What we’re watching
"Whether the asset-heavy transition is temporary or permanent — i.e. whether capital expenditure as a share of revenue falls back from 34.9% once the current build completes, or settles at a structurally higher level. Whether free cash flow re-expands above the FY2024 peak of $74.071B. Whether the equity-method artificial-intelligence investee relationship, which swung reported EPS by $1.16 between FY2025 and FY2026 and produced $6.5B of gains in FY2026 against $4.8B of losses in FY2025, remains a source of volatility or becomes a source of value. Whether the productivity franchise retains pricing power as software delivery changes. And whether the analyst target dispersion ($400 to $680) narrows, which would itself signal that the capex debate has resolved."
Confidence
Medium

Exponential Potential

Exponential Potential7/10High

"Rated 7 — genuinely high, located in cloud infrastructure and in the contracted backlog rather than in the mature software franchises. The case for the 7: an infrastructure line growing 41% at scale; a commercial backlog that grew 84% in a single year to more than two years of total company revenue; the operating leverage of a business whose gross margin is 67.9% applied to a revenue base compounding 17-20%; and a distribution position across essentially every large enterprise on earth, which the tracked lane describes as 'every enterprise runs on it.' The case against a 9: Microsoft is executing a deliberate transformation from an asset-light licensing model into an asset-heavy infrastructure operator, and asset-heavy businesses do not compound exponentially — property, plant and equipment has gone from $154.552B (FY2024) to $229.789B to $337.253B in two years, capital expenditure is 34.9% of revenue, and depreciation from that base is a permanent new claim on the income statement. Consensus reflects it: adjusted EPS growing ~13.6% next year against revenue growing 17.4%. Additionally, at $3.66 trillion the arithmetic of size applies. Exponential in backlog and in the cloud line; increasingly industrial in structure."

“The only thing that matters on macro is AI; hyperscaler cloud/AI demand keeps outstripping supply, and the adoption phase drives exponential revenue growth plus higher margins from less hiring.”
Jordi Visserconviction 85
“Microsoft's ~$13B annual AI revenue could reach ~$130B (10x) in about four years if current year-on-year growth trend continues.”
Dwarkeshconviction 60

What could take this further than the base case — and where the market may be underpricing it. Full forward-growth and acceleration math in Deeper analysis, §4 below.

Deeper analysis

Technicals, fundamentals, valuation, and the full expert-claim evidence panel — the detail behind the numbers above.

Check our number Market-implied growth ≈ 20%/yr This isn’t how we calculate fair value — it’s public math you can verify yourself. To justify today’s $514, earnings would have to compound roughly 20% a year for 10 years (9% discount rate). Analysts forecast ~20%/yr, so the market is pricing in about what the Street expects.

Reference table

Street consensus$541.88 (+10.0%); high $680, low $400, median $531.50. 0 strong buy / 66 buy / 16 hold / 0 sell — not one sell rating
Valuation25.1x FY2027E consensus EPS of $19.63 · 21.2x FY2028E · 28.5x trailing ADJUSTED EPS of $17.28 · 27.5x trailing GAAP EPS of $17.95 · 8.28x book · 11.03x sales
The forward bookCommercial remaining performance obligation $678B, +84%, weighted duration ~2.3 years, ~30% recognised within 12 months — more than two years of revenue already contracted
Corrected balance sheetVendor reports net debt of $107.873B. Counting the $55.908B of short-term investments the vendor discards, it is $51.965B. And $88.514B of the $128.808B "total debt" is finance leases, not bonds — true bond debt is $40.294B (Section 4)
Off balance sheet$329.1 BILLION of data-centre leases NOT YET COMMENCED, running fiscal 2027-2033, terms 1-20 years. In no ratio anywhere
Free cash flowDECLINED three years running: $74.071B → $71.611B → $66.987B, while net income rose from $88.136B to $133.749B. Capex $115.948B = 34.9% of revenue
ConvictionModerate-High — 244 entity-level KB claims, 35 sources; 132 bullish / 49 neutral / 63 bearish. The lane documents the underperformance in real time (Section 7)
Technicals12-month −5.97% vs SPY +24.3% (−30 points); 3-month +19.1% vs SPY +7.6%. RSI 80.6. 22.5% above the 50-DMA

What the experts actually said 202 traceable claims on MSFT · showing the highest-conviction voices

“The only thing that matters on macro is AI; hyperscaler cloud/AI demand keeps outstripping supply, and the adoption phase drives exponential revenue growth plus higher margins from less hiring.”
Jordi Visserbullishconviction 852025-08-03
“Agrees with Ackman that Microsoft is dislocated; stock near 400 is worth closer to 550-600, with Nadella one of the best CEOs and enterprise Azure/AI tailwinds.”
Quarantine Misattributedbullishconviction 702026-05-23
“Hyperscalers are a top AI winner: if intelligence is log of compute, whoever supplies the most compute wins; Azure is one of the few at-scale.”
Dwarkeshbullishconviction 902025-02-19dwarkesh-4GLSzuYXh6w:ddfa071bd1
“Azure's core future is 'token factories' — a heterogeneous infra fleet optimized via software for TCO/utilization; the compute TAM is enormous.”
All-Inbullishconviction 852026-01-21all_in-5nCbHsCG334:83614825bc
“AI revenue ($13B, Copilot) is top of first inning; as workloads move to cloud, Azure growth accelerates to 40-45%.”
Compound And Friendsbullishconviction 852025-01-31compound_and_friends-CQCA0iLGOxY:12f73078d8
“Microsoft's Bitcoin-treasury proposal would fail — proxy-voting firms follow board recommendations ~97% of the time, and there's no upside for the board yet.”
Natalie Brunellneutralconviction 902024-12-20

Every claim reconciles to a real claim_id in the Synthos knowledge base — this is the evidence the verdict is built on, not vibes. Management (the company itself) is shown but half-weighted; one cautionary voice is included on purpose.

Price & moving averages 12 months · 50 & 200-day averages · 52-week range

338393447502557Aug '25Nov '25Jan '26Apr '26Jun '26Aug '2652w hi $542Price 514200-DMA 43150-DMA 43152w lo $353

Solid line = price · dashed line = 50-day average · dotted line = 200-day average · the two thin horizontal lines mark the 52-week high and low. Price above both averages is an uptrend.

Data summary: last close $513.53, 19% above the 50-day average ($431), 19% above the 200-day average ($431) — an uptrend. 5% below the 52-week high of $542, 46% above the 52-week low of $353.

Bollinger Bands 20-day average ± 2 standard deviations

316380444508572Aug '25Nov '25Jan '26Apr '26Jun '26Aug '26Price 51420-day avg 493

The shaded band widens when the stock gets more volatile. Riding the upper edge = strong momentum (sometimes stretched); the lower edge = weak / potentially oversold.

Data summary: price $513.53 is currently at/above the upper band (stretched) (band $475–$512).

RSI (14) momentum gauge · 0–100

705030Aug '25Nov '25Jan '26Apr '26Jun '26Aug '26RSI 73.1

Above 70 (overbought zone, shaded) = overbought, below 30 (oversold zone, shaded) = oversold. Currently 73.

MACD 12 / 26 / 9 · trend & momentum

0Aug '25Nov '25Jan '26Apr '26Jun '26Aug '26signal 20.0MACD 18.7

The MACD line crossing above the signal line (bars flip to the up color) = momentum turning up; crossing below (bars flip to the down color) = turning down. Bar height = the size of that gap.

Data summary: MACD is currently below its signal line by 1.23, negative momentum.

Relative performance vs S&P 500 & its sector (XLK (sector)), set to 100 a year ago

6386109132155Aug '25Nov '25Jan '26Apr '26Jun '26Aug '26XLK (sector) 139S&P 500 119MSFT 101

Solid = MSFT · dashed = S&P 500 · dotted = XLK (sector). A rising line means it is beating that benchmark — the sector line shows whether it is a leader or laggard within its own group.

Forward revenue & earnings actual → estimate · "FY" = fiscal year, "E" = estimate

0246491737983$254BFY24EPS $12$279BFY25EPS $13$329BFY26EEPS $17$391BFY27EEPS $20$467BFY28EEPS $23$562BFY29EEPS $29$656BFY30EEPS $35$870BFY31EEPS $45

Darker bars = actual results, brighter = analyst estimates. Taller bars to the right = expected growth.

Key stats an RIA wants

Price$513.53
Market cap$3,813B
P/E trailing29×
P/E FY26E / FY27E30× / 26×
EV / Sales11.6×*
EV / EBITDA18.6×*
Gross margin67.9%
Net margin40.3%
Dividend yield0.71%
Beta1.099
52-wk range$353 – $542
RSI(14)55
50 / 200-DMA$431 / $431
12-mo return+1% (SPY +19%)
Street target$548 ($400–$690)
Analyst grades66 Buy · 16 Hold · 0 Sell
FMP ratingA-
Next earnings"2026-10-28 — Q1 FY2027 earnings. Vendor consensus EPS $4.67 on revenue $90.541B. 85 days away; Q4 FY2026 was reported 2026-07-29 and the FY2026 10-K filed the same day, so the full fiscal year is in the data."

* Enterprise value recomputed in-house: the data vendor nets cash but omits short-term investments, overstating EV for cash-rich balance sheets. EV multiples marked * use market cap + total debt − cash − short-term investments.

1. What the business actually is

The FY2026 10-K describes Microsoft as "a technology company committed to making digital technology and artificial intelligence available broadly", generating revenue "by offering a wide range of cloud-based solutions, content, and other services to people and businesses; licensing and supporting an array of software products; delivering relevant online advertising to a global audience; and designing and selling devices." It reports in three segments — Productivity and Business Processes, Intelligent Cloud, and More Personal Computing — and separately discloses revenue by product line.

Revenue by product line, FY2026 (from seg_prod, with two vendor label defects corrected):

LineFY2026FY2025GrowthShare
Server products and cloud services$129.425B$98.435B+31.5%39.0%
Microsoft 365 commercial products and cloud services$101.997B$87.767B+16.2%30.7%
Gaming / Xbox$21.790B$23.455B−7.1%6.6%
LinkedIn$19.817B$17.812B+11.3%6.0%
Windows$17.084B$17.314B−1.3%5.1%
Search advertising$15.176B$13.878B+9.4%4.6%
Microsoft 365 consumer products and cloud services$9.175B$7.404B+23.9%2.8%
Dynamics products and cloud services$9.006B$7.827B+15.1%2.7%
Enterprise services$8.260B$7.760B+6.4%2.5%
Other products and services$0.109B$0.072B—0.03%
Total$331.839B$281.724B+17.8%

The lines sum exactly to reported revenue, so this table is arithmetically clean. Two vendor label defects are corrected here. First, the vendor renders "Microsoft 365" as "Microsoft Three Six Five" in both years — a text-extraction artefact, not a real label. Second, the vendor changes the FY2025 label "Gaming" to the FY2026 label "XBOX", which makes it appear that one line vanished and another appeared; they are the same line and the year-over-year comparison is valid. We use the filing's category names throughout.

The 10-K's own summary of FY2026 versus FY2025 is more informative than the product table and we quote it directly:

Segment operating income, FY2026 versus FY2025 (from the 10-K's MD&A): total operating income "increased $26.7 billion or 21%"; Productivity and Business Processes "increased $14.1 billion or 20%"; Intelligent Cloud "increased $12.4 billion or 28%"; More Personal Computing "increased $220 million or 2%". The profit growth is concentrated in exactly the two segments where the revenue growth is, and the consumer-facing segment is effectively flat.

Geography, FY2026: United States $170.794B (51.5%), non-US $161.045B (48.5%). The split has been close to even for a decade and the vendor table is clean and consistent.

Headcount: 228,000 (profile). No contradicting figure was located in the 10-K text extract.

2. Two earnings numbers, and which one to use

Microsoft reports both GAAP and adjusted results, and the difference is entirely one investment. The 10-K states:

> "Adjusted net income and adjusted diluted earnings per share ('EPS') are non-GAAP financial measures. These non-GAAP financial measures exclude net gains and losses from investments in [a strategic artificial-intelligence partner]."

and quantifies it:

> "Current year net income and diluted EPS were positively impacted by net gains from investments ... which resulted in an increase in net income and diluted EPS of $5.0 billion and $0.67, respectively. Prior year net income and diluted EPS were negatively impacted by net losses ... which resulted in a decrease in net income and diluted EPS of $3.6 billion and $0.49, respectively."

The other-income footnote adds the gross figures: "Other income (expense), net included $6.5 billion of net gains and $4.8 billion of net losses for fiscal years 2026 and 2025, respectively ... primarily net recognized gains (losses) on our equity method investment ... The net gains recorded for fiscal year 2026 primarily relate to the dilution gain from the ... Recapitalization."

The vendor payload carries both bases without labelling either, which is a genuine trap. The inc_a/inc_q blocks are GAAP; the earn_cal block is adjusted. They disagree in three of four quarters:

Quarter (fiscal)inc_q GAAP diluted EPSearn_cal reported EPSDifference
Q1 FY26 (2025-09-30)$3.72$4.13−$0.41
Q2 FY26 (2025-12-31)$5.16$4.14+$1.02
Q3 FY26 (2026-03-31)$4.27$4.27$0.00
Q4 FY26 (2026-06-30)$4.80$4.74+$0.06
FY2026 total$17.95$17.28+$0.67

The reconciliation is exact. $17.95 − $17.28 = $0.67, precisely the per-share impact the 10-K attributes to the investment gains. earn_cal is the adjusted basis; inc_a/inc_q are GAAP; the filing settles which is which, and the vendor never says.

This matters for the multiple. Analyst estimates are on the adjusted basis — FY2026 consensus epsAvg was $17.004 against adjusted actual $17.28 (a beat) and GAAP actual $17.95. Therefore every forward multiple in this dive is computed on the adjusted basis, and we state the comparison explicitly:

BasisFY2026 EPSMultiple at $492.81
GAAP diluted$17.9527.5x
Adjusted (the 10-K's own measure, and the estimate basis)$17.2828.5x
FY2025 adjusted comparison$14.13—
Adjusted growth FY2025 → FY2026+22.3%(GAAP growth was +31.6% — flattered by a $1.16 swing)

Half of Microsoft's reported GAAP earnings growth last year was a swing in the valuation of one private investment. Adjusted growth of 22.3% is still excellent; GAAP growth of 31.6% is not the number to underwrite.

3. Growth, margins and the transition

Fiscal yearRevenueGrowthGross marginOperating incomeOp. marginCapexCapex/revenue
FY2021$168.088B—68.9%$69.916B41.6%——
FY2022$198.270B+18.0%68.4%$83.383B42.1%——
FY2023$211.915B+6.9%68.9%$88.523B41.8%$28.107B13.3%
FY2024$245.122B+15.7%69.8%$109.433B44.6%$44.477B18.1%
FY2025$281.724B+14.9%68.8%$128.528B45.6%$64.551B22.9%
FY2026$331.839B+17.8%67.9%$155.237B46.8%$115.948B34.9%

Read that table left to right and the entire debate is visible. Revenue growth accelerated to +17.8% and operating margin reached a record 46.8% — the business is performing. But gross margin has slipped 190 basis points from the FY2024 peak of 69.8%, capital expenditure has quadrupled in three years, and capex as a share of revenue has gone from 13.3% to 34.9%. Property, plant and equipment has risen from $154.552B (FY2024) to $229.789B (FY2025) to $337.253B (FY2026).

The quarterly series shows the same thing at higher resolution:

QuarterRevenueYoYGross marginOperating incomeOp. marginGAAP diluted EPS
Q4 FY24 (2024-06-30)$64.727B—69.6%$27.925B43.1%$2.95
Q1 FY25 (2024-09-30)$65.585B—69.4%$30.552B46.6%$3.30
Q2 FY25 (2024-12-31)$69.632B—68.7%$31.653B45.5%$3.23
Q3 FY25 (2025-03-31)$70.066B—68.7%$32.000B45.7%$3.46
Q4 FY25 (2025-06-30)$76.441B+18.1%68.6%$34.323B44.9%$3.65
Q1 FY26 (2025-09-30)$77.673B+18.4%69.0%$37.961B48.9%$3.72
Q2 FY26 (2025-12-31)$81.273B+16.7%68.0%$38.275B47.1%$5.16
Q3 FY26 (2026-03-31)$82.886B+18.3%67.6%$38.398B46.3%$4.27
Q4 FY26 (2026-06-30)$90.007B+17.7%67.2%$40.603B45.1%$4.80

Gross margin has declined in each of the last four quarters, from 69.0% to 67.2% — 180 basis points — and operating margin from 48.9% to 45.1%. That is the depreciation arriving. Revenue growth has held remarkably steady at 16.7-18.4% throughout. The transition is visible, it is orderly, and it is being paid for out of growth rather than out of profit.

Quarterly depreciation and amortisation in the vendor file is internally erratic — $13.061B (Q1 FY26), $9.198B (Q2), $10.167B (Q3), $6.108B (Q4) — which does not fit a business whose fixed-asset base grew 47% during the year. The four quarters sum correctly to the annual $38.534B, so we use the annual figure and flag the quarterly split as unreliable (Section 8). The 10-K states the depreciation policy without a change of estimate: "servers and network equipment, two to six years"; internal-use software, three years.

Consensus forward path (analyst counts in brackets): revenue $389.567B FY2027 (32), $465.859B FY2028 (34), $558.098B FY2029 (21), $655.797B FY2030 (14), $869.908B FY2031 (17). Adjusted EPS $19.631 (29), $23.254 (25), $28.174 (10), $35.463 (4), $44.85 (5). The FY2030 and FY2031 EPS rows rest on four and five analysts respectively and are not used for anything. Note also that FY2031 revenue implies +32.6% growth over FY2030, an outlier inconsistent with the rest of the curve.

4. Balance sheet — three separate corrections, all material

Correction one: netDebt discards short-term investments. The vendor reports netDebt of $107.873B, which is total debt ($128.808B) less cash and cash equivalents only ($20.935B). It silently discards $55.908B of short-term investments. Counted properly:

> Cash and short-term investments $76.843B − total debt $128.808B = net debt of $51.965B, not $107.873B.

The vendor overstates net leverage by 108%.

Correction two — and this one is more important: 69% of the "debt" is finance leases, not bonds. Decomposing the FY2026 balance sheet:

ComponentFY2026FY2025FY2024
Short-term debt$9.227B$2.999B$8.942B
Long-term debt$31.067B$40.152B$42.688B
Bond debt subtotal$40.294B$43.151B$51.630B
Finance-lease obligations, current$9.678B$8.596B$5.929B
Finance-lease obligations, non-current$78.836B$60.437B$40.293B
Finance-lease subtotal$88.514B$69.033B$46.222B
Reported "total debt"$128.808B$112.184B$97.852B

Microsoft's bond debt has FALLEN from $51.630B to $40.294B over two years while its finance-lease obligations have nearly DOUBLED from $46.222B to $88.514B. The cash-flow statement corroborates: net debt issuance was −$3.000B in FY2026 and −$8.962B in FY2025 — Microsoft has been repaying bonds. The leverage growth is entirely data-centre leasing. Any ratio that treats the $128.808B as homogeneous corporate debt is mis-describing the risk, and debtToEquityRatioTTM of 0.291 and netDebtToEBITDATTM of 0.520 should be read with that in mind.

Correction three — the number that is in no ratio at all. The 10-K:

> "As of June 30, 2026, we had additional leases, primarily for datacenters, that had not yet commenced of $329.1 billion, with some arrangements subject to certain contractual conditions being met. These leases will commence between fiscal year 2027 and fiscal year 2033 with lease terms of 1 year to 20 years."

$329.1 billion. That is 2.6 times the entire reported on-balance-sheet debt load, 4.3 times the cash and short-term investment balance, and it appears nowhere in the vendor payload. It is the single largest undisclosed-by-the-vendor obligation in this batch. It is not debt in the legal sense, and much of it is contingent on conditions being met — but it is a contracted commitment to pay for data-centre capacity running out to fiscal 2033, and it is the counterweight to the $678B of contracted revenue on the other side of the ledger. The two numbers should always be quoted together.

Enterprise value. km_ttm.enterpriseValueTTM reports $3,767.259B. Rebuilt (market cap $3,659.385B + total debt $128.808B − cash and short-term investments $76.843B) it is $3,711.351B, a +1.5% residual — small enough that we accept the vendor figure for order-of-magnitude purposes while noting that it embeds the net-debt error. evToEBITDATTM of 18.15x is arithmetically defensible; we do not lean on it because the EBITDA denominator includes the equity-method gains discussed in Section 2.

Other balance-sheet facts, FY2026: total assets $758.376B (from $619.003B); goodwill $119.651B and intangibles $18.609B; deferred revenue $72.965B (from $64.555B); total equity $442.387B; retained earnings $328.265B. Book value per share $59.55, so the stock trades at 8.28x book; tangible book per share $40.94. Current ratio 1.23. Interest coverage 50.9x. Return on equity 33.2%, return on invested capital 20.6%, return on capital employed 26.3%.

Contingencies are immaterial. The 10-K discloses accrued aggregate legal liabilities of $553 million, with adverse outcomes "approximately $400 million in aggregate beyond recorded amounts" reasonably possible, plus a long-running European data-protection matter concerning the professional-network subsidiary that remains under appeal. For a company of this size these are rounding errors and no valuation adjustment is made.

5. Cash flow — the number that fell three years running

Fiscal yearOperating cash flowCapexFree cash flowBuyback (cash flow)DividendsNet debt issuance
FY2023$87.582B$28.107B$59.475B$22.245B$19.800B−$2.750B
FY2024$118.548B$44.477B$74.071B$17.254B$21.771B+$0.575B
FY2025$136.162B$64.551B$71.611B$18.420B$24.082B−$8.962B
FY2026$182.935B$115.948B$66.987B$22.271B$26.445B−$3.000B

Operating cash flow grew $46.8 billion in FY2026 — a genuinely exceptional result, which the 10-K attributes to "an increase in cash received from customers and a decrease in cash used to pay income taxes." And free cash flow still fell, for the third consecutive year. Every dollar of the operating improvement and more went into property and equipment.

The 10-K's investing discussion adds a detail worth having: cash used in investing rose $66.9B to $139.5B, "primarily due to a $51.4 billion increase in additions to property and equipment and a $22.2 billion increase in cash used in other investing primarily to facilitate the purchase of components." That second item is $22.2B of component pre-purchasing that does not appear in the capex line — supply-chain commitment on top of the reported capital programme.

Capital returns. Dividends of $26.445B (a $3.56 annual rate, 0.72% yield, 19.8% payout) and buybacks. Note a reconcilable discrepancy: the cash-flow statement shows $22.271B of common stock repurchased, while the 10-K's share-repurchase note states "During fiscal years 2026 and 2025, we repurchased 36 million shares and 31 million shares of our common stock for $16.7 billion and $13.0 billion, respectively, through our share repurchase program" and adds that the table "excludes shares repurchased to settle employee tax withholding related to the vesting of stock awards." Both figures are correct: $16.7B is the discretionary programme, and roughly $5.6B is tax-withholding settlement. We use $16.7B as the measure of discretionary buyback and say so. As of 2026-06-30, $40.6 billion remained of a $60 billion authorisation approved in September 2024.

The share count barely moves. Diluted shares: 7.469B (FY2024), 7.465B (FY2025), 7.453B (FY2026) — a 0.16% reduction last year. Against $12.405B of stock-based compensation, the buyback is close to a treadmill. Total shareholder yield is therefore roughly 1.2% (dividend 0.72% plus ~0.46% net buyback), the lowest meaningful return of capital in a decade.

6. Valuation — priced in or room?

At $492.81 (market cap $3,659.4B):

FY2026AFY2027EFY2028EFY2029E
Revenue$331.839B$389.567B (32)$465.859B (34)$558.098B (21)
Revenue growth+17.8%+17.4%+19.6%+19.8%
Adjusted EPS$17.28$19.631 (29)$23.254 (25)$28.174 (10)
Adjusted EPS growth+22.3%+13.6%+18.5%+21.1%
P/E on adjusted EPS28.5x25.1x21.2x17.5x
GAAP diluted EPS$17.95———
P/E on GAAP27.5x———
Price to book8.28x———
Price to sales11.03x———
Free cash flow yield1.83%———
Dividend yield0.72%———

Peer context. The vendor peer set is largely unusable: Alphabet ($4,570.4B), Apple ($4,544.0B), NVIDIA ($5,133.4B) and Oracle ($419.7B) are legitimate comparisons; Fortinet ($123.5B), DigitalOcean ($15.1B), GoDaddy ($11.8B), SPS Commerce ($2.7B) and Rapid7 ($0.7B) are not remotely comparable to a $3.7 trillion enterprise-software and cloud franchise. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn.

The vendor's scoring model rates Microsoft A− with an overall score of 4 — the highest quality rating in this batch, with return-on-equity and return-on-assets sub-scores of 5 and a price-to-earnings sub-score of 3 (i.e. not flagged as expensive, unlike most of this batch).

6a. What today's price assumes (the inversion)

At $492.81 — 25.1x FY2027 consensus adjusted EPS of $19.631 — the price embeds roughly the following falsifiable claims:

6b. The return bridge (why the multiple moves)

Expected return over 12-24 months decomposes as: adjusted EPS growth (+13.6% in FY2027, +18.5% in FY2028 on consensus) + multiple drift (we assume a modest HOLD-to-slight-EXPANSION, from 25.1x to ~28x forward) + shareholder yield (~1.2%: 0.72% dividend plus ~0.46% net buyback).

Our base case assumes the multiple broadly holds and drifts modestly higher, and we should be explicit that this is a re-rating of sentiment rather than of fundamentals. The argument for it is not that Microsoft deserves a higher multiple than history — it is that the multiple already compressed. The stock fell 34% in the first quarter of 2026 and remains down 5.97% over twelve months while adjusted earnings grew 22.3%; that is definitionally a de-rating that has already happened. The FY2026 10-K's disclosure of an 84% increase in contracted backlog is precisely the kind of evidence that reverses a sentiment de-rating, and the three-month +19.1% move suggests the market has begun to price it.

The majority of the base-case return is still earnings growth. From $492.81, if adjusted EPS reaches the FY2027 consensus of $19.631 and the multiple simply holds at 25.1x, the price is $493 — flat. The base case of $550 therefore requires 28.0x, i.e. roughly three turns of expansion. That is the fragile leg and we name it as such. The cross-check that makes it tolerable: $550 is only 23.6x FY2028E, and it sits 1.5% above a street consensus of $541.88 built by 82 covering analysts.

The bull case at $667 requires 34x FY2027E, close to the street high of $680. That needs operating margin to stabilise and free cash flow to re-expand, and is not the central expectation.

6c. Variant perception (where we differ, what would surprise)

Synthos fair values

All three anchors are multiples of FY2027 consensus adjusted EPS of $19.631 (29 analysts), cross-checked against FY2028E of $23.254 (25 analysts).

Base is 11.6% above spot; asymmetry roughly 1.7:1 to the upside (20% down, 35% up). The destination supports ownership; the RSI of 80.6 argues against buying it all today. Buy — Tactical, staged.

7. Knowledge base — 244 claims that explain the last twelve months

Tracked knowledge-base claims naming Microsoft: 275 raw hits, of which 244 are entity-level. Search covered MSFT (case-sensitive), Microsoft, Azure, the chief executive's name and the assistant product name, across all 51,928 records.

Discards and weighting. 31 hits were thesis- or category-level only and are excluded from the entity count. Fourteen claims tagged to a quarantined-misattribution source are excluded outright under the standing attribution policy. Twenty-six management-lane claims (skill 0.5) are half-weighted and identified wherever cited. Only 3 of the entity-level hits matched on the cloud-product term rather than the company name, so contamination from theme-level claims is low here — unusually clean for a name of this prominence.

Stance mix: 132 bullish, 49 neutral, 63 bearish across 35 distinct sources, spanning 2013-01-16 to 2026-08-01. At 25.8% bearish this is a genuinely contested name, and the bearish claims cluster tightly in the first half of 2026 — exactly the period of the underperformance.

The lane's most valuable property is that it explains the chart. These four claims, in sequence, are the twelve-month story:

> 2026-03-31 · compound_and_friends · neutral · conviction 60

> "Mag 7 drove 75% of Q1 S&P losses (Microsoft −34%) as investors front-run the asset-light to asset-heavy capex transition compressing margins."

> 2026-04-10 · compound_and_friends · neutral · conviction 68

> "Mag 7 valuations compress despite record earnings — market fears datacenter capex converts to depreciation and crushes profitability in 5-10 years."

> 2026-04-10 · compound_and_friends · bearish · conviction 55

> "Retail can prop up small/retail-driven names but not trillion-dollar stocks; institutional secular selling in Microsoft leaves it with no bid."

> 2026-04-03 · compound_and_friends · neutral · conviction 70

> "Beneath a flat index, name-brand mega-caps (Microsoft, Oracle, Meta) are in deep drawdowns while leadership rotates to lesser-known utilities, industrials, and memory-chip names."

That is a coherent, dated, correct diagnosis of a 34% drawdown, recorded while it was happening. It is also, crucially, a claim about multiple compression driven by a fear about future depreciation — not a claim that the business was deteriorating. The FY2026 annual report shows the business did not deteriorate.

The bull lane, and one dated price call:

> 2026-07-03 · compound_and_friends · bullish · conviction 78

> "Sticking with Microsoft — 34% ROE, $78B on the balance sheet, 25% earnings growth quarter after quarter; being down 23-24% year-to-date makes no sense."

> 2026-05-20 · anthony_pompliano_show · bullish · conviction 75

> "Microsoft, near 400, is worth closer to 550-600; every enterprise runs on it, [the chief executive] among the best CEOs as it moves to [cloud and AI]."

> 2026-07-25 · jordi_visser (skill 2.0) · bullish · conviction 74

> "Not worried about hyperscaler debt — low debt/equity, ~$2T of RPOs and ~$2T of compute already contracted ...; demand is already there and CapEx keeps accelerating."

> 2026-08-01 · anthony_pompliano · bullish · conviction 72

> "[The four largest cloud providers'] backlog of contracted orders nears/exceeds $2 trillion; they raised CapEx with orders to match — I'm not worried about them despite widening CDS."

The 2026-05-20 claim is directly testable and is now half-right. It was made with the stock "near 400" and named $550-600 as fair value. The stock is now $492.81 — 23% higher — and our own base case of $550 sits at the bottom of that named range. A dated, specific, falsifiable call that has partially played out is worth more than a dozen general endorsements, and it is an input to the verdict.

The bear lane's strongest current claim is about timing, not thesis:

> 2026-07-28 · Michael Batnick · bearish · conviction 55

> "Bought Microsoft, Workday and ServiceNow ~3 weeks ago but doesn't love the setup into the call — it's limping into the report, uninspiring; likely to sell at the open."

That was written the day before the print that has since driven the stock up. We record it because honesty requires recording the misses as well as the hits, and because it illustrates the limit of near-term positioning claims.

> 2026-04-06 · forward_guidance · bearish · conviction 80

> "The four software Mag 7 names are troubled; Microsoft is a 'disaster' (lost the [AI-assistant] battle) — their stocks reflect it."

> 2026-04-12 · jordi_visser (skill 2.0) · bearish · conviction 70

> "Magnificent Seven names already had their run and can't deliver the multiples investors need."

Note that the single highest-skill voice in this file (skill 2.0) is bearish in April and bullish in July, on the same reasoning that the contracted backlog resolves the debt concern. That is a legible change of mind against evidence, and we treat it as such.

Management-lane claims, half-weighted. Twenty-six claims from the chief-executive lane (skill 0.5) describe custom silicon economics, on-device inference, quantum research and agent architectures. These are company statements about company products and carry no independent evidentiary weight; none is used as an input to the valuation. They are noted only to be complete about what the lane contains.

Why Moderate-High rather than High. The lane is deep (244 claims, 35 sources, thirteen years), unusually clean (only 3 theme-level product matches), and it contains a dated falsifiable price call that is playing out. But it is 25.8% bearish, the bear arguments are specific rather than lazy, and the most recent bearish claim was made one day before the print that changed the picture. Positive, not overwhelming.

8. Data integrity — what we rejected from the vendor file and why

Eight items. Three are direct filing-versus-vendor contradictions, and in every one the filing wins.

1. FILING OVERRIDES VENDOR — the earnings basis is ambiguous and the vendor never labels it. The vendor carries GAAP diluted EPS in inc_a/inc_q ($17.95 for FY2026) and adjusted EPS in earn_cal ($4.13 + $4.14 + $4.27 + $4.74 = $17.28), with no indication that the two blocks are on different bases. The 10-K resolves it: "Adjusted net income and adjusted diluted EPS ... exclude net gains and losses from investments in [a strategic artificial-intelligence partner]", and the FY2026 impact was "an increase in net income and diluted EPS of $5.0 billion and $0.67." $17.95 − $0.67 = $17.28, exactly. We use adjusted EPS of $17.28 for all valuation work, because that is the basis analyst estimates are built on, and we state both.

2. FILING OVERRIDES VENDOR — the vendor payload contains no record of $329.1B of lease commitments. The 10-K discloses "additional leases, primarily for datacenters, that had not yet commenced of $329.1 billion ... between fiscal year 2027 and fiscal year 2033 with lease terms of 1 year to 20 years." This is 2.6x the entire reported debt load and appears in no vendor field and no leverage ratio. We report it prominently and pair it with the $678B of contracted revenue.

3. FILING OVERRIDES VENDOR — the buyback figure has two correct answers and the vendor gives only one. The cash-flow statement reports $22.271B of common stock repurchased in FY2026. The 10-K states "we repurchased 36 million shares ... for $16.7 billion ... through our share repurchase program" and notes the table "excludes shares repurchased to settle employee tax withholding." We use $16.7B as the discretionary buyback and identify the ~$5.6B remainder as tax-withholding settlement, which is not a return of capital in the same sense.

4. netDebt ignores $55.908B of short-term investments — a 108% overstatement. Vendor netDebt of $107.873B is total debt ($128.808B) less cash and cash equivalents only ($20.935B). Correctly computed: net debt of $51.965B. enterpriseValueTTM of $3,767.259B carries the error; rebuilt it is $3,711.351B, a +1.5% residual — small enough to accept for scale but noted.

5. "Total debt" of $128.808B is 68.7% finance leases and should not be read as corporate borrowing. Bond debt is $40.294B ($9.227B short-term plus $31.067B long-term) and has fallen from $51.630B in FY2024; finance-lease obligations are $88.514B and have nearly doubled from $46.222B. The cash-flow statement confirms net debt issuance of −$3.000B (FY2026) and −$8.962B (FY2025). debtToEquityRatioTTM (0.291) and netDebtToEBITDATTM (0.520) are reported but explicitly qualified, not rejected.

6. Quarterly depreciation and amortisation in inc_q is internally erratic — quarterly D&A REJECTED, annual accepted. The FY2026 quarterly series reads $13.061B, $9.198B, $10.167B, $6.108B — a declining pattern in a year when the fixed-asset base grew 47% (from $229.789B to $337.253B). The four quarters sum correctly to the annual $38.534B, so we use only the annual figure and do not compute any quarterly cash-flow or margin metric from the quarterly D&A line. The 10-K discloses no change in useful lives; the stated policy remains "servers and network equipment, two to six years."

7. Vendor segment labels are defective in two ways — corrected against the filing. The vendor renders "Microsoft 365" as "Microsoft Three Six Five" (a text-extraction artefact), and relabels FY2025's "Gaming" as FY2026's "XBOX", which superficially reads as one line disappearing and another appearing. They are the same line; we use the filing's category names and treat the year-over-year comparison as valid. The product lines sum exactly to reported revenue in both years.

8. Two minor internal inconsistencies, flagged. (a) quote.yearHigh of $553.72 and yearLow of $349.20 are intraday extremes; tech.hi52 of $542.07 and tech.lo52 of $352.83 are closing extremes. We use the closing series throughout. (b) The insider file contains a null row — Hogan Kathleen T (officer) with a blank transaction type, zero shares, zero price and zero securities owned — which is a corrupt record and is excluded from Section 10.

Not defects, correctly reported: freeCashFlowYieldTTM of 1.83% and priceToFreeCashFlowRatioTTM of 54.6x look wrong for a company of this quality but are correct, and reflect the genuine three-year decline in free cash flow documented in Section 5. est EBITDA and EBIT rows pass the usual sanity checks for this file (net income never exceeds EBITDA in any row, and EBITDA always exceeds EBIT) — an uncommon outcome, and worth stating, because roughly 70% of files in this feed fail that test. We still use epsAvg exclusively for forward valuation.

Non-equity tripwire: checked and clear. MSFT is common stock, beta 1.13, average volume 38.8M shares, dividend variable and rising ($3.56 trailing), and a 52-week closing range of $352.83-$542.07 — a 54% band. Nothing resembling a preferred or baby-bond profile.

9. Technicals — the best three months and the worst twelve

10. Insiders — routine compensation mechanics, nothing else

DatePersonRoleActionTypeSharesPriceHeld after
2026-07-15Amy ColemanEVP, Chief Human Resources OfficerDisposalF-InKind (tax withholding)32$384.9345,413
2026-06-15Alice L. JollaChief Accounting OfficerAcquisitionA-Award5,004$076,153
2026-06-15Amy ColemanEVP, Chief Human Resources OfficerDisposalF-InKind (tax withholding)36$390.7445,445
2026-06-11Emma N. WalmsleyDirectorAcquisitionA-Award (RSU)3$01,263
2026-06-11John D. RaineyDirectorAcquisitionA-Award (RSU)1$0296
2026-06-11Penny S. PritzkerDirectorAcquisitionA-Award (RSU)33$014,323
2026-06-11Mark MasonDirectorAcquisitionA-Award (RSU)1$0232

One further record — Kathleen T. Hogan, officer, with a blank transaction type, zero shares, zero price and zero securities owned — is a corrupt row and is excluded (Section 8).

The honest reading is that this table contains no signal. Every acquisition is a zero-price grant of restricted stock units; every disposal is an in-kind tax withholding of 32 and 36 shares. Not one insider bought or sold a share on the open market in the period covered. The three-share and one-share director awards are dividend-equivalent accruals, not decisions.

The absence is worth one sentence, though: through a period in which the stock fell to a closing low of $352.83 and recovered to $492.81, no officer or director sold discretionarily. That is weak evidence and we weight it as such — but the alternative, heavy selling into the recovery, would have been meaningful, and it did not occur.

11. Verdict, kill-criteria and flip conditions

Buy — Tactical.

The business case is strong enough to own. The entry point is poor enough to stage.

What the FY2026 annual report established, six days ago. Commercial remaining performance obligation of $678 billion, up 84% — more than two years of contracted revenue against $331.839B of annual sales, with a 2.3-year weighted duration and 30% recognisable within twelve months. Cloud revenue up 27% to $214.4B; the infrastructure line up 41%. Segment operating income up $26.7B or 21%, concentrated in the two cloud-facing segments. Adjusted EPS up 22.3% to $17.28. Return on equity 33.2%, return on invested capital 20.6%, gross margin 67.9%, operating margin at a record 46.8%. The vendor's own quality model rates it A−, the highest in this batch, and 82 covering analysts produce zero sell ratings.

What you are accepting, stated without softening. Free cash flow has fallen three years running — $74.071B, $71.611B, $66.987B — while net income rose from $88.136B to $133.749B. Capital expenditure is 34.9% of revenue, up from 13.3% three years ago. Gross margin has fallen in each of the last four quarters, from 69.0% to 67.2%. There is $329.1 billion of data-centre lease commitment that has not yet commenced, running to fiscal 2033. And Microsoft's own risk factors state that these investments "are in advance of fully developed revenue streams" and that "the associated revenue may not be realized in the expected timeframes or at expected levels." Our knowledge base spent the first half of 2026 arguing exactly this, and it was right about the stock for nine months.

Why Tactical rather than Core. Base fair value of $550 is 11.6% above spot and sits within 1.5% of a street consensus of $541.88 — we do not have a variant perception on the destination, only on the earnings basis, the balance-sheet composition, and the entry. RSI is 80.6 after a 19.1% three-month move, and the price is 22.5% above its 50-day average. Those two facts together mean the right action is to establish a position and hold capital back for a retracement, not to complete the purchase today. A staged entry is Buy — Tactical by definition.

Pre-registered KILL criteria — what would break the thesis:

Pre-registered FLIP TO BUY — CORE (upsize) criteria:

Where MSFT fits in the Synthos Framework Portfolio. The mega-cap platform sleeve, at a 2-3% target weight built in three tranches over 4-8 weeks — roughly one third now, one third on any move back toward the 200-DMA at $433.35, one third reserved. The house standing rule that mega-cap incumbents are guilty until proven innocent on a forward basis applies, and Microsoft's defence is the strongest available: an 84% increase in contracted backlog to $678 billion is forward evidence, not trailing performance. Note also that it is the only mega-cap in this batch that has underperformed the index over twelve months while growing adjusted earnings 22.3% — the combination the flagship-selection philosophy is designed to find.

On overlap within the batch. Microsoft and Alphabet (also covered 2026-08-04) are both mega-cap platforms funding very large infrastructure programmes, but they are not redundant and they resolve differently. Alphabet has risen 97.6% over twelve months, has stopped repurchasing stock, has issued $30.5B of equity and $19.0B of preferred, has negative free cash flow, and carries $85.2B of uncommenced leases — we rate it Hold at roughly 7% base upside. Microsoft is down 5.97% over twelve months, is still repaying bond debt, retains positive if declining free cash flow, and discloses $678B of contracted backlog against $329.1B of uncommenced leases — we rate it Buy — Tactical at 11.6% base upside. If only one mega-cap platform is held, on today's prices it should be this one. Logged as a tracked Synthos call (Buy — Tactical) as of 2026-08-04 at $492.81.

Single biggest risk: that the depreciation from the capital programme arrives faster than the revenue it was built for. Capital expenditure of $115.948B in one year, $329.1B of leases not yet commenced, gross margin already down 180 basis points in four quarters, and the company's own risk factors conceding that the spend is "in advance of fully developed revenue streams." The $678B backlog is the best available counter-evidence, but backlog is a promise of revenue, not of margin — a contract can be honoured at a price that does not cover the cost of the capacity built to serve it.

Most fragile assumption in the price: that operating margin holds near 45-47%. It was 48.9% two quarters into fiscal 2026 and 45.1% by the fourth quarter. Every element of the valuation — the 25.1x forward multiple, the consensus EPS path, the base case of $550 — assumes that slide arrests. If margin instead continues at the four-quarter rate, FY2028 earnings would fall materially short of the $23.254 consensus and the multiple would compress at the same time.

Provenance & disclosures