Amazon [AMZN]
Task 5 · as of 2026-07-27 · spot $231.39
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 Sections below that predate this update are retained as the historical record and are annotated where superseded.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 Nbelow 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.
portfolio-book contract executedportfolio_book.json read at 2026-07-27.
| Book state | Value |
|---|---|
| Positions | 0 — the book is 100% cash |
| Watchlist | MU, ISRG, NET, SMR, NBIS |
| Max single-name weight | 5.0% |
| Cash hurdle (annual) | 4.7% |
| Max pairwise correlation without disclosure | 0.60 |
| Max sector concentration | 25% |
| Position hard stop | −2.0% of book |
| Drawdown ladder | review −5%, de-gross −10%, stop −15% |
251 trailing daily log returns to 2026-07-27, Alpaca SIP feed, adjusted closes.
| AMZN vs | ρ | AMZN vs | ρ |
|---|---|---|---|
| META | 0.47 | SNDK | 0.22 |
| GOOGL | 0.45 | NET | 0.20 |
| MSFT | 0.35 | NBIS | 0.20 |
| NVDA | 0.30 | CIEN | 0.16 |
| AVGO | 0.24 | AAOI | 0.16 |
| MU | 0.22 | ORCL | 0.15 |
| SPY | 0.58 | QQQ | 0.55 |
Mean correlation of AMZN against the concurrent AI-infrastructure underwriting set (MU, NBIS, NET, SNDK, CIEN, AAOI, MSFT, GOOGL): 0.245.
The assignment stated that "correlation is the dominant portfolio fact right now" because MU, NBIS, NET, SNDK, CIEN, AAOI, MSFT and GOOGL are all under concurrent underwriting in the AI-infrastructure complex. Measured on actual returns, that premise is wrong, and the correction matters for sizing.
The "AI-infrastructure complex" is not one correlated block. It is at least two, and they barely move together:
| Cluster | Members | Internal pairwise ρ | Cross-cluster ρ |
|---|---|---|---|
| A — Memory / optics / hardware | MU, SNDK, CIEN, AAOI, NBIS | 0.30 – 0.75 (MU–SNDK 0.75, SNDK–CIEN 0.58, MU–CIEN 0.54) | — |
| B — Hyperscalers / software | AMZN, META, MSFT, GOOGL, NET | 0.11 – 0.47 (AMZN–META 0.47, AMZN–GOOGL 0.45) | 0.00 – 0.35 vs cluster A |
The most striking single number: MSFT versus MU is −0.02, and MSFT versus SNDK is −0.03. Two names that "everybody knows" are the same AI trade are statistically unrelated over the last year.
Why: capital has been rotating out of the AI capex spenders and into the AI capex beneficiaries. Over the trailing 12-1 window MU is +919%, SNDK +4,822%, CIEN +441%, AAOI +400%, NBIS +368% — while ORCL is −39%, MSFT −27% and META −23%. The complex is not correlated; it is a transfer.
Consequence for this decision: no pairwise correlation involving AMZN exceeds the book's 0.60 disclosure threshold. The highest is AMZN–META at 0.47. Correlation is not a binding constraint on an AMZN position. The binding constraint is Gate 2, not the book.
The cash-hurdle test is retired (criteria.md: "The cash hurdle, replaced by slot competition in the
strategy"), so the paragraph below is retained as the historical record only. AMZN's scenario-weighted
E[R] is −49.5% for a long; the excess over hurdle is −54.2%. A long fails the capital-competition test
outright. A short would need to clear the gates first — it does not (§2).
1. Fundamental conclusion. Amazon is a strong business getting stronger at the operating line and weaker at the reported line. EBITDA margin expands from 17.0% to 21.2% through FY2030 on genuine leverage. Operating margin contracts from 11.2% to 9.6% because depreciation rises from 5.8% to 11.6% of revenue. Economic free cash flow is negative in every forecast year. AWS is re-accelerating (28% in Q1-2026, fastest in 15 quarters) with a 35.4% segment margin. The business is fine; the reported earnings and the cash flow are not what they appear.
2. Expectations conclusion. Consensus and the house forecast are, on the numbers, the same. Q2-2026 revenue: consensus $196.18bn, house ~$196bn. AWS: consensus ~$40.5bn, house ~$40.3bn — a 0.5% gap. FY2026 revenue growth sits inside the consensus band. Revisions are drifting positive (+2.8% over 90 days). The Street's target is $312.19, 35% above spot. ΔE = house − Street ≈ 0 on every metric that matters.
3. Valuation conclusion. Rich on cash flow, roughly fair on multiples. The reverse DCF is the sharpest statement: at an 8.5% WACC the price requires a 26.6% terminal consolidated operating margin against an all-time high of 11.2%. Probability-weighted value $116.81 versus $231.39 spot.
4. Portfolio conclusion. No position. The valuation conclusion is bearish, and it is the only bearish conclusion. There is no expectations gap for the valuation to translate. This is the NET case exactly — a tech/growth name that is expensive without being shortable.
Per ~/.claude/skills/investment-memo/references/criteria.md, every test is a named Criteria carrying a
type, returning PASS / FAIL / INDETERMINATE. BINDING = a long-only absolute-return book treats
failure as disqualifying. MEASURED = always scored and stored, informs timing or sizing, never blocks.
A missing input is INDETERMINATE, never FAIL (calibration item D1). The memo scores every Criteria and blocks on none of them — the type labels describe the strategy that consumes this analysis.
| Criteria | Type | Result | Evidence |
|---|---|---|---|
| Quality | BINDING | PASS | COMPOUNDER. FY2025 operating margin 11.2%, above the 10% archetype line and rising three years running (6.4% → 10.8% → 11.2%). Gross margin 50.3% — corrected; the scan's gross_margin_pct: 0.8 is a tag artifact, Amazon files no usable GrossProfit tag. Revenue +12.4% YoY to $716,924m FY2025. The one genuinely adverse quality reading is asset growth +27.5% ($641.6bn → $818.0bn), a documented negative predictor — and mechanically it is the capex programme itself. |
| Valuation | BINDING | FAIL | Implied path requires 16.3% revenue CAGR against 11.7% demonstrated → −4.6pp margin, at a GROWTH_MATCHED 22.7x EV/EBIT exit (31 peers, 6.7–17.4% growth), implying −24.3% compression from today's 30.0x. The sign flips at ~27.5x — AMZN passes only if it still trades near today's multiple in five years. This REVERSES the scan record, which reads PASS, +6.8pp, because that solve embedded a 20.0% terminal margin against Amazon's actual 11.2%. Not upgraded to PASS WITH ARGUMENT: the memo's own Base case terminal margin (9.1%) is below today's level, so the evidenced margin path runs the wrong way. Working: Valuation §0.1–§0.3. |
| Liquidity | BINDING | PASS | Equity: ~$2.5tn, unlimited liquidity. Options chain pulled, not assumed (Alpaca, 18-Sep-2026 expiry, strikes within ±12% of spot): 22 strikes carrying open interest, median 7,924 contracts, maximum 17,227, 175,947 in total (OI as of 2026-07-27); median quoted spread 2.9% of mid — the tightest of the five names — median quoted size 15 bid / 60 ask. Fillable at size. |
| Momentum | MEASURED | Scored — 2nd quintile | 12-1 momentum +3.1%, 35.0th cross-sectional percentile of the 129-name scan. 6-1 −1.9% (35.7th pct). RSI-14 35.5 — the weakest near-term reading of the five. 83.9% of the 52-week high; below the 200-day. Governs when to enter, never whether to own. |
| Catalyst | MEASURED | Scored | Q2-2026 earnings and the quarterly AWS growth and segment-margin disclosure. Dates as sourced in the Catalyst Calendar; no date is asserted that the sources do not support. |
| Downside | MEASURED | Scored | Permanent-loss case: AI capex digestion — AWS decelerates to mid-single-digit growth by FY2029 while depreciation from $700bn+ of cumulative capex still lands, compressing the terminal operating margin to 6.8%. Precedent named: AWS in 2022–23, when growth halved and segment margin fell ~700bp. The prior forward DCF put this at $54.63 (−76.4%); that figure came from an exit multiple that was not growth-matched and from a revenue base 8.4% too low, and should be read as the mechanism's direction rather than a calibrated price. No going-concern case: net cash +$24.0bn. |
| Consensus | MEASURED | Scored | Street $312.19, 36 analysts, range $230–$370 — +35.3% to spot. Recorded; no longer an admission test and blocks nothing. |
| Peer Spread | MEASURED | INDETERMINATE | No same-end-market peer's own multiple history and current percentile were assembled. INDETERMINATE, not FAIL. |
| Short Mechanism | MEASURED | FAIL | Requires decelerating growth and exhausted margin runway. Growth is decelerating only mildly (corrected TTM +14.2% YoY against an 11.7% three-year CAGR — in fact above it), and the margin has expanded three years running with the incremental leverage plainly not spent. Neither leg holds. |
| Sub-sector | MEASURED | Scored | Large-cap internet / hyperscale cloud + retail. AI-capex payer. |
| Was | Now | Why |
|---|---|---|
| Gate 4 (Expected return): FAIL — E[R] −49.5%, i.e. −54.2% vs a 4.7% cash hurdle | Valuation Criteria: FAIL, −4.6pp | Same direction, entirely different instrument and a far smaller number. The old figure came from a forward DCF whose base case put fair value at $119.38 against a $230.74 spot — a −48% base case is not a valuation, it is a statement that the model and the market disagree by a factor of two. The implied-path test instead asks what the price requires (16.3%) against what the business has done (11.7%). |
| Gates 1/2/3 | dissolved or re-typed | Gate 2 is absorbed into the Valuation Criteria; the catalyst is MEASURED and blocks nothing; the mechanism question splits into Quality (business) and Short Mechanism (thesis). |
| Gate 6 (Momentum) | Momentum Criteria: MEASURED, 2nd quintile, blocks nothing | Momentum is entry timing, not selection. |
| Scenario-probability sensitivity (30/50/20) | Sensitivity over the exit multiple | The range was being run on the parameter that could not change the answer. The flip point is at ~27.5x exit, which the old method could not locate. |
Scan record valuation: PASS, +6.8pp |
FAIL, −4.6pp | The scan's solve embedded a hardcoded 20.0% terminal margin against Amazon's actual 11.2%. Its PASS was in substance "Amazon is cheap if its margin nearly doubles" — an assumption about the most contested variable on the name, not a valuation finding. |
Scan gross_margin_pct: 0.8 |
50.3% | Tag artifact — Amazon files no usable GrossProfit tag. Corrected from revenue less CostOfRevenue. |
Scan revenue_ttm: $685,057m |
$742,776m (+8.4%) | The scan's "TTM" skips Q4 — on Amazon it omits the largest quarter the company has ever reported ($213,386m) and substitutes a year-old Q1. |
This memo does not output "Long", "Short", "Watchlist" or "Avoid". The prior version of this section
issued an INVESTMENT DECISION block. That verdict has been removed. The evidence is retained above.
AMZN — Amazon.com, Inc. framework: Criteria + two-horizon valuation (2026-07-29)
Spot $230.74 (2026-07-28) Archetype: COMPOUNDER
BINDING Quality ................ PASS op margin 11.2% and rising 3 years (6.4->10.8->11.2);
gross margin 50.3% (scan's 0.8% is a tag artifact - corrected)
Valuation .............. FAIL required 16.3% vs demonstrated 11.7% -> margin -4.6pp
exit 22.7x EV/EBIT, GROWTH_MATCHED (n=31, growth 6.7-17.4%)
implied compression: -7.3x / -24.3% from today's 30.0x
sign flips at ~27.5x exit - AMZN passes only if it still trades
near TODAY'S multiple in five years
NOT upgraded to PASS WITH ARGUMENT: the memo's own base-case
terminal margin (9.1%) is BELOW today's 11.2%
Liquidity .............. PASS Sep-2026 chain: 22 strikes with OI, median 7,924, total 175,947
median quoted spread 2.9% of mid - tightest of the five
MEASURED Momentum ............... 2nd quintile (12-1 +3.1%, 35.0th pctile, RSI-14 35.5) - TIMING ONLY
Catalyst ............... Q2-2026 earnings; quarterly AWS growth and segment-margin disclosure
Downside ............... AI capex digestion - AWS to mid-single-digit growth by FY2029 while
depreciation on $700bn+ of capex lands; terminal margin to 6.8%.
Precedent: AWS 2022-23 (growth halved, segment margin -700bp).
Prior forward DCF: $54.63 (-76.4%) on a non-growth-matched exit
multiple and a revenue base 8.4% too low - direction, not price.
No going-concern case; +$24.0bn net cash.
Consensus .............. Street $312.19, 36 analysts, range $230-$370
Peer Spread ............ INDETERMINATE - no peer multiple history assembled
Short Mechanism ........ FAIL (growth not decelerating vs its own 3y CAGR; margin runway unspent)
12-MONTH TARGET $291 (+25.9%) own four-year median 3.70x EV/Sales on NTM revenue $848.4bn
floor $263 (+14.1%) at today's 3.35x; downside $238 (+3.0%) at own p25
AMZN trades at the 32nd PERCENTILE of its own four-year EV/Sales range - the least
dislocated multiple of the five names updated today
Street $312.19 (+35.3%) implies 3.97x - a fuller re-rating than this analysis
NOTE ON THE TWO HORIZONS The 12-month target is ABOVE spot while the implied-path test FAILS.
That is not a contradiction - it is the two instruments doing their jobs. Reasonably
priced for a year, demanding for five. Reporting only one of those is the defect the
two-output method exists to fix.
CORRECTIONS scan valuation PASS (+6.8pp) -> FAIL (-4.6pp): the scan embedded a 20.0% terminal margin
against Amazon's actual 11.2%.
scan TTM revenue $685,057m -> $742,776m (+8.4%); scan gross margin 0.8% -> 50.3%.
prior probability-weighted value $116.81 (-49.5%) and ticket target $207 SUPERSEDED.
DISCLOSED GAPS NTM revenue is corrected-TTM YoY held flat, NOT consensus (Alpha Vantage quota exhausted).
Own-multiple window is 47 months, starting after the June-2022 20:1 split.
Peer Spread INDETERMINATE.
Would convert to LONG: 1. Price ≤ $150 (the bear-scenario zone, ~35% below spot) with AWS growth still ≥ 20% — at which point the reverse-DCF requirement falls to ~17% terminal operating margin, within reach of a mix-shifted Amazon. 2. Amazon beginning to disclose AWS RPO or contracted backlog. This would for the first time make the regime-change test applicable and could convert a Tier 3/4 story into Tier 1 evidence. 3. Economic free cash flow turning positive while capex stays above $200bn — proof the spend is earning.
Would convert to SHORT (all of the first three, plus a tape condition): 1. E1 — AWS revenue growth printing below 20% year on year in any quarter while FY-forward capex guidance stays above $200bn. 2. E2 — 30-day FY2027 EPS revisions turning net negative (currently +2.8% over 90 days). 3. E3 — Any Anthropic down-round or a reversal of the accumulated ~$32bn of marks, which would cut reported EPS by several dollars with no change in the operating business. 4. T1 (tape) — weekly close below the 200-day moving average (~$234.6) with a lower-high / lower-low sequence. Required, because Gate 6 is currently neutral and initiating a short into a neutral tape with no variant is the exact NET error.
Conditions that would materially worsen this analysis (no verdict exists to downgrade): - Q2-2026 (30-Jul) showing retail segment operating margin contracting year on year, confirming that the FY2025 improvement was substantially the heavy-equipment life extension. - Any further extension of AI server useful lives, reversing the §5 quality signal.
Required even where the decision is no position.
| Vehicle | Economics | Verdict |
|---|---|---|
| Outright short equity | Borrow ~0.3%/yr, no dividend liability, deep liquidity, unbounded loss. E[R] +49% gross on the model. | Rejected — Gate 2 fails. The E[R] is a valuation artefact, not a thesis. |
Long-dated put spread — AMZN270115P00220000 / AMZN270115P00190000 |
Debit $9.52, width $30, max profit $20.48, R:R 2.15:1, delta −0.344 / −0.167, IV 36.0% / 37.5% | Rejected. Structurally the correct vehicle if a short were warranted — defined risk, sells back the inflated wing — but it expresses a view I do not hold. |
Aug-2026 put spread into earnings — AMZN260821P00225000 / AMZN260821P00205000 |
Debit $5.53, width $20, max profit $14.47, R:R 2.62:1, IV 44.6% / 46.1% | Rejected — Gate 5 fails. IV/RV is 1.42x. Buying this is paying a 42% volatility premium for an earnings view the house does not have. |
| Long equity | — | (previously rejected on E[R] −49.5% vs a 4.7% cash hurdle — superseded, that test is retired). No verdict issued; the live result is Valuation Criteria FAIL by −4.6pp with a 12-month target of $291. |
| Cash | +4.7% risk-free | SELECTED. |
Pre-authorised conversion mechanism, documented so a future conversion is executable without re-deriving it — and NOT to be used before the trigger fires: if price reaches ~$150 with AWS growth intact, the entry vehicle would be outright equity, not options, because at that level realised vol would likely exceed implied and the thesis would be structural rather than event-driven.
Scenario-weighted E[R] for a long: −49.5%, computed from the position's perspective using the Task 2/3 probabilities (bear 30% / base 50% / bull 20%) and outcomes (−76.4% / −48.4% / −12.0%). Net of the 4.7% cash hurdle: −54.2%. No borrow, slippage or premium costs apply because no position is taken.
Simple version for the ticket: risking ~26% (a move to the Street's $312.19 target from a short) to make ~10% (a move to the $207 house target from spot). A 0.4:1 payoff before costs — before any consideration of whether the thesis exists.
The honest summary: the point-estimate risk/reward on a long is poor and the point-estimate risk/reward on a short looks superficially excellent (+49% E[R]). The second number is a trap. It is generated entirely by a discount-rate assumption, has no supporting variant, no resolving catalyst, and a scorecard that argues against it. Recording it here and then declining to act on it is the point of this task.