Microsoft

Phase B — Measure performance (re-run)

Lens 5A · THE CAPEX VERDICT

The primary figure, stated exactly.

"For calendar year 2026, we expect the mix of short-lived assets to remain similar to Q3. We expect to invest roughly $190 billion in capital expenditures, which includes approximately $25 billion from the impact of higher component pricing." — Amy Hood, CFO, Microsoft FY2026 Q3 earnings call, 2026-04-29 [primary: transcripts/2026-q1.md]

Its exact definition, on all three axes:

The three numbers for one quarter — this is the whole problem in miniature. Fiscal Q3 FY26 (Jan–Mar 2026):

You cannot reconstruct Microsoft's headline capex from the cash-flow statement. The naive sum overstates it by $3.0B in this quarter and understates it in others. Two reconciling items: (a) accrual-vs-cash timing on goods received, (b) an unreconciled ~$0.7B/quarter difference between Hood's stated finance-lease figure and the XBRL ROU-additions tag (9M FY26: Hood $22.5B vs XBRL $19.486B — a persistent $3.0B gap, flagged as an open item).

Why "$120B" exists — three routes, all arithmetically defensible, none of them calendar-2026.

Route 1 — the analyst's half-year residual (most likely origin). On the same call where Hood said $190B, the next question was:

"Maybe, Amy, could you elaborate a little bit on the CapEx guidance you just provided? Obviously, it requires a fairly material pickup in CapEx in the second half of the calendar year, maybe to the tune of $120 billion." — Karl Keirstead, analyst, Microsoft FY26 Q3 earnings call, 2026-04-29 [primary: transcripts/2026-q1.md]

The arithmetic: CY2026 H1 = fiscal Q3 FY26 actual $31.9B + fiscal Q4 FY26 guided ">$40B" ≈ $72B. Residual for Jul–Dec 2026 = $190B − $72B ≈ $118B. The "$120B" is a six-month figure, said out loud on the call, quoted since as if it were the year.

Route 2 — calendar-2025 actual (the base, not the guide). Management-definition capex, four calendar quarters of 2025:

Cross-check: CNBC reported the $190B guide as "up 61% from 2025". $190B / $118.0B = 1.610 → +61.0%. The base reconciles exactly. A source quoting "$120B+" is, on this route, quoting last year.

Route 3 — fiscal 2026, cash-only. The route the fiscal-vs-calendar hypothesis predicts — but it only lands on $120B if you strip both axes:

So the fiscal-vs-calendar axis alone does NOT explain the gap. Fiscal 2026 including leases is ~$144B — nowhere near $120B. It takes fiscal year and cash-only together to get there.

THE VERDICT. Microsoft's 2026 capex is ~$190B, and that figure is calendar year 2026, capital expenditures including finance leases, on an accrual basis, as guided by the CFO on 2026-04-29. The "$120B+" figure is not a competing estimate of the same quantity — it is (most likely) the H2-CY2026 residual an analyst named on the call, or the CY2025 base, or a fiscal-year cash-only cut. All three land at $113–118B, which is exactly why they are indistinguishable once the definition is dropped. The $70B "dispersion" in the big-four aggregate is not a disagreement about Microsoft — it is an undeclared-definition error. For any aggregate that sums management-guided, calendar-year, lease-inclusive capex, Microsoft's line is $190B.

The trap to avoid on the way out. There is a primary-source Microsoft number near $190B that has nothing to do with 2026 capex: $196.6B of leases signed but not yet commenced as of 2026-03-31, commencing across fiscal 2026 through fiscal 2031 [primary: 10-Q, Note 12 — Leases]. It is a five-year commitment stock, not an annual flow. Anyone citing "$190B" without saying which one is 50/50 to be citing the wrong one.

Desk rule this establishes: every capex figure in the KB and in any aggregate must carry (period basis, lease treatment) — e.g. MSFT $190B (CY2026, incl. finance leases, guided). Without both tags the number is not verifiable.

One more forensic consequence — Microsoft's reported FCF excludes finance-lease capex entirely. Reported free cash flow is OCF minus cash PP&E additions only:

[primary: XBRL 10-Qs; Hood quotes from transcripts]

On a lease-adjusted basis, 9M FY26 economic free cash flow is closer to ~$27.9B, a 41% haircut to the reported figure. Only ~$4.0B of finance-lease payments touched the cash statements in the period (operating cash flows from finance leases $1.829B + financing cash flows from finance leases $2.179B) [primary: 10-Q, Note 12]. As the finance-lease mix moves, reported FCF moves for reasons that have nothing to do with the capital actually deployed — Hood said as much in Q2: "free cash flow was $5.9 billion and decreased sequentially, reflecting the higher cash capital expenditures from a lower mix of finance leases."

---

Lens 5B · THE BACKLOG — what is disclosed and what genuinely is not

Disclosed (primary):

The RPO time series — the single most useful table for 07-29 [primary: XBRL RevenueRemainingPerformanceObligation, 10-K/10-Q]:

The entire OpenAI commitment landed in one quarter. Since then the backlog has grown $2B — 0.3% sequential. Hood's framing is that commercial RPO ex-OpenAI grew 26% "in line with historic seasonality," and that the OpenAI contract "represents multiyear demand needs from them, which will result in some quarterly volatility in both bookings and RPO growth rates going forward." Both readings are on the record; the 07-29 sequential print is the tiebreak.

Confirmed still undefined (the KB flag is correct on two of three):

• Cancellability — NOT disclosed. The 10-Q says only that estimating RPO "can involve significant judgments, including identifying and assessing variable consideration and potential renegotiation of commitments" and that commercial RPO is "reported net of reserves." No termination-rights disclosure, no minimum-commitment/take-or-pay split, no penalty structure. Microsoft nowhere states what fraction of the $627B is legally non-cancellable.
• Customer concentration beyond OpenAI — NOT disclosed. Only the OpenAI ~45% figure exists. A second large counterparty is now visible but unquantified: Hood attributed FY26 Q2 bookings growth partly to "the previously announced Anthropic commitment from November" [primary: FY26 Q2 call], with no dollar figure. No other named counterparty, and no Azure-only RPO breakout.
• Duration — NOW RESOLVED. ~2.5 years weighted average, ~25% inside 12 months. This flag can be closed in the KB.
• The "$80B unfulfillable Azure backlog" — no primary corroboration; see Corrections. If the KB carries it, it should be demoted to `` or dropped.

---

Lens 6 · Earnings calls (sentiment trend, re-run on three now-on-disk transcripts)

Tone across FY26 Q1 (2025-10-29) → Q2 (2026-01-28) → Q3 (2026-04-29):

• Q1: capex framed as accelerating — "we now expect the FY '26 growth rate to be higher than FY '25." Finance leases spiked to $11.1B of a $34.9B quarter (32%) while cash PP&E was only $19.4B — the highest lease-mix quarter in the series.
• Q2: the OpenAI contract lands; RPO steps $398B→$631B; commercial bookings +23% cc. Capex $37.5B, lease mix drops to 18%. FCF collapses to $5.9B — and Hood explicitly attributes it to lease mix, not spend.
• Q3: lease mix drops again to 15% ($4.7B of $31.9B); cash paid for PP&E ($30.9B) now runs ahead of accrual capex as prior accruals settle. The $190B CY2026 guide is introduced. Language shifts decisively to supply: "constrained at least through 2026," "demand continues to exceed supply," with Azure growth expected to "show modest acceleration in the second half of the calendar year."

The phrase that changed: Q1/Q2 discussed balancing supply across first-party vs Azure demand; Q3 introduced a forward supply-constraint horizon with a date attached ("at least through 2026") and paired it with a specific reacceleration claim (H2 CY2026). What is new and quantified: AI business annual run rate $37B, +123%; Copilot paid seats surpassed 20 million, +250% YoY; first-party model efficiency claims (MAI Transcribe One, 67% GPU efficiency gain; MAI Image Two, up to 260%) and Maia 200 "with over 30% improved tokens per dollar compared to the latest silicon in our fleet" now live in Iowa and Arizona — i.e. the COGS-reduction lever is being disclosed with numbers for the first time.

OpenAI agreement mechanics, as management describes them post-restructure: royalty-free IP access through 2032; revenue share running through 2030 with Microsoft's rev-share to OpenAI eliminated. Hood: "having the revenue share exist through 2030 — the predictability of that is a real positive for us — and, as Satya pointed out, the IP being royalty-free with the elimination of our rev share to them." [primary: FY26 Q3 call]

---

Lens 8 · Stock-price catalysts (updated)

• 2026-04-29 (FY26 Q3): beat on revenue and EPS, Azure +40%, stock −~3.9% on the $190B CY2026 capex guide vs ~$154.6B Visible Alpha consensus.
• 2026-07-06: 4,800 job cuts announced (~2.1% of workforce; ~1,600 Xbox).
• Price: $393.82 at 2026-07-20 — effectively flat vs $393 at the prior dossier (2026-06-18), but −~23% across H1 CY2026. Traded below both its 50-day ($398) and 200-day ($426) EMAs at the July 6 close of $386.74.
• Sell-side: consensus "Strong Buy"; average price target ~$558 (58 analysts, S&P Global). The gap between an unchanged spot and a ~$558 average target is itself a datapoint about how much of the de-rate the Street has accepted.

Unchanged read: this name trades on the Azure growth rate and the capex/FCF trajectory, not on the headline beat.

---

Lens 11 · Forward arithmetic (mechanics only — NOT a projection)

Presented as arithmetic on management's own guide plus published consensus. No base/bull/bear paths and no target are set in this file — that is Connor's, after his own read.

• Q4 FY26 guided revenue $86.7–87.8B, +13–15% [primary: Hood, FY26 Q3 call]. Segment guides sum to the same range: PBP $37.0–37.3B (+12–13%), IC $37.95–38.25B (+27–28%), MPC $11.75–12.25B.
• Q4 FY26 consensus diluted EPS $4.21, vs $3.65 a year ago.
• Nine-month FY26 non-GAAP diluted EPS $13.14 [primary: 10-Q filed 2026-04-29]. Adding consensus Q4 → FY2026 non-GAAP diluted EPS ≈ $17.35.
• Guided Q4 adjusted tax rate ~19%; other income/expense ex-OpenAI ~−$100M; ~$900M of one-time retirement costs in Q4; FY26 operating margin expected up ~1 point YoY [primary: FY26 Q3 call].
• FY27 opex growth guided mid-to-high single digits with headcount down YoY [primary: FY26 Q3 call] — the July 6 layoff is the first execution of that.

No forecast.ts created — logging a Brier-scored EPS call would be taking the directional position this refresh is explicitly withholding.

---

Phase C — Judge people & books (partial re-run)

Lens 9 · Management (re-run — board changes only)

Nadella and Hood unchanged; the capital-allocation judgment is unchanged from 2026-06-18 and is carried by reference. Two governance deltas:

• Carmine Di Sibio (ex-EY global chairman & CEO) joined 2026-05-14, seated on the Audit Committee and Compensation Committee; board to 13 [primary: 8-K + Ex-99.1, 2026-05-14]. Adding a Big-Four ex-CEO to Audit at the exact moment the accounting questions are lease-classification, useful-life and equity-method-OpenAI is at minimum a well-timed appointment.
• Reid Hoffman will not stand for re-election at the 2026 annual meeting; director since 2017; explicitly "not as a result of any disagreement with management" [primary: 8-K filed 2026-06-05]. Note the AI-adjacent conflict surface a Hoffman departure removes (LinkedIn co-founder; Inflection AI co-founder; OpenAI board history).
• Capital return, 9M FY26: $13.3B of buybacks (27M shares), $20.3B of dividends declared, $44.0B remaining of the $60B authorization [primary: 10-Q filed 2026-04-29].

Lens 10 · Forensic red flags (re-run on the capex/lease axis)

Three items, all sharpened by this pass:

1. Finance leases are the single largest reporting-quality issue in this name. They are PP&E additions that never enter the cash-flow statement or reported FCF (Lens 5A). They ran $19.5B (XBRL) / $22.5B (management-stated) in 9M FY26 — and the mix is volatile quarter to quarter (32% → 18% → 15% of capex), which mechanically swings reported FCF independent of capital deployed. Watch on 07-29: the FY2026 10-K restates the full-year lease footnote; any change in finance-vs-operating lease classification policy, or in the leases-not-yet-commenced disclosure, is material.
2. Leases not yet commenced at $196.6B vs a $633B total RPO. The commitment stock is now ~31% of the contracted revenue stock, up from ~25% at Dec-2025 ($155.1B / $631B) and ~25% at Jun-2025 ($92.7B / $375B). The commitment side has been growing at least as fast as the contract side — worth tracking as a ratio.
3. Useful-life assumptions — unchanged so far, and the 10-K is the place to check. Hood's Q3 disclosure that "roughly two-thirds of our CapEx was for short-lived assets, primarily GPUs and CPUs" with the remainder "for long-lived assets that will support monetization over the next fifteen years and beyond" is the clearest depreciation-mix statement Microsoft has given. Any lengthening of server/network useful lives in the FY2026 10-K would flatter margins and must be caught on 07-29/30. No change detected through the Mar-2026 10-Q.
4. OpenAI equity-method optics — unchanged. 9M FY26 other income included $5.9B of net gains from OpenAI, primarily the Recapitalization dilution gain, vs $2.7B of net losses a year prior [primary: 10-Q]. Q3 FY26 in isolation was near-neutral ($14M net loss, minimal EPS impact) [primary: Q3 FY26 earnings release]. The non-GAAP ex-OpenAI line remains the only readable earnings series.
5. IRS NOPAs ~$28.9B (tax years 2004–2013, transfer pricing) — unchanged, contested, no resolution expected within 12 months. Still the largest quantified contingency.
6. SEC enforcement: no Litigation Releases or AAERs naming Microsoft — carried from the 2026-06-18 EDGAR EFTS scan; no new filings in the window to change it.

---

Lenses carried forward unchanged

Three questions added to Lens 14, ordered by information value:

1. Of the $627B commercial RPO, what portion is legally non-cancellable versus subject to termination or renegotiation rights — and what does "net of reserves" reserve against?
2. Finance leases were 32%, 18% and 15% of capex across the last three quarters. What drives that mix, is it a policy choice or a delivery-timing artifact, and what should investors use as the denominator for capital deployed given reported FCF excludes it entirely?
3. Leases not yet commenced went $92.7B → $196.6B in nine months. What is the annual commencement schedule across fiscal 2026–2031, and how much of the $190B CY2026 capex is lease commencement versus new cash outlay?

---

What to watch on 2026-07-29 (after close)

Ranked by how much each would change the file.

1. The CY2026 capex guide — reaffirm, raise, or cut? $190B was set on 2026-04-29 with ~$25B attributed to component pricing. Memory pricing has been the swing input. Any restatement changes the desk's whole big-four aggregate. Listen for the exact phrase: does Hood say "calendar year 2026" again, and does she still say "including finance leases"?
2. Q4 capex actual vs the ">$40B" guide, and the finance-lease split. The lease share (32% → 18% → 15%) is the number that determines whether reported FCF looks like a collapse or a recovery. Hood flagged that Q4's sequential increase includes ~$5B of component pricing "as well as the impact from finance leases."
3. Sequential RPO. $631B → $633B was +$2B. A third flat quarter says the backlog is an OpenAI artifact; a resumption of $20B+ sequential says the ex-OpenAI base is still compounding. Also watch whether Microsoft repeats the ~45% OpenAI share — it has been given once, verbally, for Dec-2025 only.
4. Azure vs the +39–40% cc guide — and whether the "modest acceleration in H2 CY2026" claim is reaffirmed, softened, or given a number.
5. Free cash flow, and whether Microsoft changes how it presents it. 9M reported FCF $47.3B; lease-adjusted ~$27.9B. If the FY2026 full-year FCF is presented without a lease-inclusive alternative, that is itself the disclosure choice worth noting.
6. The FY2026 10-K (files ~07-30, one day after the release). Three specific checks: (a) server/network useful-life assumptions — any lengthening; (b) the leases-not-yet-commenced figure and its FY2026–FY2031 commencement schedule; (c) Item 1A for any newly quantified customer-concentration or backlog-cancellability language.
7. Revenue vs the $86.7–87.8B guide and non-GAAP EPS vs the $4.21 consensus; FY26 operating margin vs the "up ~1 point" commitment; the ~$900M one-time retirement charge landing where guided.
8. Whether the AI run-rate ($37B, +123%) and Copilot paid seats (20M, +250%) are repeated. These are voluntary disclosures — the quarter they stop being given is the quarter to pay attention.

---