01 · Equity deep-dive — synthesized analyst desk
AMZN
$254.92 ▼ 11% off the Aug 3 record
NASDAQ · E-COMMERCE & CLOUD INFRASTRUCTUREMKT CAP $2.75T52-WK $196.00 – $287.20AS OF SEP 2, 2026

Amazon has never earned more. It has never kept less of it.

Operating income rose 43% last quarter and AWS grew 37% — its fastest in eighteen quarters. And trailing free cash flow is negative $7.6 billion, because every dollar the business generates, plus $60 billion of fresh debt, is being poured into concrete and silicon. The conveyor still runs; the question is what comes off the far end. Six analyst lenses, three scenarios, four horizons.

The verdict · TL;DR
One question decides the stock: is the $220‑billion‑a‑year pour the widest moat ever built — pre-sold through 2028 by a $496B backlog — or a debt-funded defence of cloud share Amazon is already losing? AWS accelerated five straight quarters yet its market share still slipped roughly two points year-over-year. Trailing earnings look cheap at 21× only because a $53.4B paper mark on Anthropic is sitting in them; the forward multiple is higher than the trailing one. Our desk lands materially below the Street.
5-yr · prob-weighted
$447
+75% vs $254.92
52-week playback · where the tape sits »» Mid-band, off the record
$254.92 · Sep 1, 2026 close consensus $327.67 · +29%
$196.00 · 52-wk low · Feb 17 ’26 $287.20 · record · Aug 3 ’26
Price history + cone of outcomes · Dec 2024 → Sep 2031
HISTORICALBULLBASEBEARPROB-WTD
$700$500$360 $260$190$140 20252026 202720282029 20302031 $161 · Apr ’25 $259 · late ’25 $196 · Feb 17 ’26 $287 record · Aug 3 $447 Street $328 $267$313$359 $689 $456 $232 TODAY · $254.92 »»»

Left of the divider: Amazon’s actual price into today — $161 trough (Apr ’25) → $259 late-’25 high → $196 low on Feb 17, 2026 after the $200B capex guide landed → a 15.3% single-day jump on July 31 to a record $287.20 on Aug 3 → $254.92 now. Dotted markers are dated closes from S&P Global Market Intelligence; the connecting path between them is drawn to scale but smoothed. Right of the divider: synthesized scenario midpoints, probability-weighted at bear 30% · base 45% · bull 25%. Log scale, annual marks. Wall Street’s 12-month consensus is $327.67 (61 analysts, S&P Global; range $230–$405, zero sell ratings).

Re-weight the outcomes yourself

Our desk starts bear-tilted — 30 / 45 / 25 — because the forward multiple sits above the trailing one, free cash flow is negative, and long-term debt has doubled in six months. Disagree? Drag. The base case fills whatever is left, and the blended target, the dotted line on the chart above, and the clay row in every scenario card all recompute live.

BEAR 30%BASE 45%BULL 25%
Blended 5-yr target $447 +75% vs $254.92
02 · The quarter, in eight numbers

Q2 2026 — reported July 30, 2026

The operating business had one of the best quarters in Amazon’s history. The cash statement had one of the worst. Both are true, and the gap between them is the investment case.

$200.6B
Net sales +20% · first $200B quarter
$42.2B
AWS sales +37% · fastest in 18 qtrs
13.7%
Operating margin · vs 11.4%
39.4%
AWS operating margin +650bp
$19.8B
Advertising +26% · highest-margin line
$496B
AWS backlog · +$132B in one quarter
−$7.6B
Trailing free cash flow · was +$18.2B
$169.0B
Trailing net capex +64%

Source: Amazon Q2 2026 results (Form 8-K exhibit 99.1, filed July 30, 2026) and the Q2 2026 Form 10-Q. Backlog figure per management commentary on the July 30 call. Percentages are year-over-year unless stated.

03 · The analyst panel

Six lenses, reasoned independently, then made to argue

Each desk was run as a self-contained pass before reading the others, so the disagreement is real rather than negotiated. They span $196 to $355 — an 81% spread on the same set of filings. The panel mean, $277, sits well below the Street’s $328.

Growth · momentum PM

Nothing this large has ever accelerated like this

A $150B run-rate business does not speed up from 17% to 37% over five quarters unless demand is structurally short of supply — and the backlog says the next two years are already sold.

  • AWS growth 17% → 20% → 24% → 28% → 37% across five quarters; backlog $496B, up $132B in a single quarter
  • Advertising +26% to $19.8B — the highest incremental-margin line Amazon owns
  • AI and custom-chips businesses each past $25B annualised, growing triple digits
  • Operating leverage is live: income +43% on sales +20%
12-mo target · 37× FY27E $9.55$355
ConvictionHIGH
Moat · competitive strategy

The moat is power, land and silicon — not software

$220B a year buys a physical position in megawatts and fabs that no rival can replicate on a timeline shorter than a decade. Software moats get cloned; substations do not.

  • Graviton runs at 98% of the top-1,000 EC2 customers; Graviton5 is 25% faster than Graviton4
  • Trainium holds multi-year, multi-gigawatt commitments from both leading AI labs
  • AWS power capacity set to double by end-2027 versus 2025, when power is the binding constraint
  • Three profit pools: AWS (~61% of operating income), advertising, and logistics now sold as a service
12-mo target · sum-of-parts$315
ConvictionHIGH
Quant · technical

Trend intact, valuation percentile is the problem

The tape is fine. What the February drawdown revealed is exactly what the marginal buyer is sensitive to — and it is not revenue.

  • $254.92 sits above the 50-day ($250.94) and well above the 200-day ($239.94) as of Aug 27
  • But 11.2% has come off in under a month since the Aug 3 record
  • Feb 2026 saw a 24% peak-to-trough drawdown on a capex guide alone — the sensitivity is capital, not sales
  • Short interest 0.88% offers no covering cushion; beta 1.45; 66 open-market insider sales, zero buys, in six months
12-mo target · retest of the record$285
ConvictionMEDIUM
Macro · sector strategy

A cash machine has become a duration asset

Amazon has taken on consumer cyclicality and long-bond sensitivity in the same ticker, at the moment the market is re-pricing debt-funded AI capex.

  • The buildout is bond-financed: $37B raised in March, $25B more in July; long-term debt $65.6B → $128.9B in six months
  • Interest expense $1.31B in Q2 versus $0.52B a year earlier — a 2.5× step-up that compounds into 2027
  • Memory-chip inflation is the stated reason for the $20B capex raise: input costs hit the balance sheet before the P&L
  • The consumer is holding — paid units +17%, North America +16% — but shipping costs are up 19%
12-mo target · 28× FY27E, rate-discounted$265
ConvictionMEDIUM-LOW
Value · FCF & quality

The trailing multiple is an optical illusion

A company posting negative free cash flow, doubling its debt, and booking two-thirds of pre-tax income from marking up a private stake is not a quality compounder this year. It is an infrastructure project with a retailer attached.

  • Trailing P/E reads 21× — but forward is 28.8×. When forward exceeds trailing, the trailing number is the fiction
  • TTM free cash flow −$7.6B versus +$18.2B a year ago; a $66.1B swing in net capex
  • Return on invested capital 11.9% and falling as the asset base compounds faster than profit
  • Offsetting it: $161B of real operating cash flow, $123B of cash and securities, and a marked Anthropic stake worth ~7% of market cap
12-mo target · 25× FY27E + stake$245
ConvictionMEDIUM
Bear · short-seller

Accelerating and still losing share is the worst combination

Spending like a winner while the scoreboard moves the other way. And the largest new customer is a company Amazon owns a fifth of.

  • AWS grew 37% — and its infrastructure share still slipped from roughly 30% to 28% year-over-year, because Google Cloud grew 82%
  • Circularity: roughly $100B of the $132B backlog jump was the Anthropic commitment; Amazon marks its own ~21% stake up by $53.4B, then books the customer’s spend as backlog
  • The depreciation wave: $169B of trailing net capex on three-year server lives; D&A is already $75.2B and up 28%
  • Regulation on both flanks: the FTC and 22 states allege $20B+ of hidden ad surcharges (filed Aug 31, 2026); the monopoly trial opens March 29, 2027
12-mo target · 21× haircut FY27E$196
ConvictionHIGH

These lenses are synthesized analytical frameworks built from published research methodologies — not real individuals, and not real firm ratings. Targets are the output of the stated multiple applied to the stated earnings estimate; the arithmetic is shown, so you can disagree with the inputs rather than the conclusion.

04 · Wall Street targets

Sixty-one analysts. Zero sell ratings.

The consensus is $327.67, about 29% above the last close. It is also the most one-sided book on any mega-cap we track: 59 buys, 2 holds, no sells. That unanimity is itself a risk factor — there is nobody left to upgrade.

12-month price targets · sorted low to high
BUYHOLDSELL · none
$0$100$200 $300$400 Street lowWedbushMizuho Morgan StanleyTelseyRosenblatt JPMorganStreet high $230 $310 $330 $335 $335 $345 $365 $405 CONSENSUS $327.67 TODAY $254.92

Consensus, range and rating distribution per S&P Global Market Intelligence via StockAnalysis, as of Sep 1, 2026 (61 analysts; 59 buy / 2 hold / 0 sell). Named targets are post-Q2 revisions: JPMorgan (Doug Anmuth) $365 and Wedbush $310 on July 30–31, 2026; Rosenblatt $345, Telsey $335 and Mizuho $330 on July 31; Morgan Stanley (Brian Nowak) $335 on Aug 16. Other aggregators show slightly different means — ChartMill lists $323.84 across 75 analysts — because coverage universes differ; we use the S&P Global figure throughout.

05 · Scenarios by horizon

Four horizons, four outcomes, one live row

Every price below is an exit multiple applied to a projected earnings path — nothing more sophisticated, and nothing hidden. The clay row is wired to the sliders above. Bars are normalised within each card, so bull always reads full width and you are comparing shape, not scale.

1 year

Sep ’27
Bear$190−25%
Base$267+5%
Bull$322+26%
Prob-wtd$258+1%

2 years

Sep ’28
Bear$196−23%
Base$313+23%
Bull$408+60%
Prob-wtd$302+18%

3 years

Sep ’29
Bear$205−20%
Base$359+41%
Bull$499+96%
Prob-wtd$348+36%

5 years

Sep ’31
Bear$232−9%
Base$456+79%
Bull$689+170%
Prob-wtd$447+75%
Show the assumptions & the math

Everything starts from operating EPS — earnings from the actual business, excluding the investment marks that have distorted reported GAAP figures all year. FY2026E operating EPS is $7.90, derived from H1 operating income of $51.3B, the Q3 guide midpoint of $24.5B, an estimated ~$29B in Q4, roughly a 20% ongoing tax rate and ~10.9B diluted shares. Reported GAAP EPS for 2026 will land far higher — consensus near $13.10 — because it includes the Anthropic revaluation. We do not capitalise a paper mark.

Bear — 30% · the digestion year arrives

AI demand normalises through 2027; AWS decelerates toward the low-20s by 2028 and mid-teens thereafter. Depreciation from the $220B and $260B capex years lands on the P&L and pins operating margin near 13%. Ad remediation plus a settlement costs $1.5–2B a year. The market re-rates Amazon as a capital-intensive utility.

EPS path   2027 $8.60 · 2028 $9.40 · 2029 $10.30 · 2031 $12.80
1yr  $8.60 × 22× = $189 → $190
2yr  $9.40 × 20.8× = $196 → $196
3yr  $10.30 × 19.9× = $205 → $205
5yr  $12.80 × 18.1× = $232 → $232

Base — 45% · the backlog converts, slowly

AWS grows ~30% in 2027 and ~24% in 2028 before settling near 18%. Capex peaks around $260B in 2027 then grows slower than revenue, so free cash flow crosses back above zero during 2028. Advertising compounds ~20%. Operating margin drifts toward 15% by 2029. The multiple holds roughly where it is.

EPS path   2027 $9.55 · 2028 $11.60 · 2029 $13.80 · 2031 $19.00
1yr  $9.55 × 28× = $267 → $267
2yr  $11.60 × 27× = $313 → $313
3yr  $13.80 × 26× = $359 → $359
5yr  $19.00 × 24× = $456 → $456

Bull — 25% · 2028 demand is as striking as management says

AWS holds above 30% growth through 2028; Trainium takes genuine inference share, so gross profit per dollar of capex improves structurally rather than just scaling. Advertising re-accelerates past 25%. Anthropic lists at $2T or more and the ~21% stake — carried at $190.4B on June 30 — is re-rated toward $400B+ and valued separately by the market.

EPS path   2027 $10.40 · 2028 $13.60 · 2029 $17.20 · 2031 $25.50
1yr  $10.40 × 31× = $322 → $322
2yr  $13.60 × 30× = $408 → $408
3yr  $17.20 × 29× = $499 → $499
5yr  $25.50 × 27× = $689 → $689

Blending

At the default 30 / 45 / 25 weights:
1yr  190(.30) + 267(.45) + 322(.25) = $257.65  (+1% vs $254.92)
2yr  196(.30) + 313(.45) + 408(.25) = $301.65  (+18%)
3yr  205(.30) + 359(.45) + 499(.25) = $347.80  (+36%)
5yr  232(.30) + 456(.45) + 689(.25) = $447.05  (+75%)
Implied 5-year compound annual return ≈ 11.9%

Read the 1-year number carefully. A probability-weighted $258 against a spot price of $254.92 says the next twelve months are, on this framework, roughly a coin flip — while the Street sits at $327.67. The disagreement is not about whether AWS is growing. It is about what multiple a business deserves while it is consuming cash.

06 · Capital & cash

The chart the whole argument lives in

We plot operating cash flow rather than revenue here, because at Amazon’s scale a revenue bar would dwarf everything and hide the actual story. That story is simple: in 2023 the clay bar was two-thirds of the blue one. In 2026 it overtakes it — and the olive bar goes through the floor.

Operating cash flow vs capex vs free cash flow vs debt · $ billions
OP CASH FLOWCAPEXFREE CASH FLOWLONG-TERM DEBT
$240$180$120 $60$0 85116140176 5383132220 373811−40 585366155 20232024 20252026E

2023–2025 are reported figures from Amazon’s annual results (FY2025 10-K and prior 8-K exhibits): capex is gross purchases of property and equipment; free cash flow is Amazon’s own definition — operating cash flow less purchases of property and equipment net of proceeds; debt is long-term debt excluding the current portion. 2026E is our estimate, built on management’s ~$220B cash capex guidance given July 30, 2026, H1 actuals, and the $128.9B of long-term debt on the balance sheet at June 30 plus the $25B raised in July. Trailing-twelve-month free cash flow through June 30, 2026 was already −$7.6B.

07 · The earnings ladder

Every target above is this chart times a multiple

Solid bars are operating earnings — what the business actually produces. The dashed clay outline is where reported 2026 GAAP EPS is likely to land once the Anthropic revaluation is included. That $5.20 gap is why Amazon’s trailing P/E looks like 21× while its forward P/E is 28.8×.

Diluted EPS · reported and estimated
REPORTEDESTIMATED2026 GAAP incl. paper mark
$14$10.50$7 $3.50$0 $13.10 GAAP $2.90 $5.53 $7.17 $7.90 $9.55 $11.60 202320242025 2026E2027E2028E reportedoperating estimate

2023–2025 are reported diluted EPS from Amazon’s annual results. 2026E–2028E are our operating estimates, which exclude non-operating investment revaluations; they are broadly consistent with published normalised consensus (roughly $7.7–8.0 for 2026 and $9.3–9.6 for 2027). The $13.10 GAAP figure is the consensus reported number and includes the Anthropic mark. In the trailing twelve months to June 30, 2026, reported EPS was $12.44 against roughly $7.50 of operating earnings — the same distortion, already in the price data.

08 · Growth scorecard

The core is steady. The frontier is on fire.

Latest reported year-over-year growth, sorted low to high. Olive is the mature core; clay is the frontier that now carries the thesis. Read it against the price: the operating business is accelerating while the stock is 11% below its record. That disconnect is the bull case — and the bear case is that it is being bought with borrowed money.

Year-over-year growth · Q2 2026
STEADY COREFRONTIER
0%25%50% 75%100% Physical stores Subscriptions International 3P seller services Worldwide paid units Total net sales Advertising AWS net sales Operating income AWS operating income AI & chips run-rate +4% +12% +15% +16% +17% +20% +26% +37% +43% +64% triple digits

All figures from Amazon’s Q2 2026 results, quarter ended June 30, 2026. “AI & chips run-rate” is management’s disclosure that each of those two businesses passed a $25B annualised revenue run rate growing at triple-digit percentages; the bar is drawn at 100% as a floor, not a measured value. Physical stores excludes fuel and other adjustments Amazon does not break out.

09 · The debate

Bull versus bear, on the same set of filings

Both sides here are reading the identical 8-K. They disagree about one thing: whether $220 billion a year of spending is an asset being built or a margin being destroyed.

▲ THE BULL CASE

  • Five straight quarters of acceleration at enormous scale. AWS went 17% → 20% → 24% → 28% → 37%, the fastest in eighteen quarters — while its operating margin expanded 650bp to 39.4%. Accelerating revenue and widening margin at a $169B run rate is close to unprecedented.
  • The backlog is contracted, not hoped for. $496B of committed AWS revenue, up $132B in a single quarter and roughly 2.5× year-over-year. 2027 capacity is largely reserved; management called 2028 demand “striking.”
  • Advertising is the quiet compounder. +26% to $19.8B in the quarter — an ~$80B annual run rate at very high incremental margin, growing faster now than it was three years ago.
  • Vertical silicon is a genuine cost moat. Graviton runs at 98% of the top-1,000 EC2 customers and Graviton5 is 25% faster than Graviton4; Trainium holds multi-year, multi-gigawatt commitments from both leading AI labs. AI and chips each cleared $25B annualised.
  • The spending is demand-pulled. Management raised 2026 capex to $220B and still says it cannot meet demand in 2026 — or 2027. You do not get a capacity shortage in a speculative bubble.
  • The Anthropic stake is a real asset, not just an accounting artefact. $13B invested; carried at $190.4B on June 30 — about 7% of market cap. A listing near $2T would push that toward 15%, and Amazon collects the customer’s AWS spend on top.
  • Operating leverage is already visible. Operating income +43% on sales +20%; margin 13.7% against 11.4% a year ago.

▼ THE BEAR CASE

  • Accelerating and still losing share. AWS grew 37% — and its cloud infrastructure share still slipped roughly two points year-over-year, to about 28%, because Google Cloud grew 82% and Azure around 40%. Spending like the leader while the scoreboard moves the other way is the worst of both.
  • Free cash flow is gone. −$7.6B trailing versus +$18.2B a year ago, and our 2026 estimate is roughly −$40B. All $161B of operating cash flow, plus tens of billions of new debt, goes into the ground.
  • The balance sheet changed shape in six months. Long-term debt $65.6B → $128.9B, plus $25B more raised in July. Interest expense is up 2.5×. Asked how the rest gets funded, the CEO said: “Nothing to share today.”
  • Earnings quality is poor right now. Two-thirds of Q2 pre-tax income was a non-operating mark on a private stake. The forward multiple (28.8×) is higher than the trailing one (21.4×) — the tell that the trailing number is fiction.
  • Circularity. Roughly $100B of the $132B backlog jump was the Anthropic commitment. Amazon owns about a fifth of Anthropic, marks that stake up by $53.4B, and books the same company’s spending as contracted revenue. The asset, the customer and the growth are one bet.
  • The depreciation wave is arithmetic, not opinion. $169B of trailing net capex on roughly three-year server lives. D&A is already $75.2B and up 28%. If AI revenue growth merely normalises to 25%, depreciation catches gross profit.
  • Two regulatory fronts at once. The FTC and 22 states allege $20B+ of hidden ad surcharges since 2019 — aimed squarely at the highest-margin segment — and the separate monopoly bench trial opens March 29, 2027.
10 · Risk map

Likelihood × impact, over three to five years

Where each risk sits, not merely how loud it is. Note that Amazon’s hottest cell is not a competitor or a regulator — it is its own depreciation schedule, which is the one risk that arrives on a fixed timetable whether or not anything goes wrong.

Low impact
Medium impact
High impact
Likely
  • FX & shipping-cost drag
  • AWS relative share erosion
  • Memory / input-cost inflation
  • Depreciation wave
Possible
  • Anthropic mark reversal
  • Ad-pricing remediation
  • Financing & rate shock
Tail
  • Structural antitrust remedy
  • AI demand air-pocket

Depreciation wave

Likely × High

What breaks: $169B of trailing capex on ~3-year server lives lands as D&A faster than AI revenue scales, and operating margin goes backwards even with revenue growing.

Ad-pricing remediation

Possible × High

What breaks: the FTC suit forces auction changes or restitution on a business the agency says gathered $20B+ in hidden surcharges — hitting Amazon’s highest-margin revenue line.

Financing & rate shock

Possible × High

What breaks: the long end stays elevated and a bond-funded $220B annual programme reprices, turning a growth story into a spread story.

Structural antitrust remedy

Tail × High

What breaks: the March 2027 monopoly trial produces conduct or structural remedies separating marketplace, logistics or Prime economics. Low odds, but it reprices the retail flywheel outright.

AI demand air-pocket

Tail × High

What breaks: a funding squeeze at the AI labs converts a capacity shortage into a capacity glut, and $496B of backlog turns out to be renegotiable. Amazon would own the depreciation without the revenue.

AWS relative share erosion

Likely × Medium

What breaks: AWS keeps growing but slower than the market, so the premium multiple attached to cloud leadership migrates to Google Cloud.

Memory / input-cost inflation

Likely × Medium

What breaks: memory prices already forced a $20B capex raise. More of the same buys the same capacity for more money, compressing returns on invested capital.

Anthropic mark reversal

Possible × Medium

What breaks: a soft listing or a down round reverses part of the $190.4B carrying value, producing headline losses and exposing how thin operating earnings are underneath.

FX & shipping-cost drag

Likely × Low

What breaks: an 80bp FX headwind is already in Q3 guidance and worldwide shipping costs are up 19%. Chronic, manageable, but a persistent tax on retail margin.

11 · Plain-language glossary

The jargon, decoded

Hover the dotted terms up in the metrics strip, or scan the desk’s working definitions here.

Free cash flow
Operating cash flow minus money spent on property and equipment. Amazon’s is currently negative: the business generates $161B a year and spends more than that building.
Capex
Capital expenditure — cash spent on data centres, power, servers and chips. Guided to roughly $220B for 2026, up from $132B in 2025.
Backlog
Contracted revenue not yet delivered. AWS’s $496B is work customers have already committed to pay for, which is why management can talk about 2028 with confidence.
Operating margin
Profit from running the business, divided by sales — before interest, tax and investment marks. 13.7% last quarter, versus 11.4% a year earlier.
EV / EBITDA
Enterprise value (market cap plus net debt) divided by earnings before interest, tax, depreciation and amortisation. About 17× here — a measure that deliberately ignores the depreciation the bears care most about.
Trailing vs forward P/E
Price divided by the last twelve months’ earnings, versus the next twelve months’. Amazon’s forward multiple is higher than its trailing one — unusual, and a sign the trailing figure is inflated by one-off gains.
Non-operating mark
A paper gain from revaluing an investment, booked through the income statement without any cash changing hands. Amazon recorded $53.4B of these in Q2 2026, mostly on its Anthropic stake.
Depreciation wave
The lagged cost of today’s building. Servers are written off over roughly three years, so 2026’s $220B of spending becomes an earnings headwind from 2027 onward whether or not revenue follows.
ROIC
Return on invested capital — profit earned per dollar of capital employed. Currently 11.9%, and mathematically falling while the asset base grows faster than profit.
Exit multiple
The price-to-earnings ratio assumed at the end of a forecast. Multiply it by projected EPS and you have a target price. Every number in this report is built that way.
Prob-weighted
Each scenario’s price multiplied by its probability, then summed — a single expected value across bear, base and bull. The clay figures throughout.
Circularity
When an investor, a supplier and a customer are the same party. Amazon owns ~21% of Anthropic, sells Anthropic cloud capacity, and books the resulting commitment as backlog.