Business

Know the Business

Amazon is not one business — it is a high-volume, low-margin retail logistics machine bolted onto a hyperscaler and a digital-ad network, and the second two now produce most of the profit. AWS contributes 18% of revenue but 57% of operating income; advertising — over $80B annualized — is the silent third leg the market still under-counts. The most likely market mistake is anchoring on consolidated margins (about 11%) and missing that the durable economic engine is AWS plus ads, while first-party retail is essentially a fee-collection platform for third-party sellers.

How This Business Actually Works

Amazon runs three economically distinct businesses under one P&L. Look at where revenue and profit come from — the gap between the two reveals the engine.

The retail businesses generate most of the revenue but almost none of the incremental profit — North America earns about 7 cents per dollar of sales, International about 3. AWS earns 35. So the question is not "is Amazon a good retailer" but "what does the retail business actually do for Amazon."

Three things, mainly. First, it builds Prime, the subscription that locks the customer into the ecosystem and converts them into a high-frequency shopper. Second, it builds traffic — about 600 million shopping visits a month — that the ads business sells back to merchants ($21.3B in Q4 alone, growing 22%). Third, third-party sellers now do 61% of unit volume, paying Amazon commissions, FBA fulfillment fees, and ad placement fees — so a large slice of "retail" economics is actually a take-rate on someone else's inventory. First-party retail itself is a thin-margin operating expense to manufacture customer attention; ads and 3P seller fees are how that attention is monetized.

AWS is a different business entirely: long-duration enterprise contracts (backlog $244B, +40% YoY), capital-intensive but with strong unit economics once a data center is full. Each marginal hour of EC2 utilization drops largely to operating profit. The constraint is no longer demand — management said in Q4 they are "monetizing capacity as fast as we can install it" — but power, GPUs, and the lead time on new data center capacity. That bottleneck is why capex is doubling.

The Playing Field

Amazon's peer set is unusual: it competes against hyperscalers, against retailers, and against ad platforms, and looks unlike any single one of them. The peer table makes the hybrid nature concrete.

Amazon sits in a no-man's land between two clusters. The hyperscalers (MSFT, GOOGL, META) earn 32–46% operating margins on lower revenue — they have software economics. The retailers (WMT, COST, TGT) earn 4–5% on similar or larger revenue — pure retail economics. Amazon's blended 11% margin reflects that AWS is dragged down by retail, and retail looks better than it should because of AWS's dollar contribution. The right comparison is not consolidated; it is segment-by-segment: AWS's 35% margin is below Azure's mid-40s but ahead of Google Cloud, and Amazon Ads' growth rate (22%) and likely margin profile sit firmly inside the META/GOOGL band.

What "good" looks like in this peer set — the trait Amazon does not yet match — is FCF conversion. MSFT, GOOGL, and META each run capex/revenue near 23–35% and still deliver 18–25% FCF margins because their underlying gross margins are 70%-plus. Amazon delivered just 1.1% FCF margin in FY25 because retail's gross margin is 50% and it spent $128B on capex. The market is paying 32× earnings on the bet that AWS's per-unit economics — once new capacity fills — will pull the consolidated FCF margin sharply higher.

Is This Business Cyclical?

Amazon is cyclical, but not in the way most retail names are. The cycle hits through capex and utilization, not gross margin — every five-to-seven years Amazon over-builds capacity, free cash flow collapses, and operators then monetize that capacity into multi-year operating leverage.

There have been three FCF collapses in the past decade: 2014 (logistics overbuild ahead of Prime expansion), 2021–22 (COVID-era doubling of fulfillment plus AWS infrastructure), and now 2025–26 (AI infrastructure). Each was followed within 18–24 months by sharp operating-margin expansion as new capacity filled. The current cycle is the largest in dollar terms by far — capex roughly $128B in 2025, guided to about $200B in 2026 — and the recovery thesis depends on AWS demand absorbing that capacity at attractive returns. The transcript provides one quantitative anchor: AWS backlog rose 40% year-on-year to $244B, suggesting demand is running well ahead of installed capacity for now.

The retail business is also cyclical in the conventional sense — it is consumer-discretionary spending — but the impact is muted by the structure. Prime is a $25B+ subscription run-rate that does not flex with the cycle. Third-party seller revenue scales with units, not unit economics, so margin is more stable than first-party retail. The line that does cycle hard is advertiser demand — when CMOs cut budgets, the ad segment compresses fast, and that flows almost entirely to operating income because incremental ad revenue is near-pure margin. AWS is enterprise-IT cyclical (it lagged the broader downturn in 2022–23 as customers optimized spend, then re-accelerated), not GDP-cyclical.

The Metrics That Actually Matter

Standard ratios mislead on Amazon. Consolidated operating margin blends two incompatible businesses. Reported earnings include large non-cash gains on the Anthropic stake (about $15B in 2025). The metrics below are the ones I would put on a one-page dashboard.

Two ratios I would not anchor on: P/E (distorted by Anthropic mark-to-market and capex-driven D&A) and EV/Sales (mixes a 35%-margin business with a 7%-margin business). EV/EBITDA at the segment level is the cleaner cut, and a sum-of-parts at peer multiples — AWS at hyperscaler EV/EBITDA, ads at META's multiple, retail at WMT's — is the one that tracks the stock.

What I'd Tell a Young Analyst

Stop valuing Amazon as one business. The market still partly anchors on retail — that framing has been wrong for at least five years. Build the model as a sum of three: a hyperscaler, an ad network, and a global retailer. Then ask which of the three is most likely to surprise the consensus.

What to watch every quarter. AWS year-over-year growth rate (especially against Azure's), AWS backlog growth (forward demand), advertising revenue growth (the silent margin engine), and capex versus operating cash flow (the cycle indicator). If AWS growth re-accelerates while capex is still ramping, you are early in the cycle. If AWS growth fades while capex stays high, you are late.

What the market may be missing. The custom-silicon business (Trainium and Graviton) is now a $10B+ run-rate growing triple digits. It is not in any sell-side model as a separate line. If it works, AWS's gross margin structurally re-rates because Amazon stops paying NVIDIA's take. If it fails, the AI capex bill gets a lot harder to digest. This is the asymmetric bet inside the stock and it is largely unpriced.

What would change my mind. A sustained slowdown in AWS backlog growth — that signals demand has caught up with capacity, not that capacity is short. Retail operating margin compression in North America back below 5% — would mean Amazon is having to cut take-rates to defend share against Walmart and Shein/Temu. And any sign that ad growth is decelerating below 15% — the platform is supposed to be in early innings, not maturing.

The honest framing. Amazon at this price requires you to underwrite (1) AWS holding share against Azure as AI workloads concentrate at the hyperscaler tier, (2) the chips bet at least partly working, and (3) the retail business not having to give back margin to Walmart's e-commerce push. None of these is obviously true. All three are plausible. That is what 32× earnings is paying for.