AMZN › analyze
AMZN · Analyze from before
Price: $256.26 (was $277.42, −7.6% in 24 days) · Market cap: $2.76T · Beta 1.454 Verdict: 🟡 HOLD — conviction [6.3] (baseline: HOLD 6.0) · Differential re-analysis, no new earnings print since baseline.
A portfolio-specific passage was removed from the public build.
The one-paragraph version
Nothing fundamental changed in 24 days — there has been no new earnings print, and every ttm financial figure in this file (revenue $775.68B, opinc $93.71B, FCF −$7.6B, debt $251.6B) is identical to the baseline because it is the same Q2 2026 quarter. What changed is the price (−7.6%) and the news flow, and on balance the news flow is a wash to mildly positive: the FTC monopoly trial was pushed from Oct 13, 2026 to Feb 9, 2027 (a real risk deferral, not a resolution), a new New Jersey AG antitrust suit over delivery-contractor wage suppression was filed (Aug 4) and adds to the regulatory pile without yet being quantifiable, insider selling continued at scale (Bezos sold another ~$4.07B on top of the ~$5.65B already sold, executives keep selling at $255–268 — right at today's price, zero purchases, as flagged before), and the Anthropic IPO now has a concrete shape — investors reportedly targeting a $2 trillion valuation for an October listing, which if it holds would mark Amazon's stake up again, hugely, in the same non-cash, unsellable, ASU-2016-01 way that inflated Q2. The honest read: the business is unchanged, the price is 7.6% cheaper, and it has moved from the top quartile of its own fair-value range to roughly the midpoint — a modestly better entry on an unchanged thesis, still short of the add zone.
Claim Ledger — every baseline claim, re-tested
| # | Baseline claim (2026-08-04) | Status | Fresh evidence (2026-08-28) |
|---|---|---|---|
| 1 | ttm revenue $775.68B, opinc $93.71B, net margin 17.4% > op margin 12.1% (non-operating mark tripwire) | CARRIED | No new quarter filed. get_stock_info shows identical totalRevenue ($775.68B) and profitMargins (17.44%). Same fourth-instance-of-the-pattern finding from [[pattern-net-margin-above-operating-margin-is-a-tripwire]] stands unchanged. |
| 2 | Core EPS ex-marks ~$6.84; headline GAAP 22.2x; core ~40.6x; consensus forward 26.9x is the honest anchor | REFRESHED (price-adjusted only) | Same EPS split, cheaper price: GAAP ttm now 20.9x ($12.25 EPS), core ex-marks now ~37.5x, forward now 24.7x on forwardEps $10.39. Forward still sits below trailing/current-year ($10.39 < $12.25 < $12.48 epsCurrentYear) — the decisive tell from the baseline is intact and gets a second same-shape confirmation from epsCurrentYear/priceEpsCurrentYear per [[pitfall-vendor-forward-eps-is-the-wrong-fiscal-year]]'s diagnostic. |
| 3 | Anthropic carrying value $190.4B, unsellable, Level 3, ~15.9–19.7% ownership | DRIFTED — toward a larger, not smaller, mark | Reporting now points to a $2T IPO target for October, vs. the $965B primary / $1.2T secondary used to size the baseline mark. If that holds, Amazon's stake re-marks sharply upward again in Q3/Q4 — bullish for the headline number, but it is the same unsellable, non-cash mechanism, now larger. The bear-side trigger ("IPO prices below ~$700B") looks unlikely to fire; the realistic path is a further inflation of GAAP EPS, which should be discounted the same way, not treated as validation. |
| 4 | ~$58B carrying-value reconciliation gap, unresolved | UNTESTED (no new 10-Q since baseline; still flagged, not re-checked against footnote) | |
| 5 | ttm FCF −$7.6B vs +$18.2B a year ago; FY2026 capex raised to $220B on memory costs | CARRIED | No new cash flow statement. Independent corroboration arrived instead: cloud-industry-wide DRAM/NAND capex share is now reported rising from 47% (2026) to 68% (2027) of hardware capex, and Amazon itself raised consumer hardware prices (a budget Echo speaker +60%, $49.99→$79.99, Aug 25) — direct evidence the memory shock is a real, economy-wide cost shock hitting Amazon on both the capex and COGS side, not a one-off accounting choice. |
| 6 | AWS 76% of H1 capex, retail capex flat | CARRIED — no new segment disclosure since Q2. | |
| 7 | Zero buybacks/dividends in a decade; $251.6B debt | CARRIED — unchanged in this cycle; no new financing event surfaced this pass. | |
| 8 | Moat WIDE, composition deteriorating; Walmart "landing hardest," Shopify/OpenAI "landing quietly," AWS losing relative share | REFRESHED, one nuance | Walmart's Q2 FY27 (Aug 20) shows US eCommerce decelerating to +24% from Q1's +45% cited in the baseline — the assault is still real (ads +38–43% vs Amazon's ads +26.2%) but less acute than the prior print implied. Directionally unchanged: Walmart still outgrowing Amazon's core, just not accelerating further. |
| 9 | FTC trial 2026-10-13; EU DMA decision Nov 2026 | DRIFTED (FTC only) | FTC trial rescheduled to Feb 9, 2027 — the FTC itself sought the delay citing resource constraints (post-DOGE staffing losses), not case weakness. This removes a near-term catalyst from the Oct window but does not resolve the underlying liability question — just defers it five months. EU DMA still on track: Commission must decide by November 2026, final decision expected late Oct; both firms have until Sept 2026 to respond. CARRIED as scheduled. |
| 10 | Insider selling: zero opens purchases, heaviest selling $255–275, Bezos's last sale July 2025 | REFRESHED, same shape, larger scale | Confirmed again with fresh Form 4 data through Aug 24, 2026: still zero open-market purchases — every entry is a sale, a $0.00 gift, or an RSU vest. New Bezos sale: 1,209,649 sh at $286.41 (Aug 3) plus a disclosed 15M-share, ~$4.07B Rule 10b5-1 program (filed same week) — on top of the ~$5.65B already sold Jun–Jul 2025, cumulative recent Bezos selling is ~$9.7B. Executive cluster (Jassy, Garman, Herrington, Zapolsky, Reynolds, Olsavsky) sold again on Aug 21 and Aug 24 at $257–260 — essentially today's price, the same pattern as the baseline's $255–275 cluster. [[pitfall-yahoo-insider-purchases-counts-rsu-grants]] applies unchanged. |
| 11 | 62 analysts, zero sells, mean target $323.29 | CARRIED | Now 60 analysts, still strong_buy, mean target $327.00 — essentially flat, sell-side conviction unmoved by the 7.6% pullback. |
| 12 | Fair value $225–290, midpoint ~$255; at $277.42 stock in top quartile of own range | REFRESHED — materially better entry, same range | Underlying SOTP/reverse-DCF inputs are unchanged (no new financials), so the $225–290 band carries forward unmodified. At $256.26 the stock now sits almost exactly at the fair-value midpoint, not the top quartile — the single biggest change this pass. |
| 13 | HOLD, conviction 6.0; add zone $215–240; trim 28x fwd ≈ $289 | REFRESHED verdict, price context improved | Trim recomputed on current forwardEps ($10.39): 28x → ≈$291, materially the same level. Entry zone unchanged at $215–240; price is now 6.8% above the top of that zone, down from 15.6% above at baseline. Not yet an add, but the gap has closed by more than half. |
| 14 | New: NJ AG antitrust suit over DSP delivery-contractor wage suppression | NEW — not in baseline | Filed Aug 4, 2026 in the District of New Jersey: alleges Amazon holds monopsony power over delivery-driver labor, bars DSPs from poaching each other's drivers, and retaliates against unionization efforts. Amazon disputes the claims. Adds to, but does not yet quantify, the regulatory attack surface — now three live matters (FTC, EU DMA, NJ AG) plus the still-dormant Xockets ITC complaint. |
| 15 | Berkshire's historical Amazon exit (Q1 2026, pre-baseline) | Not a fresh catalyst | Confirmed as Q4 2025/Q1 2026 history (Buffett sold 77% in his final quarter, Greg Abel liquidated the remainder by Q1 2026, reportedly because the position's internal sponsor left for JPMorgan) — this predates the baseline and is being re-circulated in commentary explaining the August pullback, but it is not new information. Correctly excluded as a driver of the last 24 days' move. |
1. General Health (§1) — carried in full
No new financial statement has been filed since the baseline (next print ~2026-10-30). Every figure below is identical to the 2026-08-04 report:
| FY2023 | FY2024 | FY2025 | ttm Jun-26 | |
|---|---|---|---|---|
| Operating cash flow | $84.95B | $115.88B | $139.51B | $161.40B |
| Capex | −$52.73B | −$83.00B | −$131.82B | −$169.01B |
| Free cash flow | $32.22B | $32.88B | $7.70B | −$7.60B |
FY2026 capex guide $220B, still driven by memory. Zero buybacks, zero dividends, $251.6B debt, 2.0%/yr dilution — all carried unchanged. See baseline for full detail; nothing here needed re-deriving.
What is new: independent, economy-wide corroboration of the memory-cost mechanism. Server DRAM pricing is tracked up ~270% YoY industry-wide by end-2026; cloud operators' DRAM/NAND share of hardware capex is projected to rise from 47% (2026) to 68% (2027). Amazon itself passed the cost through to consumers on Aug 25 (a budget Echo device rose from $49.99 to $79.99, ~60%). This is the strongest evidence yet that the capex raise is a real, external, sector-wide supply shock — not a discretionary AI-infrastructure choice Amazon could easily dial back. That cuts both ways for the thesis: it validates the FCF-destruction mechanism as real and durable rather than a one-quarter anomaly, which argues for patience on the depreciation-wave concern raised in the baseline, not urgency.
2. Moat (§2) — WIDE, composition still deteriorating, one competitive nuance
The baseline's segment map, moat-source table, and circularity quantification ($68B in vs. $238B+ of AWS commitments from Anthropic/OpenAI, ~48% of the $496B backlog) are unchanged — no new segment disclosure exists to re-test them against. Carried in full.
One refresh: Walmart's assault has not accelerated further. Walmart's Q2 FY27 (reported Aug 20) showed US eCommerce growth decelerating to +24% from the +45% cited in the baseline's Q1 FY27 read, even as advertising accelerated further (+38–43% globally vs. Amazon's own +26.2%). Read this as: the competitive pressure is real and structurally still outgrowing Amazon's core retail, but it is not compounding at an accelerating rate quarter over quarter. Not a reversal of the "landing hardest" finding, just a second data point that moderates the trend line.
Regulatory attack surface widened, not narrowed. Three live matters now, not two: 1. FTC monopoly trial — pushed to 2027-02-09 (was 2026-10-13). The FTC requested the delay citing post-DOGE staffing losses, not case weakness — treat this as a deferral of risk, not a reduction of it. 2. EU DMA cloud-gatekeeper designation — still on track for a final decision by November 2026 (Amazon/Microsoft have until September to respond to the June preliminary finding). This is the nearer, and arguably sharper, structural risk: a designation would mandate interoperability and ban AWS self-preferencing, a direct hit to the switching-cost moat the baseline called "the deepest moat Amazon owns." 3. New Jersey AG v. Amazon (DSP monopsony) — filed 2026-08-04, alleging Amazon suppresses delivery-driver wages and retaliates against unionization through its Delivery Service Partner program. Too early to size financially; noted as a new, credible addition to the regulatory pile rather than a re-run of the FTC marketplace case.
Moat verdict unchanged: WIDE, STABLE trend, composition still deteriorating — capital still flows disproportionately to AWS (the segment with the weakest network effects) at the expense of advertising (the strongest moat in the company, still receiving almost no capex).
3. Valuation (§3) — same model, cheaper entry
The SOTP and reverse-DCF machinery from the baseline rests entirely on ttm and FY2027E figures that have not changed since no new quarter has printed. The fair-value band is therefore carried forward unmodified: $225–290, midpoint ~$255. What changed is only the price relative to that band.
| 2026-08-04 (baseline) | 2026-08-28 (this pass) | |
|---|---|---|
| Price | $277.42 | $256.26 |
| Position in FV band ($225–290) | Top quartile | ~Midpoint |
| Distance above entry zone ($215–240) | +15.6% | +6.8% |
| Trim (28x fwd) | ≈$289 (on $10.33 fwd EPS) | ≈$291 (on $10.39 fwd EPS) |
| Headline ttm P/E | 22.2x | 20.9x |
| Core ex-marks ttm P/E | ~40.6x | ~37.5x |
| Forward P/E (consensus) | 26.9x | 24.7x |
Graham IV remains unusable for the same reason as the baseline — $118.74 now vs. $120 then, capitalizing an equity base still carrying a nine-figure Level 3 mark. DDM/DYT remain N/A (no dividend).
The Anthropic IPO is the valuation wildcard, and it has grown, not shrunk, since the baseline. A $2T IPO target (vs. the $965B–$1.2T range the baseline's SOTP haircut was built on) would push the "equity stakes" line of the SOTP well above the baseline's $50–160B range if it holds — but the same caveat applies with more force: a bigger unrealized mark is a bigger reversal risk, not a bigger realized asset. Treat any Q3/Q4 EPS beat driven by this mark exactly as the baseline treated Q2's — as the fourth confirmed instance of [[pattern-ai-build-inflates-earnings-while-destroying-fcf]], not as fundamental progress.
4. Sentiment — selling into strength, unchanged
- Sell-side unmoved: 60 analysts (was 62), still zero sells, mean target $327.00 (was $323.29) — a 7.6% price drop moved the target by essentially nothing, meaning the Street's model didn't change, only the market's near-term risk premium did.
- 🚩 Insider selling continues at scale, still zero purchases. Bezos disclosed a further ~$4.07B, 15M-share sale under a pre-arranged 10b5-1 plan in the first week of August — on top of the ~$5.65B already sold in the baseline window — for ~$9.7B in recent Bezos selling alone. The full executive bench (Jassy, Garman, Herrington, Zapolsky, Reynolds, Olsavsky) sold again on Aug 21 and Aug 24, at $255–260 — essentially today's price. [[pitfall-yahoo-insider-purchases-counts-rsu-grants]] applies without modification: every "purchase" Yahoo's summary counts is an RSU vest or a $0.00 gift.
- Macro framing: WebSearch attributes part of the pullback to "sticky PCE inflation" concerns and Trump-administration chip-tariff headlines layering onto the memory-cost story — consistent with [[Market/regime]]'s standing finding that September rate expectations price a hike, not a cut, and that rate-duration-sensitive names should not be underwritten on an easing assumption.
- Net sentiment read: cautious, not euphoric. The baseline called Q2's reaction "optimism tipping into euphoria." The subsequent 7.6% giveback, continued heavy insider selling at the new lower price, and a flat analyst target suggest the market has partially — not fully — walked back the euphoria while the fundamental picture is unchanged. That is a healthier setup than the baseline found, not a worse one.
5. Verdict
🟡 HOLD — conviction [6.3] (up from 6.0). Still not an add at $256.26, but closer than at any point since the baseline.
Why conviction moved up, not the letter grade: nothing about the business changed in 24 days — no new quarter, no capex reversal, no moat erosion. What changed is that the market gave back roughly half of the re-rating the baseline flagged as "1.5–2x too large," for reasons (macro rate jitters, a new but unquantified state AG suit, continued insider selling, memory/tariff headlines) that do not touch the core AWS-reacceleration thesis. A stock getting cheaper for reasons unconnected to its moat or its cash generation is, by definition, a better entry on an unchanged thesis — precisely the scenario the framework's first principle exists to catch.
Why not higher, and why not ACCUMULATE yet: the price is still 6.8% above the top of the $215–240 add zone, the FCF picture is unchanged (still −$7.6B ttm, still funded by debt, still no plan disclosed for the incremental $20B), and the single largest swing factor — the Anthropic IPO — now points toward an even larger non-cash mark rather than a smaller one, which means the "how much of AWS's growth is circular" question the baseline flagged as the single most important undisclosed number is, if anything, more urgent, not less. The FTC delay is a genuine but modest risk reduction (a near-term catalyst removed), fully offset by the new NJ AG suit and the imminent EU DMA decision, which is arguably the sharper structural threat of the two US/EU matters because it targets the AWS switching-cost moat directly rather than the retail marketplace.
A portfolio-specific passage was removed from the public build.
The single number that would most change this answer — unchanged from baseline, now more urgent
The dollar amount of AWS revenue and RPO owed by counterparties in which Amazon holds an equity stake or funding commitment. Untested this pass — no new disclosure. If circular AWS revenue is under ~$5B of $148B ttm, conviction goes to 7.0 and the verdict to ACCUMULATE on a pullback into the $215–240 zone. If it is over ~$25B, conviction goes to 3.5 and the verdict to AVOID. The Anthropic IPO pricing event (median est. Oct 26) is now the closest thing to a real-world test of this question the market will get — a $2T print validates nothing about circularity per se, but the roadshow disclosures that accompany a public offering of this size are the best chance yet to see Anthropic's own customer-concentration and compute-supplier economics from the other side of the table.
Break triggers (falsifiable, dated) — carried, corrected, and one added
| # | Trigger | Date | Status vs. baseline |
|---|---|---|---|
| 1 | AWS operating margin <36% (≥340bps sequential compression), or consolidated op income below Q3 guide floor | Q3 print ~2026-10-30 | Carried, date refined (Yahoo's earningsTimestampStart now points to ~Oct 30 vs. the baseline's Oct 29 estimate). |
| 2 | Anthropic IPO prices or trades below ~$700B | Median est. 2026-10-26 | Direction of risk flipped from the baseline's framing. Reported investor target is now $2T — the realistic scenario is a large mark-up, not the downside case originally described. Re-read this trigger as "does the mark reverse after pricing," not "does it price low." |
| 3 | FY2027 capex guided ≥$270B while AWS growth guides below 30% | ~Feb 2027 | Carried unchanged, still untested. |
| 4 | FTC monopoly trial | CORRECTED: 2027-02-09 (was 2026-10-13) | Rescheduled; removed as a near-term (Oct 2026) catalyst, now coincides roughly with trigger #3's window. |
| 5 | EU DMA cloud-gatekeeper final decision | Nov 2026 (unchanged, response window closes Sept 2026) | Carried, nearer-term and arguably sharper than the FTC matter — targets the AWS switching-cost moat directly. |
| 6 (new) | NJ AG v. Amazon (DSP monopsony) — first substantive ruling or discovery milestone | Not yet scheduled | New; watch for a motion-to-dismiss ruling as the first dated marker. |
Data-quality notes
- No new 10-Q or 10-K has been filed since the baseline; every ttm financial figure carried in §1–§3 is the Q2 2026 print, not re-derived.
- 🚩
epsCurrentYear($12.475) sits abovetrailingEps($12.25) whileforwardEps($10.39) sits below both — a second, independent confirmation of the baseline's core finding (consensus excludes future marks) via the diagnostic prescribed in [[pitfall-vendor-forward-eps-is-the-wrong-fiscal-year]]. - roic.ai's free tier still returns annual-only data (confirmed again this pass) — [[pitfall-roicai-free-plan-caps-history-at-two-years]] applies; quarterly/ttm cross-validation was not attempted via that source.
- 🚩 Yahoo's
freeCashflowfield again shows a small positive figure ($3.22B) against the company's own −$7.6B ttm FCF — same finance-lease definitional gap as the baseline; use the company's number. - The $2T Anthropic IPO figure is a reported investor target, not a filed valuation — treat as directional, not a number to build a fair-value line around, until an S-1 or pricing actually lands.
Sources
.mcp/fin.py AMZN --news, Yahoo Finance MCP (get_stock_info, get_holder_info insider_transactions, get_recommendations), retrieved 2026-08-28 · FX Leaders — AMZN retreats on tariffs/capex · CNBC — Bezos files to sell $4B · Fortune — Anthropic $2T October IPO · futuresearch.ai — Anthropic IPO date/valuation forecast · CNBC — NJ AG antitrust suit over DSP drivers · NJ OAG press release · Bloomberg Law / MLex — FTC trial rescheduled to Feb 2027 · Digital Markets Act EC — AWS/Azure gatekeeper preliminary position · Walmart Q2 FY27 earnings release · The Register — memory crunch, cloud capex to DRAM/NAND · 24/7 Wall St — Berkshire's Amazon exit context
Related agency knowledge: [[pattern-ai-build-inflates-earnings-while-destroying-fcf]] · [[pitfall-unrealized-equity-marks-break-headline-pe]] · [[pitfall-vendor-forward-eps-is-the-wrong-fiscal-year]] · [[pitfall-yahoo-insider-purchases-counts-rsu-grants]] · [[Market/regime]] · baseline AMZN 2026-08-04