§ 01 Business Overview
Amazon is the hardest company in the public markets to value because it is not one business. It is five, stacked inside each other like nesting dolls, each one changing the economics of the one beneath it.
The first business is e-commerce. Amazon sells goods directly to consumers and operates the largest third-party marketplace in the United States. North America retail generated $387 billion in revenue in 2025. International retail generated $143 billion. Together they represent roughly 74% of total revenue. They are also the lowest-margin part of the business, historically operating near breakeven and only recently turning consistently profitable.
The second business is cloud infrastructure. Amazon Web Services generated $108 billion in revenue in 2025, up 19% year over year, at a 35.4% operating margin. AWS is the dominant cloud provider globally with approximately 31% market share, ahead of Microsoft Azure at 24% and Google Cloud at 12%. It is also the engine that funds everything else. AWS contributes roughly 57% of Amazon's total operating income despite representing only 18% of total revenue.
The third business is advertising. Amazon's advertising segment generated approximately $69 billion in 2025, growing at over 20% annually. It is, by margin profile, closer to Meta's advertising business than to retail. Every sponsored product listing on Amazon.com is a high-margin advertising placement sold against purchase intent data that no other platform can replicate. A user searching "running shoes" on Amazon is further down the purchase funnel than anyone searching the same phrase on Google or Meta.
The fourth business is custom silicon. Trainium, Graviton, Inferentia, and Nitro are Amazon-designed chips running inside AWS data centers. The custom chip business now generates more than $20 billion in annualized revenue growing at triple-digit percentages year over year. Andy Jassy said on the Q1 2026 earnings call that if the chips business were a standalone company selling chips produced this year to AWS and third parties as other chip companies do, its annual revenue run rate would be $50 billion. That is not a projection. It is a statement about what Trainium and Graviton are already producing, valued at market pricing.
The fifth business is logistics and physical infrastructure. Amazon operates the largest private delivery network in the United States, with over 1 billion same-day or overnight deliveries completed so far in 2026. The grocery business surpassed $150 billion in gross sales in 2025, making it the second-largest U.S. grocer by that measure, with same-day perishable delivery growing over 40 times year over year.
In Q1 2026, Amazon reported $181.5 billion in revenue, up 17% year over year, beating the $177.3 billion Wall Street consensus by over $4 billion. Operating income reached $23.9 billion, a 13.1% margin that CFO Brian Olsavsky called the highest in Amazon's history. Free cash flow collapsed 95% to $1.2 billion as capital expenditure hit $44.2 billion, up 77% year over year, with a full-year 2026 capex plan of approximately $200 billion. That last number is the entire analytical debate about Amazon right now.
§ 02 Competitive Moat · Strong
Amazon has four separate moats, and unlike most conglomerates where the businesses compete for capital without reinforcing each other, Amazon's moats compound. Each business makes the others stronger.
AWS: the data gravity moat. Cloud computing has a structural lock-in property that goes beyond switching costs. When an enterprise moves its databases, analytics, and applications to AWS, it creates what the industry calls data gravity: the tendency of compute and applications to move toward where the data lives rather than the other way around. Moving data at petabyte scale is slow, expensive, and disruptive. Once a company's core data is in AWS, the path of least resistance for every subsequent application is to build on AWS as well. This compounds over time as the data grows and the application stack deepens. AWS's 28% growth in Q1 2026 on a base of roughly $150 billion annually is not a company with a weakening competitive position. It is a business where the lock-in is strengthening as more enterprise workloads migrate permanently to the cloud.
The AWS backlog reached $364 billion at the end of Q1 2026, a figure that excludes the recently announced Anthropic deal worth over $100 billion. Management emphasized the backlog has reasonable breadth and is not concentrated in only one or two large customers. A $464 billion contracted backlog against a $150 billion annual run rate means more than three years of current revenue is already committed. That is not demand speculation. It is signed contract.
Advertising: the purchase intent moat. Amazon's advertising moat comes from a data source that Google, Meta, and every other advertising platform cannot replicate: actual purchase behavior at scale. When a brand advertises on Amazon, it is reaching consumers who are actively shopping for something. The conversion rate on Amazon advertising is structurally higher than display or social advertising because the intent signal is stronger. This is why Amazon's advertising business generates margins closer to pure software than retail, and why it can sustain 20%+ annual growth while already running at $69 billion in annual revenue. The data flywheel reinforces itself: more purchases create more behavioral data, which improves targeting, which delivers better returns for advertisers, which attracts more advertising spend.
Custom silicon: the vertical integration moat. This is the newest layer and the one most investors are still underweighting. Trainium2 delivers approximately 30% better price performance than comparable GPUs. Trainium3, which began shipping in early 2026, improves on that by another 30 to 40%. Trainium commitments now exceed $225 billion in total revenue commitments. Trainium2 is largely sold out. Trainium3 is nearly fully subscribed. Much of Trainium4, still roughly 18 months from broad availability, is already reserved. The mechanism behind the moat is operating leverage: every dollar of Trainium that displaces Nvidia GPU purchases inside AWS saves Amazon approximately the same margin that Nvidia would have earned on that chip. Jassy said Trainium will save Amazon tens of billions in annual capex and provide several hundred basis points of operating margin advantage versus relying on others' chips for inference. That margin advantage widens the cost gap between AWS and any cloud competitor that still depends on Nvidia for the bulk of its AI compute.
Retail and logistics: the scale moat. Amazon's fulfillment network is the largest private delivery infrastructure in the United States. The unit economics of this network improve with every incremental package because the fixed cost of the network - facilities, vehicles, and routing algorithms - is spread across a larger volume base. Third-party sellers on the marketplace generate high-margin revenue for Amazon while simultaneously filling the logistics network, which reduces per-unit delivery cost for first-party goods. This is a flywheel that has been running for 25 years and is still accelerating.
The financial snapshot connects directly to the moat. AWS's 35.4% operating margin in 2025, recovering from 27.1% in 2023, is the result of operating leverage as the fixed infrastructure base fills with contracted workloads. Each additional dollar of AWS revenue beyond the infrastructure cost flows to operating income at a much higher margin than the segment average. That is the same mechanism Amazon's custom silicon strategy is designed to amplify further.
§ 03 Financial Snapshot
The three-year direction of travel is unambiguous on every line. Revenue grew from $514 billion to $716.9 billion, a 39% increase across three years. Operating income grew from $12.2 billion to $80 billion, a 555% increase across the same three years. Net income went from a $2.7 billion loss in 2022 to $77.7 billion in profit in 2025. That margin expansion from 2.4% to 11.2% is driven by a single mechanism: AWS and advertising, the two highest-margin segments, growing faster than retail, the lowest-margin segment. As the revenue mix shifted toward high-margin businesses, consolidated operating margin expanded even without dramatic improvement in any single segment.
| Year | Total Revenue | AWS Revenue | Operating Income | Op. Margin | Net Income |
|---|---|---|---|---|---|
| 2022 | $514.0B | $80.1B | $12.2B | 2.4% | ($2.7B) |
| 2023 | $574.8B | $90.8B | $36.9B | 6.4% | $30.4B |
| 2024 | $638.0B | $107.6B | $68.6B | 10.8% | $59.2B |
| 2025 | $716.9B | $108.0B+ | $80.0B | 11.2% | $77.7B |
AWS segment operating margin moved from 27.1% in 2023 to 37% in 2024 before moderating to 35.4% in 2025 as the capex build began hitting depreciation ahead of the associated revenue. In Q1 2026, AWS operating margin was 37.7%, the highest since the capex surge began, because Trainium displacement of external chip purchases started flowing through the income statement. The direction of AWS margin in 2026 is upward.
Q1 2026 results:
| Metric | Q1 2026 | Change YoY |
|---|---|---|
| Total Revenue | $181.5B | +17% |
| AWS Revenue | $37.6B | +28% (fastest in 15 quarters) |
| AWS Operating Income | $14.2B | margin 37.7% |
| North America Operating Income | $8.3B | margin 7.9% |
| International Operating Income | $1.4B | margin 3.6% |
| Advertising Revenue | $17.2B | +22% |
| Total Operating Income | $23.9B | margin 13.1% (highest ever) |
| Net Income | $30.3B | incl. $16.8B Anthropic gain |
| Capital Expenditure | $44.2B | +77% |
| Free Cash Flow | $1.2B | -95% |
| Q2 Revenue Guidance | $194-$199B | - |
| Q2 Operating Income Guidance | $20-$24B | - |
The Anthropic accounting issue. The Q1 net income of $30.3 billion includes a $16.8 billion pre-tax gain from Amazon's mark-to-market revaluation of its Anthropic investment. Under current accounting rules, Amazon must mark its Anthropic investment to fair value whenever a significant financing event provides a new reference price. Anthropic's shares have been trading at an implied $1 trillion valuation on secondary markets, and the company is reportedly in early discussions for an IPO that could come as soon as October 2026. Each upward revaluation flows directly into net income without any corresponding cash inflow. The true operating beat in Q1 was real and significant - operating income cleared the top of management's own guidance by $2.4 billion - but the net income figure requires adjustment to understand what the core business actually earned.
Capex and free cash flow: 2024 capex was approximately $83 billion. 2025 came in at approximately $105 billion. 2026 full-year guidance is approximately $200 billion. Against trailing twelve-month free cash flow of $1.2 billion, this is a year where operating cash generation is substantially below capital deployment. The AWS backlog of $464 billion and Trainium revenue commitments of $225 billion are the contracted demand base that makes this deployment rational.
Valuation: At approximately $207 to $215, Amazon trades at 32 to 35x forward earnings and approximately 18x EV/EBITDA. EBITDA margin expanded from 14.6% in 2022 to 23.7% in 2025, with consensus projecting approximately 35% by 2030. Analyst consensus sits at 46 Buys, 15 Outperforms, 5 Holds, and zero Sells, with a mean 12-month price target of approximately $285 - implying 33 to 38% upside from current prices.
§ 04 Risk Rating
Free cash flow is negative for the duration. Amazon spent $44.2 billion in a single quarter on infrastructure. The full year 2026 plan is $200 billion. Free cash flow fell 95% to $1.2 billion in Q1. Management has guided that free cash flow will remain constrained through 2026 and into 2027 as the infrastructure build lags monetization by 6 to 24 months. The thesis requires believing that $200 billion in annual capex will generate returns that exceed the cost of that capital on a timeline measured in years, not quarters. The AWS backlog of $464 billion provides more revenue visibility than most companies ever achieve. But backlog converts to cash over time, and in the interim Amazon is deploying more capital than its operations are generating.
The Anthropic accounting distortion creates a visibility problem. When net income includes $16.8 billion in non-cash investment gains in a single quarter, the headline earnings figure becomes unreliable as a measure of operating performance. Every analyst adjusting for this arrives at a different number. The distortion will grow if Anthropic moves toward an IPO and the mark-to-market adjustments become larger and more frequent. Investors relying on GAAP net income to evaluate Amazon are looking at the wrong number, and the market does not always make that distinction cleanly in real time.
Cloud growth rate comparison creates a narrative risk. AWS grew 28% in Q1 2026. Google Cloud grew 63%. Azure grew approximately 40%. In absolute dollar terms, AWS added more revenue in a single quarter than Google Cloud generates in a quarter, because it is starting from a much larger base. But the percentage comparison is unfavorable and will continue to be unfavorable as long as Google Cloud and Azure grow faster on a percentage basis. That comparison creates a recurring narrative that AWS is losing the cloud AI race, even when the absolute growth numbers tell a different story.
The Amazon LEO satellite bet is a capital call with uncertain returns. Amazon acquired Globalstar to secure global spectrum for direct-to-device capabilities and announced a deep relationship with Apple, which plans to use these services for iPhones and Watches. The satellite constellation requires significant manufacturing and launch infrastructure investment alongside the AWS build. LEO manufacturing costs are expected to weigh on Q2 2026 operating income by approximately $1 billion, and the commercial service has not yet launched. This is a third major capital bet running simultaneously alongside AWS infrastructure and custom silicon, and it is the least proven of the three.
The risk rating is 5 because the business is executing at record levels on the metrics that matter, the AWS backlog provides revenue visibility that most companies cannot achieve, and the custom silicon strategy creates a margin improvement path that is now showing up in the actual numbers. The free cash flow compression is real but it is a timing issue created by deliberate capital allocation against contracted demand, not deteriorating unit economics.
§ 05 Bull vs. Bear
Bull case: Amazon is building infrastructure that compounds in value the longer it runs. The AWS data gravity moat deepens with every workload migration because moving data out of AWS becomes more expensive and disruptive as the data grows. The advertising moat deepens with every purchase because behavioral data at transaction level improves targeting in ways that click and scroll data cannot. The custom silicon moat deepens with every Trainium generation because the price-performance gap versus external chips widens while Amazon's cost base stays fixed.
The $200 billion capex build is not speculation. It is being deployed against $464 billion in contracted AWS backlog plus $225 billion in Trainium commitments. The return on that capital is contractually guaranteed to a significant degree before the infrastructure is even built. When free cash flow recovers as the depreciation cycle normalizes and the contracted revenue scales, Amazon will generate more free cash flow than at any point in its history. EBITDA margins moving from 23.7% today toward 35% by 2030 on a revenue base approaching $1 trillion implies free cash flow at a scale that makes the current market cap look conservative.
The custom silicon business is the most underappreciated asset on Amazon's balance sheet. At $50 billion in implied annual revenue growing at triple-digit rates, Trainium and Graviton are already competing with AMD in absolute revenue terms and approaching Nvidia in strategic importance. Meta signed a multibillion-dollar deal for Graviton5 processors. Uber joined the Trainium customer roster in Q1. The external chip business is beginning to scale.
Bear case: The free cash flow collapse is not a timing issue at the margin. Amazon committed $200 billion in 2026 capex and has already signaled 2027 will be higher. The gap between operating cash flow and capital expenditure is not narrowing on a quarterly basis. It is widening. Every quarter that free cash flow stays near zero while the stock trades at 32x forward earnings requires the market to extend significant faith in a return timeline that management has been consistently vague about in terms of specific quarters and specific returns.
Google Cloud's 63% growth and approximately $462 billion backlog reflect AI-native workload capture. Azure's 40% growth reflects enterprise AI adoption via Microsoft's distribution. AWS's 28% growth on a $150 billion base reflects scale-driven execution. The concern is not that AWS is losing the cloud market. It is that Google Cloud is winning the AI-native workload category disproportionately because Google built Gemini, TPUs, and AI Overviews into an integrated product that sells AI infrastructure alongside AI models. Amazon's AI strategy relies on being the platform where Anthropic, OpenAI, and other model builders deploy, which means Amazon benefits from AI demand without owning the AI model relationship. That is a viable strategy but it is a different moat than owning the model.
The Anthropic accounting distortion is a genuine analytical risk. Half of Amazon's Q1 net income came from a non-cash investment gain. If Anthropic's valuation declines, those gains reverse and flow through net income as losses with no cash impact in either direction. The investment stake creates significant earnings volatility that has nothing to do with Amazon's operating performance.
Buy on Weakness. Entry interest: $185 to $200. At current prices around $207 to $215, Amazon is fairly valued for a business executing at this level. Thirty-two times forward earnings on 17% revenue growth with the highest operating margin in the company's history and a $464 billion AWS backlog is not an aggressive multiple. It is a reasonable one for a business with this many compounding growth vectors.
The reason the entry range sits below current prices is the free cash flow situation. Buying at 32x forward earnings in a year when free cash flow has collapsed 95% requires trust that the capex converts to revenue on the timeline management projects. That trust is warranted given the backlog, but the margin of safety is thin at current prices. At $185 to $200, the forward multiple compresses to approximately 28 to 30x on a business whose operating income is growing 30% year over year, AWS is accelerating, advertising is expanding, and custom silicon is creating a margin improvement path that does not yet exist in consensus models.
At $185 to $200, the capex risk is more adequately compensated, and the upside from free cash flow normalization in 2027 and 2028 is significant.
§ 06 What to Watch
AWS operating margin each quarter. The Q1 2026 figure of 37.7% was the highest since the current capex build began, because Trainium displacement of external chips started flowing through the cost structure. If that margin holds or expands at 28 to 30%+ revenue growth in Q2 2026, it confirms that Jassy's claim of several hundred basis points of margin advantage from Trainium is materializing ahead of schedule. Any quarter where AWS revenue grows but AWS margin compresses signals the capex is outpacing monetization.
Trainium external revenue scaling. Management is considering selling Trainium in rack form to third parties within the next couple of years. The first external customer wins beyond Meta and Uber will be the earliest signal that Amazon is transitioning from a captive chip producer to a chip vendor with external pricing power. That transition, if it happens, is worth several turns of multiple expansion because it creates a revenue stream with Nvidia-like margins rather than cloud infrastructure margins.
Free cash flow trajectory quarter by quarter. The market is pricing the eventual recovery. Any quarter where free cash flow turns materially positive ahead of consensus expectations would be a significant catalyst, because it would validate that the capex cycle is converting to revenue faster than the model assumes.
Anthropic IPO timeline and valuation. Anthropic is reportedly in early discussions for an IPO that could come as soon as October 2026. An Anthropic IPO would crystallize the value of Amazon's investment stake from a mark-to-market accounting entry into a tradeable asset, and depending on the IPO valuation, could generate a significant cash event for Amazon through secondary sales. It would also eliminate the non-cash accounting distortion that currently makes Amazon's net income unreliable as an operating metric.
Advertising growth rate sustainability. At $69 billion in annual run rate growing at 22%, Amazon advertising is approaching the scale where the law of large numbers begins to constrain percentage growth. Watch for any deceleration below 18% in the quarterly advertising growth rate as a signal that the purchase intent moat is saturating in certain categories.
This analysis introduced the concept of vertical integration as a margin lever, and why it is more powerful at Amazon's scale than it appears from the outside.
Most businesses buy inputs from suppliers and sell outputs to customers. The margin they earn is the spread between what they pay for inputs and what customers pay for outputs. Vertical integration means owning more of the supply chain, which has two effects. It reduces input costs by eliminating supplier margins. And it creates differentiation that competitors without the same vertical integration cannot easily replicate.
Amazon's custom silicon strategy is vertical integration in its most financially sophisticated form. Instead of buying Nvidia GPUs to run AWS, Amazon builds its own chips. The margin Nvidia would have earned on those GPUs stays inside Amazon. At the scale of $200 billion in annual capex, even a 20% reduction in chip costs from using proprietary silicon instead of purchased GPUs represents tens of billions in annual savings. That is the mechanism behind Jassy's claim of several hundred basis points of operating margin advantage.
The reason this matters for understanding the broader MCI portfolio is that the same principle explains why Nvidia's moat is simultaneously powerful and fragile. Every hyperscaler that successfully builds custom silicon is a customer that partially exits the Nvidia ecosystem. Google built TPUs. Amazon built Trainium. Microsoft built Maia. Meta built MTIA. Each of those programs, at scale, reduces Nvidia's addressable market from inference workloads at that customer. Nvidia's moat is strongest in training, where CUDA platform dependency is deepest, and it is eroding at the inference margin as hyperscalers vertically integrate.
The Amazon analysis and the Nvidia analysis from Week 09 are actually two sides of the same story: one company building the infrastructure to reduce its dependency on the other. Understanding both analyses together gives a more complete picture of how the AI infrastructure economy is actually evolving than either analysis gives alone.