Phase Space AI

Microsoft Valuation Analysis 2026-07-27

Microsoft [MSFT]

Microsoft Corporation [MSFT] — Valuation Analysis

Task 3 · investment-memo skill v1.4.0 · 2026-07-27 · spot $389.10

Model: Microsoft_Financial_Model_2026-07-27.xlsx. Every figure below was read back from the live workbook in Excel after all other workbooks were closed; formulas were not trusted because openpyxl wrote them without error.

---

1. DCF

Sensitivity (live grid in the workbook):

(The workbook's grid recomputes the full explicit period via SUMPRODUCT at each WACC; the $357.11 headline uses the base 8.5%/3.5% cell of the main bridge, which applies mid-year discounting to the same cash flows.)

Honest caveat: at 82.8% terminal-value weight this DCF is a terminal-multiple statement dressed as a cash-flow model. It is reported because the method requires it, not because it is the load-bearing evidence. The scenario table and the comps carry more weight.

1.1 The depreciable-life sensitivity — the input that actually matters

Depreciation is computed in this model from a two-pool vintage roll-forward seeded on the disclosed 30-June-2025 gross balances (10-Q R55: servers/network/software $132,836m; buildings/leasehold/furniture $156,445m; land $9,338m excluded as non-depreciating). Calibration check: the model returns FY2026E depreciation of $33,907m against a disclosed 9M FY2026 actual of $24,000m (R56), i.e. ~$32–34bn annualised. The engine is calibrated.

Modelled depreciation: FY2025A $21,000m → FY2026E $33,907m → FY2027E $50,343m → FY2029E $89,153m → FY2032E $154,964m. The FY2027E step-up alone is +48%.

The server life on the Assumptions tab is a live driver. Microsoft's disclosed policy is 6 years (raised from 4 in the FY2022 10-K). The Bear scenario runs it at 4.0 years — the pre-2022 policy — which is the cleanest way to express the market's actual worry.

---

2. Comps

Implied values: peer-median P/E × house FY2027E EPS of $18.32 = $421.42. Peer-median EV/Sales × house FY2027E revenue, bridged to equity = $376.24.

Sourcing honesty: MSFT's and GOOGL's FY+1 P/E use sourced consensus ($16.77 and $14.22). AMZN, META, ORCL and NVDA multiples are estimates and are flagged as such rather than presented as sourced.

The load-bearing caveat, per references/trade-construction.md: this comp set is the AI-capex cohort. Every member is re-rating for the same structural reason at the same time. MSFT trading at 23.2x against a 23.0x peer median is therefore not a stock-specific signal — it is the observation that the whole group is priced together. "If the whole group is expensive together, that is a sector-regime question, not a stock-specific short thesis." The same logic blocks a relative-value long.

---

3. Street & positioning

3.1 Consensus

Alpha Vantage EARNINGS_ESTIMATES was NOT called — the daily quota was exhausted before this session. The consensus figures above are sourced from public aggregators per references/consensus-bridge.md, which is a legitimate route and is named rather than presented as a paid feed.

3.2 House vs Street — the Gate 2A test

This is the finding, and it is decisive. With one quarter left in FY2026, a house model built independently from EDGAR actuals and company guidance lands exactly on consensus revenue and EPS. For FY2027 the house is 7.5% below the mean — but inside the published range, at the 11th percentile. At least three of 56 analysts already carry a lower number.

A number inside the consensus range is not a variant. It is agreement with the low end of consensus. The distinction matters: a variant requires a named mechanism the Street has missed, and the mechanism producing my lower FY2027 EPS (a ~150bp gross-margin decline from AI depreciation loading COGS) is the single most-discussed issue on the name — asked verbatim on the FY26Q2 call by Bernstein and answered at length by the CFO.

3.3 The consensus bridge — decomposing a 43% gap (required, references/consensus-bridge.md)

The Street target of $557.25 is 43.2% above spot. Something must reconcile that, and the decomposition is the whole story:

The gap is overwhelmingly about the multiple, not the numbers — 25 points of multiple against 7.5 points of earnings. The Street is applying roughly 28x forward earnings; the market is currently paying 23.2x FY2026E and 19.7x FY2027E. This memo does not know which is right, and says so.

The real risk to the house view if the Street is right: Microsoft delivers Azure at 39–40%, guides FY2027 capex below $190bn, discloses Copilot seats and ARPU for the first time, and the multiple re-rates from ~20x back toward its own decade-long high-20s norm. That single sequence is worth 40%+ from here and it is not far-fetched — it requires no new information beyond a good print on 29 July. Anyone bearish on MSFT at $389 is implicitly forecasting a permanent de-rating of the highest-quality large-cap franchise in the index, on a 6-year depreciation policy the company has disclosed and defended.

3.4 Short interest, ownership, and the options market

• Short interest: MSFT short interest is structurally negligible (well under 1.5% of float; days-to-cover ~1–2). The precise current FINRA bi-monthly figure could not be sourced at the time of writing and is not fabricated. Directionally: there is no short-side crowding to exploit and no squeeze risk to fear.
• Institutional ownership: ~72%, dominated by Vanguard / BlackRock / State Street index vehicles. This is passive capital, not active conviction. Any "smart money agrees" claim in either direction is weak here.
• Options market — the sharpest signal on the name. From Alpaca snapshots, 2026-07-27:

Realised volatility: 27.7% (1-year), 33.3% (3-month). The 31-Jul ATM straddle ($390 strike, call mid ~$13.6) prices roughly a ±7% move on the print, i.e. ~±$27. The front-week/September term-structure kink is a textbook event premium. Buying MSFT options into 29 July is systematically expensive — this is precisely the variance-risk-premium case references/alpaca-options.md warns about, and it is decisive for Gate 5.

---

4. Factor & Anomaly Scorecard (REQUIRED)

Direction assessed for a long, since the DCF/scenario work points at fair-to-slightly-rich rather than at a short. All inputs computed from data already wired: Alpaca SIP bars and EDGAR XBRL.

Synthesis — the factors SPLIT, and they split along an informative seam. The quality factors are unambiguously positive: 0.313 gross profitability and −0.055 accruals are excellent, and there is no earnings-quality issue anywhere in the data. The price factors are unambiguously negative: −27.8% momentum, 72% of the high, 10.4% below the 200-day. The investment factor (+20.9% asset growth) is negative and is mechanically the same thing the price factors are reacting to.

That combination — clean books, strong profitability, broken price, heavy investment — is the classic profile of a quality compounder in a capex-driven de-rating. It is not a short (accruals and F-score would have to corroborate a deterioration mechanism, and they contradict it). It is also not yet a momentum long. It is, on the factor evidence alone, a wait.

---

5. Regime-Change Test (references/regime-change-test.md)

The claim under test: AI has changed Microsoft's demand and margin structure such that the old multiple and the old growth path are the wrong anchors.

Tiered evidence

Verdict: the regime-change claim PASSES on Tier 1 evidence — and that is a real, substantive conclusion. It is also, and this is the point, not a variant, because every element of it is disclosed and quantified by management on the record.

Duration matching (Principle 3)

The evidence reaches 2.5 years from March 2026, i.e. to roughly mid-FY2029. The model does exactly that: the Intelligent Cloud contracted tranche is modelled at contracted terms through FY2029, and from FY2030 the RPO balance is allowed to decline ($745bn → $700bn → $660bn → $620bn) with the conversion rate reverting toward the pre-AI 40–45% norm. Renewal at similar terms is a Tier 4 assumption and is not made.

Two-tranche output (Principle 4) — contracted share of revenue by year

Peaks at the contract horizon, then declines. That decline is the honest expression of uncertainty and is exactly the shape Principle 4 requires.

Principle 5 shape check — does the regime change raise the FLOOR more than the ceiling?

The signature is correct. The bear case is a de-rating to 16x on a 14%-lower EPS, not a collapse — because ~$157bn of commercial RPO converts within twelve months whatever sentiment does. The contractual floor truncates the left tail. Crucially, the RPO gives the bull case no lift at all: contracted volume cannot be repriced upward into a shortage. That asymmetry is what a genuine contractual regime change looks like, as opposed to a demand narrative wearing an evidence costume.

Base rate being overridden, stated explicitly: essentially no company above $250bn of revenue has compounded

15% for five consecutive years (Chan, Karceski & Lakonishok 2003). This model does not extrapolate — total revenue growth decays from 17.0% (FY2026E) to 6.4% (FY2032E). The one override is Intelligent Cloud through FY2029, made on Tier 1 contractual evidence covering exactly those years, and no further.

---

6. Valuation summary

Every house method lands within ±8% of spot. The scenario-weighted expected return is −3.0% gross and −8.1% after the 4.7% cash hurdle and frictions. That is not a valuation gap in either direction; it is a name trading at fair value with an unusually wide distribution around it.

Reported for completeness (and it is the number to watch, not to trade): MSFT trades at 19.7x consensus FY2027 EPS, against a decade in which its forward multiple has spent most of its time in the high-20s to low-30s. Per references/trade-construction.md, a multiple below its own history is not evidence of anything on its own — the peer group is re-rating together for a shared structural reason, and the correct reading of a cheap multiple on a capex-heavy name is that the market has revised its view of the capital intensity, not that it has made an error.