Business

Know the Business — Amazon.com, Inc.

Amazon is not one business — it is a low-margin retailer wrapped around a hyperscaler and a retail-media engine, with the storefront acting as a marketing surface for the services that actually make the money. Most of the operating income comes from a slice of revenue that gets miscategorized as "the cloud thing on the side." The recurring valuation error is putting one multiple on the consolidated company; the 2026-specific error is assuming the $200B capex bet earns the same returns the last $300B of AWS capex did.

1. How This Business Actually Works

Amazon reports three segments — North America, International, AWS — but the economic engine is better described as two profit pools subsidising one storefront. Retail (NA + International) generated 82% of FY2025 revenue but only 43% of operating income. AWS produced 18% of revenue and 57% of operating income at a 35% margin. Advertising (~$70B run-rate, growing 22%) is buried inside the retail segments and is the highest-margin large business inside the company.

The flywheel inside retail is well-rehearsed: low prices and Prime delivery pull customers, traffic creates first-party intent data, that data sells sponsored-product ads at 40–50% margins, ad profits fund faster shipping, shipping deepens Prime, Prime drives basket frequency. The under-told part: 3P seller fees and advertising — both byproducts of the storefront, not the storefront itself — do most of the heavy lifting on retail-segment margin. 1P direct retail in FY25 runs at a low-single-digit operating margin; 3P fees (~30% take-rate on GMV) and ads (40%+) produce the 6.9% North America blend.

AWS is a separate engine, not a retail extension — enterprise compute, storage, database, and AI services priced per CPU-hour, GB, and API call. The bottleneck is no longer chips broadly; it is data-center power and grid interconnect. FY25 capex ran $132B and is guided to roughly $200B in 2026, "predominantly AWS." The economic test is whether each dollar of capex returns 15%+ on incremental revenue at AWS's 35% margin; management says yes, the build is "demand-led," and the $244B AWS backlog (up 40% YoY) is the public evidence.

What truly drives incremental profit:

Bargaining power runs in Amazon's favor everywhere except labor and content: 3P sellers absorb most fee increases (FBA is sticky), AWS contracts are multi-year with high migration costs, ad budgets follow purchase intent that Amazon owns. Where it does not: price competition for hard-discount staples (Walmart and Costco set the floor) and enterprise AI distribution (Microsoft's Office and OpenAI relationships open more doors than AWS account teams). SBC at $19.5B in FY25 is 2.7% of sales — not extreme by software standards, but a real, recurring transfer that headline operating margin doesn't fully capture.

2. The Playing Field

Amazon doesn't have a single peer; it has different peers per layer. Walmart is the only same-scale retailer. Microsoft is the only same-scale cloud rival. Meta and Alphabet are the closest comps for advertising. Costco is the cleanest comp for Prime's unit economics. The table is the right peer set for enterprise-level valuation; no single row tells the AMZN story alone.

Amazon's blended operating margin (11.2%) sits between the retail floor at 4% (WMT/COST) and the platform plateau at 32–46% (GOOGL/MSFT/META). Its EV/Sales of 4.3x lands roughly halfway between Walmart at 1.5x and Microsoft at 11.1x. The market is implicitly pricing about a third of Amazon as a hyperscaler, two-thirds as something better than Walmart but not yet platform-quality. That is the central valuation question of section 5.

What the peers reveal that consolidated AMZN does not:

• Microsoft's Azure operating margin is materially higher than AWS's (segment disclosures imply Azure-cloud at 40%+ vs AWS at 35%). Microsoft sells software with compute; Amazon sells compute alone. AMZN's response is custom silicon + the broadest service catalog, but the margin gap is structural unless AWS finds a software bundle.
• Walmart converts a 4% operating margin into a 21% ROE, almost identical to Amazon's 19% ROE on triple the margin. Walmart's edge is asset turnover, not pricing — store-based fulfillment is more capital-efficient than warehouse-only fulfillment for grocery and bulky goods.
• Meta runs ads at 41% operating margin with no inventory or fulfillment to feed. That is the upper bound on what Amazon Advertising could earn if separated. Embedded inside retail today, ad margins are diluted by everything around them.
• Costco's 93%+ membership renewal sets the bar that Prime (mid-80s, by triangulated estimates) needs to clear. Membership economics are the cleanest read on customer lock-in, and Costco's model — fewer SKUs, higher trip value, physical bulk — is durable in ways online-first models still need to prove.

"Good" in this peer set: a 30%+ AWS margin held through the AI capex cycle, a 7%+ North America retail margin held through a normal demand year, ad revenue growing 20%+, and capex/D&A converging from 2x toward 1.2x. Less than that puts Amazon expensive versus the retail comp set; more puts it cheap versus the platform comp set.

3. Is This Business Cyclical?

Amazon runs three different cycles at three different frequencies; the consolidated number averages them out in misleading ways.

The 2014 and 2022 troughs are the canonical "over-built into a normalizing market" episodes. Demand went from accelerating to merely growing, and the fixed fulfillment cost layer (warehouses, sortation, last-mile) didn't shrink fast enough. Operating margin halved within two quarters; recovery took 18–24 months and a layoff cycle. The 2024–25 margin breakthrough to 10–11% is the post-rationalization regime — regionalized US fulfillment, robotics deployment, and headcount discipline hitting at once.

The 2021–22 build absorbed roughly $20B of cumulative negative FCF. The current cycle is bigger — capex jumped from $83B (FY24) to $132B (FY25), with $200B guided for FY26. This is the central cyclical question for the stock. If AWS revenue keeps compounding at 20%+ and AI-driven workloads keep adding to backlog, the build pays back through depreciation absorption against rising revenue. If AI demand decelerates before capex lands as D&A, Amazon runs a 2014-style margin trough on a much bigger fixed-cost base.

The cycle hits margin before revenue. Watch North America operating margin (currently 6.9%, target 7–8%) and AWS operating margin (currently 35%, sustainable floor 30%) more closely than total revenue.

4. The Metrics That Actually Matter

Five operating metrics carry most of the information about whether this business is creating or destroying value. Generic ratios (P/E, EV/EBITDA, ROE) describe price; these describe the engine.

Two readings stand out. First, Amazon's FCF margin (1.6%) is the lowest in the peer set — below the bricks-and-mortar retailers — because $132B of cash capex outran $66B of D&A. This is mechanical; FCF is structurally suppressed during a build phase. Second, Amazon's capex/sales (18%) is below the platform peers (MSFT 23%, GOOGL 23%, META 35%) but far above the retailers (WMT 4%, COST 2%). The 2026 guide takes capex/sales toward 25%+, moving Amazon firmly into platform-capex territory. The tension: management says revenue will follow capex; the bear says capex is decoupled from short-term revenue and depreciation will arrive faster than the AI bookings.

Consolidated gross margin (50.3%) is not a useful metric here. It blends physical-goods cost (~70% gross margin sink) with software-style services (~80%+ gross margin) and should be ignored in favor of segment operating margins.

5. What Is This Business Worth?

The right valuation lens is sum-of-the-parts, not because Amazon is a holding company, but because the consolidated financials blend Walmart-economics with Microsoft-economics inside a single share class, and a single multiple cannot honestly weight both. The company itself does not break out advertising, third-party services, subscription, and AWS-AI separately, so the SOTP below is necessarily directional, but the framework is the right one.

What ultimately determines value here is the pace and quality of AWS revenue growth at sustainable 30%+ margins, plus whether North America retail can hold a 7%+ operating margin through a normal demand cycle, plus whether the $200B+ annual capex bet earns 15%+ unlevered returns. Everything else (Whole Foods, devices, LEO, physical stores) is a small option, not a value driver.

What would make the multiple look undemanding: AWS revenue holding 25%+ YoY for multiple quarters with operating margin at 35%, advertising sustaining 20%+ growth toward a ~$100B run-rate, North America retail margin reaching 8%, and capex/D&A converging back toward 1.5x by 2027. Any three of four checks the box.

What would make it look stretched: AWS revenue decelerating below 18% with margin slipping below 32%, advertising growth slipping into the teens, North America retail margin retracing toward 5%, capex/D&A staying above 1.8x for multiple years. Any two and the SOTP collapses fast.

6. What I'd Tell a Young Analyst

Don't value Amazon as one company. The retail you see and the cloud you don't see have nothing in common economically. Build a SOTP every quarter; track each part on its own metrics; ignore consolidated gross margin entirely.

The single most important number is the gap between AWS capex and AWS revenue growth. When capex/D&A is rising and AWS revenue growth is decelerating, you are watching margin compress in slow motion. When both rise together, the build is working. Right now both are rising. That is the bull case in one fact, but it requires the AI demand to keep showing up.

Watch advertising growth more than retail revenue growth. Ads are growing 22% on a $70B+ base at 40%+ margins inside a segment whose blended margin is only 7%. Each percentage point of ad-mix acceleration is worth more than a full point of total retail growth. The bear scenario nobody is pricing yet: AI shopping agents (ChatGPT, Perplexity) bypass Amazon search and erode the highest-margin business. That is the most underrated terminal-value risk in the stock.

Stock-based compensation is real cash to employees and dilutive to you. $19.5B in FY2025. Subtract it from operating income before judging margin progress. The "headline" 11.2% operating margin is closer to 8.5% post-SBC. Most peers have the same gap; the comparison still works, but use the post-SBC number when you talk about cash-generative quality.

The thesis can hinge on one number — quarterly AWS growth. Q3 FY25 was 17.5%; Q4 was 24%. That re-acceleration drove this year's multiple expansion. A print below 18% with backlog flat would make the SOTP math ugly: the AWS multiple compresses and the capex bill is already locked in.

Three things that would invalidate the bull case: (1) AWS revenue growth below 18% for two consecutive quarters with backlog decelerating, (2) North America retail margin below 5% on a normal demand year, (3) clear evidence that AI-agent traffic is taking measurable share of product discovery from Amazon search. None of those, the cross-layer flywheel is intact; any of them, the case for waiting outweighs the case for owning.