
Amazon
Three-engine: 1P+3P marketplace retail (thin-margin, scale), AWS cloud (high-margin profit engine), and high-margin ads/subscriptions (Prime); reinvests retail+ads cash flow into AWS capex and custom AI silicon.
The thesis on this name
State of AI Compute
AWS is the largest cloud plus the Trainium/Inferentia custom-silicon program — demand anchor and a vertical-integration risk to merchant chips.
State of the AI Cloud
The emerging second axis of the capacity market — GOOGL (TPU) and AMZN (Trainium) can both cut their own cost AND rent a Nvidia-alternative. AWS re-accelerated to ~28% YoY (fastest in 15 quarters) on a ~$150B run-rate with a >$225B Trainium revenue-commitment book and Trainium3 nearly fully subscribed; Anthropic's $100B/5GW Trainium deal anchors years of demand. The cleanest validation that the merchant-silicon renter base is real.
State of the AI Cloud
Largest installed cloud base, re-accelerating to ~28% YoY, with a custom-silicon (Trainium) book already ~fully subscribed through Gen3.
State of the AI Cloud
Largest installed cloud base re-accelerating to ~28%, with a Trainium book ~fully subscribed through Gen3; the second in-house-silicon axis.
State of Frontier AI
The diversified way to ride Anthropic's enterprise-share lead now that ANT itself is listed — large Anthropic stake + Trainium silicon + Bedrock (Q1'26 tokens exceeded all prior years combined, customer spend +170% QoQ) — with AWS/retail cash cushioning AI capex.
State of Frontier AI
Holds a large Anthropic stake + Trainium silicon + Bedrock (token volume in Q1'26 exceeded all prior years combined; customer spend +170% QoQ). With Anthropic itself listed as ANT since Nov 19 2026, AMZN is the diversified way to ride its enterprise-share lead (34.4% > OpenAI) and the mark-up its IPO delivered, with retail/AWS cash cushioning AI capex.
Earnings, margins, COGS & capex
Q3 FY26 was a clean operating beat — $201.9B sales (+12%, +13% ex-FX) and a record 13.9% operating margin — led by AWS reaccelerating to +39.3%, its fastest growth in nearly five years, on AI demand. Q3 EPS of $1.89 (vs $1.95) carried no Anthropic gain, unlike the first half ($16.8B pre-tax in Q1 and $50.5B in Q2); trailing-twelve-month free cash flow is an outflow of $42.4B on a ~$220B 2026 capex guide, and long-term debt reached $173.1B.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~47¢ is cost of goods and ~39¢ operating expense, leaving ~14¢ of operating profit.
Revenue trend
Margins
up from 9.7% Q3 FY25 (13.1% in Q1 FY26, 13.69% in Q2)
strong; key profit engine (39.4% in Q2, ~37.7% in Q1; ~34.5% in Q3 FY25)
expanding on logistics efficiency + ads (4.5% in Q3 FY25; ~8.0% in Q1)
improving, structurally thinnest (2.9% in Q3 FY25; ~3.5% in Q1)
COGS structure
Retail COGS dominated by product/fulfillment/shipping; not broken out as a single consolidated gross-margin line. AWS COGS increasingly weighted to data-center depreciation as the capex base scales.
Capex
FY25 ~$128.3B (+65%); FY26 guided ~$220B (+71%, computed; raised from ~$200B in July on memory costs and reiterated on Oct 29) — majority to AWS AI data centers, Trainium/Inferentia silicon, plus robotics and Leo satellites. This is the dominant swing factor for FCF and the stock.
Latest earnings
Beat on both lines against Amazon's own guidance. Net sales $201,944M (Q3 FY26), +12% Y/Y (+13% ex-FX), at the top of the $197.0–202.0B guide and held down by about four points of growth because Prime Day moved into Q2. Operating income $28,076M vs the $22.5–26.5B guide, +61% Y/Y, a record 13.9% consolidated margin (9.7% in Q3 FY25). GAAP diluted EPS $1.89 (vs $1.95 in Q3 FY25) on net income $20,685M: the decline reflects $9.5B of pre-tax Anthropic gains in Q3 FY25 and none in Q3 FY26 (total non-operating expense of $1,207M), so this quarter's EPS is operational, which the first half was not (Q2 EPS of $5.75 included $53,396M of pre-tax non-operating income 'primarily from our investments in Anthropic', Amazon's own wording, 3.2x the Q1 FY26 Anthropic gain). The company behind that stake went public on Nov 19 2026 (Anthropic, PBC listed on Nasdaq as ANT after pricing its IPO at $83 on Nov 18, following a $65B Series H at a $965B valuation reported by Fortune in Jun 2026), so from Q4 the stake is marked to a public share price, with large mark-to-market swings in both directions. Street-consensus comparisons were NOT verified, so no consensus-based beat magnitude is claimed; the ~$1.97 ex-gain 'adjusted' EPS circulating in LSEG-sourced press for Q2 is secondary, non-GAAP, and not published by Amazon.
Q4 FY26 guidance (as of Oct 29 2026, from the release): net sales between $240.0 billion and $247.0 billion, or growth of 12% to 16% compared with fourth quarter 2025, anticipating an unfavorable impact of approximately 60 basis points from foreign exchange rates. Advertising revenue is expected to grow in the high teens year over year in Q4, after +12% ex-FX in Q3, which was held down by Prime Day and other timing anomalies. Operating income between $28.5 billion and $33.5 billion, compared with $25.0 billion in Q4 2025. 'We continue to expect 2026 cash capital expenditures to be approximately $220 billion' (raised from ~$200B in July, with 'the higher cost of memory pushing this number up'); no FY27 capex figure was given. Assumes no impact from energy derivative contract remeasurements and no additional acquisitions, restructurings, or legal settlements. No full-year FY26 guidance was issued — Amazon guides one quarter at a time. For reference, the Q3 guidance of Jul 30 2026 was net sales of $197.0-202.0B and operating income of $22.5-26.5B; actuals were $201.9B and $28.1B. Management also projects AWS revenue growth of over 40% in both 2027 and 2028 (Jassy, release; a projection, not guidance).
- AWS net sales
- $45,984M (Q3 FY26), +39.3% Y/Y (headline '39%') — fastest growth in nearly five years, ~$184B annualized run rate (Q2: $42,232M, +36.8%)
- AWS operating income
- $17,773M (Q3 FY26, derived from the release's segment table) — a 38.7% segment margin vs ~34.5% a year earlier (Q2: $16,621M, 39.4% vs 32.9%)
- Total operating income
- $28,076M (Q3 FY26), +61% Y/Y — above the top of Amazon's own $22.5–26.5B guide; 13.9% consolidated margin vs 9.7% in Q3 FY25 (Q2: $27,461M, 13.69%)
- GAAP diluted EPS
- $1.89 (Q3 FY26) vs $1.95 (Q3 FY25), on net income $20,685M vs $21,187M — Q3 FY25 included $9.5B of pre-tax Anthropic gains and Q3 FY26 none; Q2 FY26 was $5.75 on net income $62,647M, of which $53,396M was pre-tax non-operating income 'primarily from our investments in Anthropic'
- Trailing-twelve-month free cash flow
- OUTFLOW of $42.4B (TTM to Sep 30 2026), vs a $14.8B inflow a year earlier — roughly -5.3% of TTM revenue (an outflow of $7.6B at Jun 30 2026)
- Trailing-twelve-month capex
- $215.2B of purchases of property and equipment ($210.7B net of proceeds and incentives), against FY26 cash-capex guidance of ~$220B (raised from ~$200B in July, reiterated Oct 29)
- North America net sales
- $112,852M (Q3 FY26), +6% Y/Y (Q2: $116,177M, +16%; Prime Day timing moved into Q2)
- International net sales
- $43,108M (Q3 FY26), +5% Y/Y (+8% ex-FX) (Q2: $42,197M, +15%)
- Gross margin (derived — Amazon prints no gross-profit line)
- 53.17% (Q3 FY26: $201,944M net sales less $94,573M cost of sales), vs 50.79% in Q3 FY25 (Q2 FY26: 52.26%)
- Custom silicon and AI businesses
- chips business >$25B run rate and AI business >$25B run rate, both at triple-digit growth (Q2 FY26 release); Graviton5 shipped GA
- Third-party seller services
- $46.8B in Q2 FY26, ~23% of revenue
- Multiples (3 Nov 2026 close; computed from aggregator consensus EPS, estimates)
- trailing P/E 20.3x — distorted downward by the first-half Anthropic marks; 23.3x FY2026 and 23.0x FY2027 consensus EPS (29.47x forward at the 31 Jul 2026 close)
- Headcount
- 1,595k, up sequentially from 1,575k despite the July AGI-division layoffs — reallocation, not contraction
Growth drivers
- AWS AI demand — Bedrock, Trainium2 (~30% better price/perf vs comparable GPUs, ~sold out) and Trainium3 (shipping, near-fully subscribed)
- Anthropic — $100B+ AWS commitment over 10yr, up to ~5GW compute; locks long-dated AWS backlog
- High-margin retail advertising (~$70B-scale business)
- Prime subscriptions + logistics efficiency lifting NA/Intl margins
- Custom silicon (Trainium/Inferentia/Graviton) — >$10B revenue this year; reduces NVIDIA dependence
Reported financials — SEC EDGAR
Audited GAAP figures pulled from SEC filings · latest filing 2026-02-06. The audited primary-source spine — not financial advice.
Revenue — annual (GAAP)
Margins & balance sheet — FY’25
Bull & bear
AWS accelerated again to +39.3% Y/Y in Q3 FY26 ($46.0B, fastest growth in nearly five years, ~$184B annualized run rate; +36.8% in Q2) at a 38.7% segment operating margin vs ~34.5% a year earlier - the profit engine reaccelerating harder than the July framing assumed, with segment margin well above last year while the depreciation base scales. Consolidated revenue was $201.9B (+12%, +13% ex-FX; about four points higher excluding the Prime Day shift into Q2) and operating income printed $28.1B vs $17.4B and above the top of Amazon's own $22.5-26.5B guide, so operating leverage held through peak investment (13.9% consolidated margin vs 9.7% a year earlier). The silicon demand story broadened from Anthropic-centric to both leading labs - Amazon names 'Anthropic and OpenAI, making multi-year, multi-gigawatt commitments' to Trainium - with chips and AI each above a $25B run rate at triple-digit growth (Q2 release). The entry has changed again: at $250.69 (Nov 3 2026 close) the stock sits ~12.7% below its 52-week high of $287.20 and ~7.7% below its Jul 31 close.
- AWS $46.0B (Q3 FY26), +39.3% Y/Y - fastest growth in nearly five years; $184B annualized run rate; Jassy projects AWS revenue growth of over 40% in both 2027 and 2028 (management projection, not guidance)
- AWS segment operating income $17.8B (Q3 FY26), a 38.7% margin (derived) vs $11.4B and ~34.5% a year earlier (Q2 FY26: 39.4%) - margin still well above last year despite the scaling depreciation base, but a little below Q2
- Consolidated operating income $28.1B (Q3 FY26) vs $17.4B and ABOVE the top of Amazon's own $22.5-26.5B guide; margin 13.9% vs 9.7% Y/Y - operating leverage intact at peak capex
- Q3 FY26 net sales of $201,944M, +12% Y/Y (+13% ex-FX; about four points higher excluding the Prime Day shift into Q2), following the first $200B+ quarter in company history in Q2 ($200,606M, +20% Y/Y). NA $112,852M +6%, operating income $8.7B (7.7% margin, from 4.5%); International $43,108M +5% (+8% ex-FX), operating income $1.6B (3.7%, from 2.9%)
- Trainium demand broadened beyond Anthropic - Amazon names 'the two leading AI labs in the world, Anthropic and OpenAI, making multi-year, multi-gigawatt commitments', plus Uber, Pinterest and a startup roster; chips business >$25B run rate and AI business >$25B run rate, both at triple-digit growth (Q2 FY26 release, not restated in Q3); Graviton5 GA
- New machinery to convert capex into booked revenue: $1B committed to 'AWS Forward Deployed Engineering', AWS Secret Cloud for Industry GA with Northrop Grumman as first classified-workload customer, up to $1B in credits for US Intelligence Community migration
- Q4 FY26 guide: net sales $240.0-247.0B (+12-16%) and operating income $28.5-33.5B vs $25.0B in Q4 2025 - the profit trajectory is guided higher again after Q3 operating income ($28.1B) beat the $22.5-26.5B guide
- Derived gross margin 53.17% (Q3 FY26: $201,944M net sales less $94,573M cost of sales) vs 50.79% a year earlier (Q2 FY26: 52.26%) - Amazon prints no gross-profit line, so this is our derivation from the release
- Contracted backlog: about $578B of commitments in customer contracts, primarily AWS, not yet recognised as revenue at Sep 30 2026 (contracts over one year), with a 6.1-year weighted-average remaining life (10-Q) — multi-year demand visibility set against the capex strain.
The buildout has crossed from compressing cash to consuming it: TTM free cash flow to Sep 30 2026 is an OUTFLOW of $42.4B (vs a $14.8B inflow a year earlier; a $7.6B outflow at Jun 30), roughly -5.3% of TTM revenue, on $215.2B of TTM purchases of property and equipment - 28% more than operating cash flow of $168.3B (computed) - so the 'funded internally' framing published in July no longer describes the situation, and long-term debt reached $173.1B (from $65.6B at Dec 31 2025). FY26 cash capex was raised to ~$220B from ~$200B in July and reiterated on Oct 29, and Jassy named the driver himself - 'the higher cost of memory pushing this number up' - i.e. input-cost inflation, so the extra $20B buys no extra capacity and compresses the implied return on the whole program. First-half reported earnings were majority non-cash ($53,396M of Q2's pre-tax non-operating income came 'primarily from our investments in Anthropic', Amazon's own wording, 3.2x the Q1 FY26 mark); Q3 carried no comparable gain (total non-operating expense of $1,207M) and its EPS fell to $1.89 from $1.95. The valuation cushion is smaller than it was but not absent: $250.69 on Nov 3 2026, ~$2.71T cap, 23.0x FY2027 consensus EPS (estimate, aggregator), against 29.47x forward at the Jul 31 close.
- TTM free cash flow is an OUTFLOW of $42.4B (TTM to Sep 30 2026) against a $14.8B inflow a year earlier — worse than the $7.6B outflow at Jun 30 2026 — with TTM operating cash flow of $168.3B (+29%) now far below capex ($210.7B TTM net of proceeds)
- FY26 cash capex raised to ~$220B in July (Jassy, Q2 call) from ~$200B on 'the higher cost of memory', and reiterated at ~$220B in the Q3 release; Q3 purchases were $77.3B and nine-month purchases $175.7B - the same dollars buy less compute; no FY27 dollar figure is in the release
- GAAP diluted EPS of $5.75 in Q2 FY26 was not an operating result: $53,396M of the $62,647M net income was pre-tax non-operating gain 'primarily from our investments in Anthropic', vs $1,686M a year earlier. Q3 FY26 EPS was $1.89 (vs $1.95) with no such gain (total non-operating expense of $1,207M), but the first nine months' $10.42 still includes about $62.8B of upward adjustments to Anthropic preferred stock (10-Q). Trailing P/E of 20.3x at the Nov 3 close (computed) is distorted DOWNWARD by it
- The Anthropic stake now cuts both ways - a $65B Series H at a $965B valuation, then a Nasdaq listing on Nov 19 2026 (IPO priced at $83, ~$1.5T fully diluted), mean tens of billions of reported 'earnings' now move with a public share price, with zero change in operations; none of it is reflected in the valuation. At Sep 30 2026 the balance sheet carried $92.5B of Anthropic nonvoting preferred stock (marked through earnings) and $146.2B of Anthropic convertible notes at fair value (marked through other comprehensive income, with a $140.3B unrealized gain) (10-Q)
- Q3 FY26 net sales of $201.9B came in at the top of the $197.0-202.0B guide (the soft top-line read was largely the Prime Day shift: about four points of growth); the Q4 guide is $240.0-247.0B (+12-16%)
- A THIRD regulatory front: the US Senate Small Business Committee opened an investigation (22-23 Jul 2026) into alleged Chinese influence in Amazon's marketplace after reports of a bribery network among China-based employees; Republican staff cited 'compelling evidence'. It is FCPA-adjacent and aimed at marketplace integrity - 3P seller services were $46.8B in Q2, ~23% of revenue
- FTC/state marketplace trial now set for Mar 29 2027 (scheduling order signed by US District Judge John Chun, pretrial conference Mar 15 2027 - secondary reporting, docket not pulled); the FTC advertising complaint remains reported-but-unconfirmed
- Jassy's 'clear line of sight to strong financial returns' and a server payback of 'a little less than three years' are SELF-REPORTED on the call - the bull case for the ~$220B now leans on management assertion rather than on realized cash
- A partial cushion, not none: $250.69 (Nov 3 2026 close) against a 52-week range of $196.00-$287.20 - 12.7% below the high and 7.7% below the $271.58 Jul 31 close - ~$2.71T cap, 23.0x FY2027 consensus EPS (estimate, aggregator), against 29.47x forward at the Jul 31 close
What it is worth
Forward multiple + reverse-DCF sanity. Each case is FY2027 consensus EPS ($10.92, estimate, aggregator; GAAP-based, so it can carry Anthropic marks) times a named forward multiple: bear at the cheapest named competitors that, like Amazon, pair a consumer business with a cloud (Alphabet and Alibaba, 18.4x; Oracle, 15.5x, is mainly a database and enterprise-software vendor and is not used), base at the stock's own 23.0x (above the 20.3x median of the six named competitors, so no step up), bull from the 52-week high (26.3x) to the multiple the average analyst target implies (29.9x). Fiscal years are the aggregator's: Oracle ends May, Alibaba Mar, Nvidia and Walmart Jan, Microsoft Jun. At $250.69 (Nov 3 2026 close) / ~$2.71T cap (10.81B shares) the market pays 23.0x FY2027 and 23.3x FY2026 consensus EPS (computed, estimates). Trailing P/E of 20.3x (computed: $250.69 over $12.37 of diluted EPS for the last four quarters, of which $10.42 is the first nine months of FY26) is unusable - it is distorted downward by the first-half non-operating Anthropic marks. At this price the market is paying for AWS holding something close to its +39.3% trajectory, consolidated margin staying near 13.9%, AND a free-cash-flow inflection out of a -$42.4B TTM outflow - three things at once, though the stock now sits 12.7% below its 52-week high rather than at it.
~$201
20% below the $250.69 close: 18.4x FY2027 consensus EPS of $10.92 (estimate, aggregator), the multiple of Alphabet and Alibaba (FY2027 consensus, Nov 3 2026 closes; estimates, aggregator), the cheapest named competitors that, like Amazon, pair a consumer business with a cloud. Oracle (15.5x, fiscal year ends May 2027, ~$169, about a third below) is cheaper but is mainly a database and enterprise-software vendor with no consumer business, so it is not used. The $201 is 2.5% above the 52-week low of $196.00 (Feb 17 2026). The July 31 bear (compression from 29.47x toward the low-20s, roughly a quarter off $271.58, about $204) is 19% below today's close; the stock now trades at 23.0x, near the low-20s it described, so that bear has largely played out in the multiple. Same driver: memory-driven capex inflation persists, FCF stays negative into FY27, the Anthropic mark reverses as ANT trades down from its listing price, and/or a regulatory remedy hits 3P seller services ($46.8B in Q2, ~23% of revenue).
~$251
0% vs the $250.69 close: 23.0x FY2027 consensus EPS of $10.92 (estimate, aggregator), the stock's own multiple. It already exceeds the 20.3x median of the six named competitors (Microsoft 26.0x, Alphabet 18.4x, Nvidia 22.2x, Oracle 15.5x, Walmart 43.2x, Alibaba 18.4x on FY2027 consensus, Nov 3 2026 closes; estimates, aggregator), so there is no step up to a peer median. The July 31 base (a multiple holding near 29x forward, replacing a ~$255-280 band) sat about 6.5 turns above today's. Same driver: AWS settles below the Q3 +39.3% peak but above the +28% published in Q1, consolidated margin holds near 13.9%, and FCF climbs back toward breakeven as capex/sales peaks.
~$287–327
+15% to +30% vs the $250.69 close: 26.3–29.9x FY2027 consensus EPS of $10.92 (estimate, aggregator). Low end: the stock's own 52-week high of $287.20 (26.3x); high end: the multiple the average analyst target implies ($326.74, 59 analysts, estimate, aggregator = 29.9x), close to the 29.47x the stock paid on Jul 31 2026 ($321.81 on today's FY2027 EPS, computed). The July 31 bull held the 29.47x forward multiple on rising earnings at $271.58, 8% above today's close, and called it an earnings-growth case rather than a re-rating case; with the multiple back at 23.0x it is a re-rating case again. Same driver: AWS sustains high-30s% growth (+39.3% in Q3) near the 38.7% segment margin, the ~$220B converts to booked revenue (the $578B contract backlog, Forward Deployed Engineering, Secret Cloud, the Anthropic/OpenAI Trainium commitments), and FCF inflects out of the -$42.4B TTM hole.
The forward multiple is 23.0x FY2027 consensus EPS at the $250.69 close, against 29.47x at the $271.58 close of Jul 31 2026: the stock fell 7.7% while estimates rose (the Jul 31 multiple implied a forward EPS of about $9.22, computed). The average analyst target is $326.74 (59 analysts, S&P Global via stockanalysis.com; estimate, aggregator), range $230 to $405, 30% above the close; the individual post-Q2 targets (Goldman Sachs $375, Benchmark $400, Morgan Stanley $335, Telsey $335) predate the Q3 print and sit inside that range. Judge on operating income; first-half reported EPS is majority-mark, and Q3 EPS ($1.89) is not. FY2026 consensus of $10.76 less the $10.42 reported for the first nine months leaves only $0.34 for Q4 (computed; estimate, aggregator), against $1.89 in Q3 with no such gain, so the consensus EPS used here is not a clean operating measure. Scenario prices are our own arithmetic on consensus EPS and the named multiples - estimates, not a recommendation or target.
SWOT
Strengths
- AWS reaccelerated to +39.3% Y/Y in Q3 FY26 ($46.0B) at a 38.7% segment operating margin (vs ~34.5% Y/Y; Q2: +36.8% at 39.4%) - the profit engine that funds everything else, growing faster on a margin well above last year's
- Operating leverage held through peak capex — consolidated margin 13.9% in Q3 FY26 (vs 9.7% Y/Y) and operating income of $28.1B above the top of Amazon's own $22.5-26.5B guide
- Custom-silicon demand now spans both leading AI labs (Anthropic and OpenAI, multi-year multi-gigawatt commitments); chips >$25B and AI >$25B run rates, both triple-digit growth (Q2 FY26 release); Graviton5 GA
- Scale: Q3 FY26 net sales of $201,944M (+12%) after the first $200B+ quarter in Q2 ($200,606M); NA $112,852M +6%, International $43,108M +5%; ~$797.4B TTM revenue (derived)
- Derived gross margin 53.17% (Q3 FY26) vs 50.79% a year earlier (Q2: 52.26%) - mix still shifting toward higher-margin revenue
- Headcount 1,595k, UP sequentially from 1,575k despite the July AGI-division layoffs - reallocation, not contraction
Weaknesses
- TTM free cash flow is a $42.4B outflow (~-5.3% of TTM revenue) on $215.2B of TTM purchases of property and equipment - the buildout consumes more than operations generate
- The FY26 capex raise to ~$220B is input-cost driven ('the higher cost of memory'), not capacity-driven - the worst kind of increase for return on the program
- Earnings quality — $53,396M of $62,647M in Q2 net income was a non-cash Anthropic mark (none in Q3), so first-half and nine-month EPS and the trailing P/E are unusable without adjustment
- Net debt of ~$58.9B (derived) at Sep 30 2026 — cash and equivalents $55.8B plus marketable securities $58.4B against long-term debt of $173.1B (from $65.6B at Dec 31 2025), excluding lease liabilities and short-term debt
- Q3 FY26 net sales grew only +12% (+13% ex-FX), about four points held down by the Prime Day shift into Q2; North America grew +6%, and the Q4 guide of $240.0-247.0B implies +12-16%
- International segment remains the structurally thinnest margin pool (~3.7% operating margin in Q3 FY26, from 2.9% a year earlier)
Opportunities
- Services-led AI conversion — $1B committed to 'AWS Forward Deployed Engineering' (AI engineers embedded with customers) - addresses the historical delivery gap that leaves AI interest unconverted to consumption
- Classified / defense beachhead — AWS Secret Cloud for Industry GA with Northrop Grumman as first classified-workload customer, plus up to $1B in credits for US Intelligence Community migration - high-margin, sticky, hard to contest
- New named AWS agreements — Warner Bros. Discovery, Vodafone, Siemens Energy, Ryanair, Pinterest, Snowflake, Moody's, Danske Bank, Fiserv, WPP
- Zoox received an NHTSA Part 555 exemption permitting it to charge for rides - the first purpose-built robotaxi to get one; a discrete regulatory unlock in a category to which the valuation ascribes no value
- Amazon Leo reached nearly 400 satellites after four more launches, 'enough to begin initial satellite internet service this year' - moves from research project to a 2026 revenue-recognition storyline
- The Anthropic stake — the Nov 19 2026 listing (ANT; IPO priced at $83, ~$1.5T fully diluted against the $965B Series H) turns an illiquid holding into a publicly priced asset that becomes saleable as lock-up releases come due. Amazon holds 1.936B Class N shares after the offering, with 1.548B more issuable on its remaining notes as its ownership cap allows (Anthropic's final prospectus) — 3.48B shares, ~$338.6B at the $97.17 first close and ~$309.8B at the $88.91 close on 20 Nov 2026 (derived); the final prospectus puts Amazon at 14.6% of outstanding capital stock, 23.6% if its notes converted in full
Threats
- Capex-ROI: ~$220B FY26 against a NEGATIVE TTM FCF of -$42.4B; the offsetting evidence (sub-three-year server payback, 'clear line of sight to strong financial returns') is self-reported by Jassy, not realized cash
- Mark-to-market whiplash on the Anthropic stake in BOTH directions, now through a daily public price: ANT's moves flow into reported earnings and other comprehensive income each quarter with no operational change (a weak listing did not happen: the IPO priced above its range on Nov 18 2026)
- Three regulatory fronts — the Mar 29 2027 FTC/state marketplace trial (secondary source), the reported-but-unconfirmed FTC advertising complaint, and the new Senate marketplace-corruption probe with FCPA-adjacent exposure
- Memory / component cost inflation raising the price of identical compute - it lifted FY26 capex by ~$20B with no capacity gain and could recur
- Valuation risk — 23.0x FY2027 consensus EPS (estimate, aggregator) at $250.69, 12.7% below the 52-week high; the price still needs AWS growth, margin, AND an FCF inflection to all land
- Competitive share at the AI frontier — Azure grew +48% in Microsoft's Q1 FY27 against AWS's +39.3% in Q3; the Google Cloud comparator (~63% published earlier) is not refreshed, so the relative-growth question is open, not resolved
Moats, dependencies & bottlenecks
Moats
~28% global share, deep enterprise lock-in, breadth of services; the cash engine for the whole company.
Western-scale 1P+3P fulfillment + last-mile; replicating it requires decades and tens of billions.
Emerging-to-Wide Lower cost-per-token and reduced NVIDIA dependence; ~$10B+ silicon revenue; advantage compounds with each generation if it holds price/perf lead.
Prime membership retention + first-party commerce data powering a ~$70B high-margin ads business — though ads is the antitrust target.
Equity + $100B+ AWS compute commitment ties a frontier-lab customer to AWS/Trainium; also the source of EPS-distorting mark-to-market.
Dependencies
AI GPUs Still buys large GPU volume despite Trainium; Trainium's purpose is to reduce this dependence over time.
Customer + investee $100B+/10yr AWS commitment is a major demand pillar; concentration + the equity mark distorts reported EPS.
chip fabrication Trainium/Inferentia/Graviton fabbed at advanced nodes; capacity + geopolitical exposure.
ASIC design/IP Annapurna-led design leans on external ASIC/IP partners for tape-out cadence.
energy, land, electrical gear $220B buildout is gated by power availability, grid interconnect, and electrical/cooling equipment lead times.
Marketplace-monopoly trial (early 2027) + reported ads-business complaint; structural-remedy tail risk.
Advantages
- Only hyperscaler that is simultaneously a hyperscaler, a chip designer, AND a top-3 retailer/advertiser — three reinforcing cash pools
- Lowest structural cost-per-token among hyperscalers via first-party silicon
- Internal demand sink (retail, ads, Alexa+, Rufus) absorbs its own AI capacity
- Largest TTM operating cash flow ($168.3B) funds about 80% of the buildout (TTM purchases of property and equipment net of proceeds were $210.7B)
Weaknesses
- TTM FCF an outflow of $42.4B on capex — least FCF cushion among mega-cap peers right now
- AWS growth rate trails Azure and Google Cloud at the headline level
- High capex-ROI uncertainty; depreciation drag building
- Antitrust overhang on both marketplace and ads
Bottlenecks
- Power and grid interconnect for AI data centers — the binding constraint on capex deployment
- Advanced-node fab capacity (TSMC) for custom silicon
- Free-cash-flow generation while capex runs at ~$220B/yr
- Time-to-monetize AI compute relative to the depreciation schedule it creates
Top signals & trends
Top signals
AWS grew +39.3% in Q3 FY26 to $46.0B — fastest in nearly five years, ~$184B annualized run rate, after +36.8% in Q2 — with segment operating margin at 38.7% vs ~34.5% a year earlier (39.4% in Q2). Jassy projects AWS growth of over 40% in both 2027 and 2028 (management projection, not guidance). The core bull thesis is stronger than at the +28% print of Q1 FY26.
TTM free cash flow is an OUTFLOW of $42.4B (TTM to Sep 30 2026) against a $14.8B inflow a year earlier, on $215.2B of TTM purchases of property and equipment. The capex bite has gone from 'compressed' to 'cash-consuming', and long-term debt reached $173.1B (from $65.6B at Dec 31 2025).
Mixed (tilting bearish) · Raised to ~$220B for FY26 (Jassy, Q2 call): 'the higher cost of memory pushing this number up from our prior estimate of about $200 billion.' The reason matters more than the number — this is input-cost inflation, so the extra $20B buys no extra capacity and compresses the implied return on the buildout. Bull case now rests on Jassy's self-reported 'clear line of sight to strong financial returns' and a sub-three-year server payback, plus his claim that capacity still falls short of demand in 2026 and 2027; bear case gets a cheaper argument than before.
Broadened beyond the Anthropic-centric read: Amazon now names 'the two leading AI labs in the world, Anthropic and OpenAI, making multi-year, multi-gigawatt commitments' to Trainium, plus startups (NEURA Robotics, Odyssey, TwelveLabs, Decart, Poolside, Karakuri, Metagenomi, NetoAI, Splash Music) and larger firms (Uber, Pinterest). Chips and AI businesses each >$25B run rate at triple-digit growth; Graviton5 GA.
Refined and widened to a third front. Trial is set for Mar 29 2027 per a scheduling order signed by US District Judge John Chun (pretrial conference Mar 15 2027) — secondary reporting, docket not pulled. The ads complaint remains reported-but-unconfirmed. NEW: the US Senate Small Business Committee opened an investigation (Jul 22–23 2026) into alleged Chinese influence in Amazon's marketplace, following reports of a bribery network in which China-based employees allegedly took payments from sellers to lift suspensions and restore listings; Republican committee staff said they found 'compelling evidence'. That attacks marketplace integrity — 3P seller services were $46.8B in Q2, ~23% of revenue — and carries FCPA-adjacent exposure distinct from antitrust remedy risk.
Mixed — part of the post-Q2 rally has been given back · AMZN closed at $271.58 on Jul 31 2026, +15.32% on the day post-print, and at $250.69 on Nov 3 2026 (7.7% lower; 12.7% below a 52-week high of $287.20). It had rallied into a ~20.6% July drawdown in the PHLX Semiconductor Index, which argues against reading generic AI-hardware sentiment through to Amazon. Market cap ~$2.71T at Nov 3; 23.0x FY2027 consensus EPS (estimate, aggregator), against a 29.47x forward P/E on Jul 31. Part of the de-rating that framed the 'bulls see entry' argument has returned since the print, so valuation discriminates again, though the stock is still 27.9% above its 52-week low.
Trends
Hyperscaler 2026 capex collectively ~$650-725B; Amazon ~$220B of it. Powers AWS growth but compresses near-term FCF across the cohort.
Trainium/Inferentia (and Broadcom-style ASICs for peers) erode NVIDIA's share of incremental AI spend; favors vertically integrated Amazon.
AWS 28% / Azure 21% / GCP 14% (Q1 2026); Azure & GCP grow faster off smaller bases — the gap is the watched metric.
High-margin ads (~$70B) is a structural margin tailwind — but now the antitrust target.
FTC marketplace suit + ads complaint; EU scrutiny — a multi-year overhang.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
AI GPUs for AWS instances (being partially displaced by Trainium).
Fabricates Trainium/Inferentia/Graviton custom silicon (Taiwan-domiciled; named as supply-chain fact).
Custom-ASIC / interconnect design partner for hyperscaler silicon.
Networking/ASIC supplier to the hyperscaler buildout broadly; also enables rival custom silicon.
Data-center power and thermal-management equipment for the AI buildout.
$100B+/10yr AWS + Trainium commitment, up to ~5GW; flagship AI customer + investee.
Long-standing flagship AWS reference customer.
Major enterprise AWS workload customer.
Representative consumer-internet AWS customer.
Azure ~21% share growing ~40% on OpenAI; the primary cloud + enterprise-AI rival.
~14% share growing ~63% on Gemini/TPUs; fastest grower, also a custom-silicon peer.
Supplier and rival — Trainium is built to take share of AI spend away from NVIDIA GPUs.
OCI gaining via AI-training contracts and aggressive capacity deals.
Primary U.S. retail + retail-media competitor; closing the e-commerce/ads gap.
Cloud + commerce leader in Asia (mainland-China-domiciled — named for competitive analysis only, NOT a buy/own recommendation).