Amazon's earnings power is real, concentrated, and — in one important place — legally contested. Two high-margin engines, AWS and advertising, are growing fast enough to justify an investment program of extraordinary size, while the low-margin retail base supplies the traffic, the sellers, and the purchase data that make both engines valuable. The problem is not demand. The problem is measurement and durability. A federal lawsuit alleges the advertising auction at the heart of the profit mix was quietly inflated, and the capital program is large enough to drive projected free cash flow negative in 2026. The history of advertising is a history of unmeasured waste. The question is not whether Amazon's machine works. It is how you know what each dollar of it truly earns — and whether the pricing mechanism survives scrutiny intact.
Three facts frame everything below. Advertising contributes only about 10% of group revenue 40 but carries an estimated roughly 70% EBIT margin, against roughly 35% for AWS and low single digits for retail 27. Capital expenditure is cited at about $220 billion for 2026 21,46, with negative free cash flow projected for that year 15,32. And the FTC, joined by 22 state attorneys general, alleges undisclosed auction surcharges cost about 1.2 million advertisers more than $20 billion since 2019 32.
1) Introduction
Amazon operates as a dual-engine conglomerate. AWS converts cloud demand into high-margin operating income. Retail converts heavy capital spending into traffic, seller dependence, and the first-party data that feeds advertising. The engines diverge structurally: AWS margins run near 35% and retail near low single digits 27, and the current investment cycle runs through AI and AWS infrastructure rather than fulfillment bricks.
Bases and labeling. Figures are drawn from the supplied evidence set — reported quarterly results, company statements, legal filings, and named third-party commentary (Evercore on 2027 capex; Citi on advertising). Each figure below is reported, estimated, or projected, and the label travels with the number.
Known gaps, stated up front:
- Data unavailable: North America and International segment revenue and operating income; the evidence quantifies AWS and advertising only.
- Data unavailable: consolidated operating cash flow, free cash flow actuals, absolute net debt, interest coverage, credit ratings.
- Data unavailable: a like-for-like peer dataset for debt, cash flow, or valuation ratios; peer benchmarking is therefore partial and flagged.
- The 2025 profit record appears under two definitions — $80 billion of operating income 35 and $77.67 billion of reported profit 50 — and the material does not establish that the definitions match. They are treated here as non-interchangeable.
2) Financial Performance
Revenue. AWS produced $42.23 billion in the June 2026 quarter 17,46, reported as 37% year-over-year growth from $30.9 billion 46. Advertising reached $68.6 billion in 2025, up 22% for the year 40,42, then accelerated: $17.2 billion in Q1 2026, up 24% 11,31,33,40,47, and $19.8 billion in Q2 2026, up 26% 16,18,19,40. That second-quarter pace annualizes to roughly $79 billion 33. Advertising remains about 10% of group revenue 40 — a thin slice of the top line carrying an outsized share of the profit logic.
Margins. The margin ladder is the crux of the valuation. Advertising carries an estimated roughly 70% EBIT margin, compared with roughly 35% for AWS and low single digits for retail 27. The advertising figure is an estimate, not a reported segment margin; it is consistent, however, with broader characterizations of advertising as one of Amazon's highest-margin businesses 10,25,37,41. On those estimates, a dollar of advertising revenue is worth about twice a dollar of AWS revenue in EBIT terms, and something like twenty times a retail dollar. Assessment, labeled as such: with advertising accelerating across the last three reported periods, consolidated margins should drift upward — provided the legal challenge does not reach the pricing mechanism (Section 5).
Profit. One account places 2025 total operating income at $80 billion 35. A second places 2025 reported profit at $77.67 billion, following $59.25 billion in 2024 and $30.43 billion in 2023 50. The trajectory is steep on either definition. The definitions may not match; I do not blend them.
Cash flow. Capital expenditure is cited at about $220 billion for 2026 21,46; Evercore projects $320 billion for 2027 32; free cash flow is projected negative for 2026 15,32. Assessment: negative free cash flow here reads more plausibly as an investment consequence than as a demand signal, because management states AWS lacks sufficient capacity for both 2026 and 2027 46. That interpretation is comfort, not proof. The test is return on the capacity, not the size of the commitment.
Debt and liquidity. Parent-level debt-to-equity sits below 0.3x 23 — conservative by any standard. For financing-environment context, not Amazon's own balance sheet: AI-related debt issuance across the industry has reached roughly $220 billion this year against $12.5 billion in the prior-year period 23. Data unavailable: Amazon-specific net debt, maturity schedule, interest coverage, credit rating.
Key figures from the supplied evidence:
| Metric | Value | Period | Basis |
|---|---|---|---|
| AWS revenue | $42.23B, +37% YoY 17,46 | Q2 2026 | Reported |
| Advertising revenue | $17.2B, +24% YoY 11,31,33,40,47 | Q1 2026 | Reported |
| Advertising revenue | $19.8B, +26% YoY 16,18,19,40 | Q2 2026 | Reported |
| Advertising revenue | $68.6B, +22% 40,42 | FY2025 | Reported |
| Operating income | $80B 35 | FY2025 | Reported (definition A) |
| Reported profit | $77.67B; FY2024 $59.25B; FY2023 $30.43B 50 | FY2023–2025 | Reported (definition B) |
| Debt-to-equity | Below 0.3x 23 | Parent level | Reported |
| Capital expenditure | ~$220B 21,46 | 2026 | Cited |
| Capital expenditure | $320B 32 | 2027 | Evercore projection |
| Free cash flow | Negative 15,32 | 2026 | Projection |
| Market capitalization | $3 trillion 36 | Aug 3, 2026 | Reported |
Data unavailable for this table: North America and International segments, consolidated net income on one consistent definition, operating cash flow, free cash flow actuals, total and net debt.
3) Earnings & Guidance
The last three reported periods show acceleration in both engines. Advertising grew 22% in FY2025 40,42, 24% in Q1 2026 11,31,33,40,47, and 26% in Q2 2026 16,18,19,40. AWS grew 37% in the June 2026 quarter 46. Data unavailable: consensus estimates, actual-versus-expectation comparisons, and formal forward guidance. The supplied material contains none.
Management commentary, as evidenced: Andy Jassy states AWS lacks sufficient capacity for both 2026 and 2027 46 — capacity, not demand, is the binding constraint. Strategy follows the constraint. Amazon is reported to be allocating $200 billion to AI infrastructure, primarily AWS capacity serving AI companies including OpenAI, Anthropic, and Google 50, and plans to add 2 million NVIDIA GPUs globally for agentic and physical AI workloads 24. On the retail side, March 2026 seller-fee cuts intended to attract volume 26 signal that management sees merchant health as a limit on near-term extraction.
Momentum assessment: both high-margin engines are accelerating. The cash cost of that acceleration is expanding faster than cash generation, which is what projected negative free cash flow means 15,32. Whether the arithmetic resolves in shareholders' favor depends on monetization per unit of capacity, not on the size of the order book for GPUs.
4) Ratios & Peer Benchmarking
The evidence supports only a partial benchmark. I fill what it supports and mark the rest.
Valuation. Market capitalization reached $3 trillion on August 3, 2026 36. Data unavailable: P/E, EV/EBITDA, EV/FCF, and historical-range comparisons.
Profitability and market position. Amazon is the world's third-largest digital advertising business, behind Google and Meta 48, and analysts estimate it captures roughly 75% of U.S. retail-media spending 48. That retail-media share is the modern equivalent of owning the back page of the catalog: the advertiser is already in the aisle. Retail-media network economics of this kind monetize traffic Amazon has already paid to acquire. Data unavailable: ROE, ROIC, and like-for-like margin or growth comparisons for Microsoft, Walmart, Alphabet, and Meta.
Leverage. Parent-level debt-to-equity below 0.3x 23. Data unavailable: net debt/EBITDA, interest coverage, peer leverage ratios.
Benchmark table — supported cells only:
| Metric | Amazon | Peer comparison |
|---|---|---|
| Digital advertising rank | Third, behind Google and Meta 48 | Google and Meta ahead 48 |
| U.S. retail-media spend share | ~75% (analyst estimate) 48 | Data unavailable |
| Cloud growth | AWS +37% YoY, Q2 2026 46 | Azure and Google Cloud: data unavailable |
| Debt-to-equity | Below 0.3x 23 | Peer dataset: data unavailable |
| EBIT margin, advertising | ~70% (estimate) 27 | Google and Meta ad margins: data unavailable |
| EBIT margin, AWS | ~35% 27 | Data unavailable |
| EBIT margin, retail | Low single digits 27 | Walmart margins: data unavailable |
| Market capitalization | $3T 36 | Data unavailable |
A note on honesty: the evidence offers no like-for-like peer debt, cash-flow, or valuation dataset. Any firm claim that Amazon beats Microsoft or Walmart on a specific ratio would be fabrication, and I decline to make it. What the evidence does support is the internal comparison — the margin ladder inside Amazon's own P&L 27 — and that ladder explains why advertising carries so much of the valuation argument despite being about 10% of revenue 40.
5) Management & Governance
Leadership. Andy Jassy moved from running AWS to running Amazon. The supplied material documents the strategic posture but contains no evidence of management-team changes, and it offers no basis for a definitive assessment of leadership effectiveness — flagged as a gap. What it does show is a consistent expansionary stance: a capacity shortage acknowledged through 2027 46, $200 billion directed at AI infrastructure 50, 2 million GPUs planned 24. In direct-mail terms, management is buying the largest list in the category because response is capacity-constrained — not because response is proven at this spend level. That distinction is the whole investment case.
Governance: the FTC case. The FTC and 22 state attorneys general allege Amazon inflated auction prices through undisclosed surcharges, costing about 1.2 million advertisers more than $20 billion since 2019 32. The complaint followed a six-year investigation that reportedly reviewed approximately 1.5 million pages of documents 39. These are allegations, not findings. Amazon denies them 34 and argues its internal documents do not demonstrate a widespread deceptive strategy 45. Amazon further states that relevance-weighted ranking saved advertisers more than $8 billion between 2021 and 2025 40,42,43,47 — a figure that originates in Amazon's own statements 44. That claim requires evidence that is not yet public. The competing numbers cannot be reconciled from the available material.
The structure of the case matters more than the headline. Potential remedies include restitution, civil penalties, disgorgement, advertiser compensation, and mandated practice changes 41, including possible elimination of floor-bidding strategies 30. A one-time payment is absorbable against $80 billion of operating income 35. A redesign of the auction would reach recurring pricing power in a business estimated at roughly 70% EBIT margin 27. A retailer knows the difference between a fine and a new cash-register ruleset. One is an expense. The other changes what the register rings. The attribution question runs through all of it: if the auction price was inflated, every advertiser ROI calculation built on that price inherited the error.
Marketplace two-sidedness. U.S.-based independent sellers average more than $375,000 in annual Amazon.com sales 13,29, and their profitability must absorb inventory, shipping, platform fees, advertising, returns, and working-capital demands 28. The March 2026 fee cuts, intended to attract volume 26, suggest management recognizes that merchant health and marketplace liquidity can conflict with maximizing near-term take. Commentary expects muted advertiser reaction to the litigation given Amazon's dominant position 48, and Citi describes the advertising business as intact 32. Assessment: the risk is prospective and structural, not evidence of current operating deterioration. But dependence is exactly why auction transparency is a governance issue and not merely a legal one. Data unavailable: board composition changes and compensation structure figures.
6) Capital Allocation
The record is unambiguous. Amazon has historically paid no dividend 1,2,12,20,36,49. Its $5 billion Class A repurchase authorization stood unused as of the second quarter 22. Capital flows instead into infrastructure: about $220 billion of 2026 capex 21,46, a reported $200 billion AI allocation 50, 2 million NVIDIA GPUs 24, $320 billion projected for 2027 32, against projected negative free cash flow in 2026 15,32.
The Anthropic relationship shows both the strength and the exposure of this approach. Amazon is described as a major investor in Anthropic 3,4,5,6,7,8,9,14,38, while Anthropic is also an anchor customer for the compute-leasing model 24. The diversification is sound: AWS can earn from cloud workloads even if no single model provider retains dominant economics. But the exposure runs through model-provider health. Anthropic has reported losses of roughly $11 billion amid high capital and operating costs 51. A lessee losing money at that scale does not invalidate the leasing model — capacity is genuinely scarce 46 — but it concentrates the question of who ultimately pays for the compute. Note also that industry-wide AI-related debt issuance has reached about $220 billion this year versus $12.5 billion in the prior-year period 23; financing conditions belong in any capital-allocation picture, even though the supplied material does not tie that issuance to Amazon's own balance sheet.
Assessment: reinvestment-led allocation is defensible while AWS demand exceeds capacity 46. It leaves shareholder returns dependent on execution and future cash conversion rather than distributions. The $3 trillion market capitalization reached on August 3, 2026 36 expresses confidence in that model. Confidence is not a cash flow.
7) Risks & Catalysts
Risk 1 — AWS growth deceleration versus Azure and Google Cloud. The evidence shows AWS at 37% quarterly growth 46 and capacity constrained through 2027 46. Data unavailable: comparable Azure or Google Cloud growth figures. The risk is structural — competitors are chasing the same demand — and cannot be quantified against peers from this material.
Risk 2 — Regulatory and antitrust action. The FTC suit carries remedies spanning restitution, civil penalties, disgorgement, advertiser compensation, and mandated practice changes 41, with possible elimination of floor-bidding strategies 30. The conduct remedies are the dangerous category for a business estimated at roughly 70% EBIT margin 27: a cash settlement ends; a repriced auction compounds.
Risk 3 — Retail margin compression. Retail margins sit in the low single digits 27. Sellers already absorb a heavy cost stack 28, and Amazon's own fee cuts 26 trade near-term take for volume. Labor or inflation pressure leaves little cushion in that structure.
Catalyst 1 — AWS reacceleration from generative AI. Capacity is the stated constraint for 2026 and 2027 46. The $200 billion AI allocation 50 and the 2 million GPUs 24 are the direct response. Every unit of capacity added converts constraint into revenue if demand holds.
Catalyst 2 — Advertising monetization. Growth accelerated from 22% in FY2025 40,42 to 24% in Q1 2026 11,31,33,40,47 and 26% in Q2 2026 16,18,19,40, an annualized pace near $79 billion 33. Headroom is plausible — roughly 75% of U.S. retail-media spend 48 and third place overall 48 imply formats beyond retail media — but that headroom is inference, not supplied data.
Catalyst 3 — Retail cost-structure optimization. The fee cuts 26 and seller economics 28 point to a volume-for-take trade that could lift marketplace liquidity. Fulfillment efficiency metrics: data unavailable.
8) Investment Implications
Does the valuation reflect a sum-of-parts — an AWS premium, an advertising premium, a retail discount? The material supplies no segment multiples, so I cannot compute one; a precise answer would be invented. What the evidence supports is directional. The margin ladder 27 says consolidated earnings increasingly come from advertising and AWS. Capex of about $220 billion for 2026 21,46 and a $320 billion projection for 2027 32, with negative free cash flow projected 15,32, mean the market is funding capacity ahead of cash returns. The FTC case 32 says the pricing mechanism of the highest-margin business is contested.
Four questions for deeper research:
- What is the actual AWS versus Azure and Google Cloud share trajectory in generative AI workloads, and what does each dollar of capacity earn? Comparable peer growth: data unavailable.
- Where does advertising go from 26% quarterly growth 16,18,19,40, and how much of the ~$79 billion run rate 33 is reachable without the pricing practices under challenge?
- What are realistic FTC settlement scenarios — cash-only versus conduct remedies — and what does each do to an estimated ~70% EBIT margin 27 and to advertiser cost-per-acquisition integrity?
- Does capex efficiency improve — revenue per GPU, per fulfillment dollar — or does the waste fraction simply scale with the budget? Evercore's $320 billion for 2027 32 assumes the spending continues. It does not prove the spending pays.
Closing entry, in the old ledger tradition: a business growing this fast, with its cash register on trial and its spending at record scale, deserves attention and skepticism in equal parts. The question is not whether Amazon's advertising works. It is how you know it works — and what the answer is worth if a court reprices it.
Appendix — Calculations and Figure Bases
Derived figures (my arithmetic on supplied numbers):
- AWS YoY growth, Q2 2026: ($42.23B − $30.9B) ÷ $30.9B = $11.33B ÷ $30.9B ≈ 36.7%, reported as 37% 46.
- Advertising annualized run rate, Q2 2026: $19.8B × 4 ≈ $79.2B, consistent with the ~$79 billion cited 33.
- Profit trajectory, definition B: FY2023 $30.43B → FY2024 $59.25B (+94.7%) → FY2025 $77.67B (+31.1%) 50.
- Capex intensity, illustrative and cross-period: 2026 capex of ~$220B 21,46 against FY2025 operating income of $80B 35 is roughly 2.75x one year of operating income. Different periods and possibly different definitions; treat as a scale indication only.
- Margin ladder illustration, per $100 of revenue at the estimated margins: advertising ≈ $70 of EBIT, AWS ≈ $35, retail ≈ $1–$3 27. Estimates, not reported segment margins; the advertising estimate is consistent with characterizations of advertising as among Amazon's highest-margin businesses 10,25,37,41.
- Advertising acceleration: +22% FY2025 40,42 → +24% Q1 2026 11,31,33,40,47 → +26% Q2 2026 16,18,19,40.
Bases and gaps. Figures labeled reported come from quarterly results or company statements as supplied; estimated figures (advertising EBIT margin, U.S. retail-media share) are analyst or commentary estimates; projected figures (2026 and 2027 capex, negative 2026 free cash flow) are forward-looking statements from management or third parties. Data gaps carried through the text: North America and International segment detail; consolidated cash-flow actuals; absolute debt and coverage ratios; credit ratings; consensus comparisons; peer ratio datasets; Prime membership economics; fulfillment cost as a percentage of sales; management-team-change and board/compensation documentation.