Phase Space AI

03 Valuation and Factor Scorecard

Amazon [AMZN]

Amazon.com, Inc. [AMZN] — Valuation, Consensus Bridge & Factor Scorecard

Task 3 · as of 2026-07-27 · spot $231.39 · market cap $2,488bn · EV ~$2,569bn (incl. leases)

Model: AMZN_Underwriting_Model_2026-07-27.xlsx — live formulas, verified by reading computed cells back through Excel via AppleScript with all workbooks closed first.

---

Updated 2026-07-29 — Criteria + two-horizon valuation. The numbered Gates are retired and replaced by named Criteria with types (BINDING / MEASURED) returning PASS / FAIL / INDETERMINATE. This memo no longer issues a position verdict — no Long, Short, Watchlist or Avoid. The expected-return-versus-cash-hurdle test is replaced by a reverse-DCF implied-path test plus a 12-month target, with sensitivity run over the exit multiple rather than over scenario probabilities. Momentum is entry timing only and vetoes nothing. Sections below that predate this update are retained as the historical record and are annotated where superseded.

Reading the retired "Gate N" numbering. Any Gate N below is historical. The mapping is: Gate 1 → Quality Criteria (BINDING) · Gate 1b → Short Mechanism Criteria (MEASURED) · Gate 2 / 2A / 2B / Path B → dissolved, absorbed by the Valuation Criteria · Gate 2C → Peer Spread Criteria (MEASURED) · Gate 3 → Catalyst Criteria (MEASURED) · Gate 4 / 4a → Valuation Criteria (BINDING) · Gate 5 → Liquidity Criteria (BINDING) · Gate 6 → Momentum Criteria (MEASURED, entry timing only) · (new) Downside Criteria (MEASURED). A Criteria returns PASS / FAIL / INDETERMINATE; a missing input is INDETERMINATE, never FAIL. Historical calibration items keep the old numbering on purpose — the record is the record.

0. Two-horizon valuation — methodology update, 2026-07-29

This section supersedes the expected-return-versus-cash-hurdle arithmetic in §2 below (retained as the historical record and annotated). Per ~/.claude/skills/investment-memo/references/valuation.md, every name carries two valuation outputs and both must be reported: a 12-month target and the implied-path test (reverse DCF).

No position verdict is issued. This document outputs an analysis.

0.1 Inputs — from the pre-computed scan, with two corrections stated

Inputs are read from the pre-computed coverage record reports/scan/AMZN_analysis.json (as_of 2026-07-28, status: OK, notes: []) rather than recomputed. Two figures are corrected, and both corrections are shown rather than silently applied.

⚠ Correction 1 — the scan's "TTM revenue" is not a trailing twelve months

coverage_scan.ttm_revenue() sums "the last four quarterly XBRL periods." Issuers do not file a discrete Q4 — the 10-K reports the full year, so no Q4 duration fact exists. The last four filed quarterly periods are Q1, Q2 and Q3 of one year plus Q1 of the next: Q4 is skipped and a twelve-month-old Q1 substituted for it.

Amazon is the worst-affected large name in this batch because its December quarter is its largest. The scan's window omits Q4-2025 = $213,386m — the single biggest quarter the company has ever reported — and substitutes Q1-2025's $155,667m. Correct TTM to 2026-03-31: $742,776m, 8.4% above the scan's $685,057m.

Same failure class as the last-FY error already recorded (which understated MU by 136% and SNDK by 76%), and live across all 129 scanned names.

⚠ Correction 2 — the scan's valuation: PASS embeds a terminal margin Amazon does not have

The record carries required_cagr_pct: 4.9, valuation_margin_pp: 6.8, valuation: "PASS". Inverting the scan's own solve (reverse_dcf.solve(..., unknown="terminal_margin", cagr=0.049)) recovers the terminal EBIT margin it used: 20.04%. Amazon's operating margin, in the same record, is 11.2%.

On AMZN the hardcoded 20% flatters rather than penalises — it assumes Amazon's operating margin very nearly doubles, which is precisely the disputed question on the name. 20% is the hardcoded constant assets/coverage_scan.py's own docstring records as a defect ("a hardcoded 0.20 put MCK, CI, COR, CAH and UNH at the top of the ranking with NEGATIVE required growth"). The documented rule is terminal_margin = max(own current margin, sector peer median); no terminal_margin field appears in any of the 129 records, so the run that produced them pre-dates that rule.

This is the clearest illustration in the batch of why the constant matters: it turns MSFT's genuine PASS into a FAIL and AMZN's FAIL into a PASS, in the same run, purely on where each company's real margin sits relative to 20%.

0.2 The implied-path test — the Valuation Criteria

Parameters held fixed: terminal EBIT margin 11.2% (Amazon's own FY2025 operating margin — the floor of the documented max(own, sector peer median) rule), exit multiple 22.7x EV/EBIT, WACC 10.0%, horizon 5 years, EV $2,485,052m, TTM revenue $742,776m. Solved for revenue CAGR.

Result: FAIL, and the reversal against the scan record is the point of this update.

The corrected revenue base helps AMZN (a higher base lowers the required rate); the corrected terminal margin hurts it far more, and the net is a flip from PASS to FAIL. The scan's PASS was, in substance, the statement "Amazon is cheap if its operating margin nearly doubles to 20%" — which is not a valuation finding but an assumption about the single most contested variable on the name.

What a FAIL here does and does not mean. It means today's price requires revenue growth 4.6pp above what Amazon has demonstrated over three years, holding the margin at today's level. It does not mean the price is unjustifiable: holding the margin fixed is conservative for a business whose margin has risen 6.4% → 10.8% → 11.2% over three years. Per criteria.md this could qualify as PASS WITH ARGUMENT only on a specific, evidenced margin path — and the memo's own Base case (9.1% terminal margin) is below today's level, not above it. On the memo's own numbers the argument is not available, so the result stands as FAIL rather than being upgraded on narrative.

0.3 Sensitivity — over the EXIT MULTIPLE, never over scenario probabilities

The prior version ran its range across scenario probabilities (30/50/20). That is the defect this replaces. Terminal margin held at 11.2%, WACC 10.0%, 5 years:

The sign flips at roughly 27.5x EV/EBIT. AMZN passes only if it exits at or above ~27.5x — that is, only if the market five years out still pays close to the 30.0x it pays today, for a business whose growth has decelerated. That is the whole question, and it is a multiple question, not a probability question. The prior version's range over scenario weights could not have located it.

0.4 Anchoring — growth-matched, and what was discarded

The 22.7x anchor is the median EV/EBIT of the 31 scanned names whose growth brackets AMZN's 11.7% (range 6.7%–17.4%). No distant-year haircut is stacked on it. The §4 comp table is retained as descriptive context and is not used to anchor — it is the AI-capex cohort, selected on shared narrative exposure rather than growth, and a set with no growth dispersion cannot identify a growth/multiple relationship.

0.5 The 12-month target

Amazon's own EV/Sales history — 47 monthly observations, 2022-09 to 2026-07. The window starts after the 20-for-1 split of June 2022: prices are split-adjusted but as-filed XBRL share counts are not, so pre-split months mix the two bases and are excluded. That exclusion is a disclosed limitation on window length. Share counts, net cash and TTM revenue are as-filed XBRL, point-in-time by filing date, with the Q4 correction applied. No look-ahead.

Near-term revenue: corrected TTM $742,776m grown at +14.2%, the rate corrected TTM revenue has actually grown year over year, held flat → NTM revenue $848,386m. A held-flat actual, not a consensus estimate: Alpha Vantage EARNINGS_ESTIMATES was unavailable, its shared 25-request daily quota exhausted. Disclosed limitation.

12-month target: $291 (+25.9%), with a floor of $263 (+14.1%) and a downside case at the own 25th percentile of $238 (+3.0%). Amazon's multiple is the least dislocated of the five names updated today — the 32nd percentile of a comparatively tight four-year range — so the target is mostly growth rather than re-rating.

The tension worth stating explicitly: the 12-month target is above spot while the implied-path test FAILS. That is not a contradiction; it is the two horizons doing their jobs. Over twelve months Amazon plausibly compounds revenue in the mid-teens and mean-reverts inside its own recent multiple range. Over five years the same price requires 16.3% revenue growth at an unchanged margin, which is 4.6pp above what it has demonstrated. A name can be reasonably priced for a year and demanding for five, and reporting only one of those was the defect this two-output method exists to fix.

Named product-cycle events inside the window (dated in the Catalyst Criteria): Q2-2026 earnings, and quarterly AWS growth and segment-margin disclosure. No date is asserted beyond what §6 of the Catalyst Calendar already sources.

Sanity band. Street consensus is $312.19 across 36 analysts, range $230–$370 — +35.3% to spot, implying 3.97x NTM EV/Sales, between today's 3.35x and the trailing-12-month 4.30x. The Street is paying a fuller re-rating than this analysis. For the record, the prior house output was a probability-weighted $116.81, −49.5% to spot, with a ticket target of $207 — both far below spot and below the Street's entire published range, consistent with item B16 (16 of 16 house targets below spot, median 46.1% below Street). Both are superseded.

---

1. Model output (verified readback)

FY2025A ties to the 10-K exactly: revenue $716,924m ✓, segment D&A $41,860m ✓, operating income $79,994m vs reported $79,975m (a $19m rounding residual on the margin input, 0.02%).

The single most important line in the table. EBITDA margin rises from 17.0% to 21.2% — genuine operating leverage. Operating margin falls from 11.2% to 9.6%. Depreciation consumes more than all of the operating leverage, rising from 5.8% to 11.6% of revenue. That is not a forecast about demand; it is the arithmetic of $1.15 trillion of capital spending meeting a five-to-six-year useful life.

---

2. Valuation

SUPERSEDED 2026-07-29. The expected-return-versus-cash-hurdle arithmetic below is retired. It is retained as the historical record. The live valuation output is the implied-path test plus the 12-month target in §0, and no position verdict is issued.

2.A DCF — and an honest statement about its information content

This DCF has essentially no information content, and the memo says so rather than presenting the number as a target. When terminal value exceeds 100% of enterprise value — because every explicit forecast year has negative free cash flow — the output is a leveraged function of two unobservable inputs (WACC and terminal growth) and tells you nothing you did not put in. Reporting "$14.20" as a price target would be false precision of the worst kind.

Assumptions used: WACC 10.5% (CoE 10.78% from beta 1.45 measured against SPY on 251 daily returns, rf 4.25%, ERP 4.5%; after-tax cost of debt 4.0%; D/(D+E) 4.6%), terminal growth 3.5%, maintenance capex 8.0% of revenue, SBC treated as a real cost throughout.

2.B Reverse DCF — what today's price actually requires (the informative version)

Solving for the terminal operating margin that makes the DCF equal the $231.39 spot:

At an 8.5% discount rate — generous for a stock with a measured beta of 1.45 — today's price requires Amazon to earn a 26.6% consolidated operating margin in perpetuity. Amazon's best-ever consolidated operating margin is 11.2%, achieved in FY2025. AWS standalone earns 35.4%. The price therefore requires the entire company — including a retail business earning 6.9% in North America and 2.9% internationally — to converge on AWS-like profitability.

Even at a 6.5% WACC, which implies almost no equity risk premium, the requirement is 16.5% — still ~47% above the all-time high. This is the valuation conclusion, and it is unambiguous: Amazon is expensive on any defensible set of cash-flow assumptions.

(The sentence that followed here previously converted this observation into an Avoid / Watchlist label. That verdict is removed — the memo issues none.) It does not, by itself, support a Short. See Task 5.

2.C Scenarios (from the model's Scenarios tab, verified)

Probability-weighted value per share: $116.81. Scenario-weighted E[R]: −49.5%. Excess over the 4.7% cash hurdle: −54.2%.

Base-rate check on each scenario (Task 2 discipline, reproduced on the Scenarios tab): - Bear (30%) — AI capex digestion. AWS decelerates to mid-single digits by FY2029 while the depreciation from $700bn+ of cumulative capex still lands. Precedent: AWS 2022–23, when growth halved and segment margin fell ~700bp. - Base (50%) — AWS holds mid-teens terminal growth; retail margin expands ~150bp on genuine automation; depreciation absorbs roughly half the EBITDA gain. This assumes deceleration, with the base rate. - Bull (20%) — Trainium/Bedrock take real share, AWS compounds >20% through FY2029, Leo and advertising add a third profit pool. This requires the above-base-rate assumption that hyperscale growth does not decelerate at scale. Named as such and weighted at 20%.

2.D Consensus — sourced, not asserted

Alpha Vantage was exhausted; consensus was sourced from public aggregation per references/consensus-bridge.md.

On the specific metric the Gate 1 mechanism would move — AWS revenue growth and consolidated operating margin — the house forecast and the Street are within 1%. This is a Gate 2A failure, stated plainly.

2.E The consensus bridge — numbers versus multiple decomposition (required)

House probability-weighted value $116.81 vs Street target $312.19 — a gap of −63%. Decomposed:

The disagreement is almost entirely about the discount rate and terminal multiple, not about the business. That is the textbook signature of a valuation-only view — and references/trade-construction.md is explicit that a valuation-only view cannot carry a directional position.

The ±15% EPS dispersion is explained. Revenue dispersion is ±3.1%; EPS dispersion is ±15%. The difference is almost certainly whether an analyst includes the Anthropic mark. Q1-2026 carried $16.8bn of pre-tax Anthropic gains — roughly $1.17 of the reported $2.78 EPS. An analyst who includes it and one who excludes it are not forecasting different businesses; they are reporting different numerators. Anyone comparing a house EPS to "consensus EPS" without establishing which convention is in use is comparing noise.

---

3. Factor & Anomaly Scorecard

All computed from data already wired: Alpaca SIP daily bars (251 trailing returns to 2026-07-27) and SEC EDGAR XBRL/R-files.

3.1 Reading the scorecard honestly

The scorecard does NOT corroborate a short mechanism, and Gate 1 requires that it be said explicitly.

Per references/trade-construction.md: "A Short whose scorecard shows clean earnings quality and strong profitability needs correspondingly stronger narrative evidence to pass this gate, and the memo must say so explicitly rather than ignoring the disagreement."

Amazon's accruals are −0.076 — deeply negative, meaning cash flow massively exceeds accounting earnings. That is the opposite of the high-accrual profile where shorts empirically work. F-score of 6 is fine. Gross profitability is above median. Three of four earnings-quality factors argue against a short.

Only asset growth (+27.5%) points bearish, and it points there hard — this is a well-documented anomaly and Amazon is at an extreme of it. But one factor against three does not clear the raised bar this reference file demands for a tech short.

Momentum (Gate 6) is neutral in every measure — 12-1 at +0.5%, price on the 200dma, flat YTD. A position in either direction would be neither with nor against the tape. Gate 6 therefore passes trivially, and provides no support.

---

4. Comparable companies

Trailing reported figures from EDGAR XBRL. NTM consensus multiples are not shown because Alpha Vantage was exhausted and I will not present unsourced forward multiples as if they were sourced. Fiscal-year ends differ and are stated.

¹ Alphabet's FY2025 figures were not retrievable under the standard XBRL tag at time of writing; FY2024 is shown and labelled. Flagged rather than silently substituted.

Reading: Amazon carries the lowest operating margin and among the highest EV/EBIT in the mega-cap set. That is a coherent market view — investors are paying for AWS and treating retail as a call option — but it means the multiple is doing a great deal of work. Per the reference file, this is not evidence of anything on its own: the whole group may be re-rating for a shared structural reason, which is a sector-regime question, not a stock-specific thesis.

---

5. Football field

The methods do not agree, and the spread is itself the finding. Cash-flow-based methods (DCF, scenarios) put Amazon at a fraction of its price. Multiple-based methods put it roughly at its price. The entire valuation therefore rests on whether one accepts the market's prevailing multiple as the right anchor.

Per references/trade-construction.md, that is exactly the condition under which valuation must NOT be allowed to drive a directional call. The house price target for the trade ticket is set at $207 — the midpoint of the ex-Anthropic P/E method, which is the most defensible bridge between the cash-flow and multiple approaches — implying ~11% downside. That is a "rich, not shortable" conclusion, not a thesis.