AMZN — Amazon.com, Inc.

Amazon operates three distinct compounding engines — AWS at $128 billion in annualized revenue and 35% operating margins, advertising at $68 billion in annualized revenue and growing 22% annually, and a retail and Prime membership flywheel serving 260 million subscribers worldwide — and the tariff-driven selloff of early 2026 has compressed the stock to its lowest normalized earnings multiple in years while leaving untouched the two businesses that actually drive the company's intrinsic value. The business is exceptional; at approximately $205 per share, it trades at 26 times normalized pre-tax earnings, modestly above the 15 times threshold where risk-adjusted return becomes compelling. Good business, meaningfully overpriced — compelling below $116.

---

The narrative that pushed Amazon shares from approximately $230 in late 2025 to under $200 in early 2026 centers on tariffs. The argument runs as follows: Amazon imports a significant portion of the goods it sells directly and the goods sold by its third-party marketplace sellers; tariffs increase the cost of those goods; higher costs reduce consumption or compress margins; and therefore Amazon's earnings power is impaired. The argument is not wrong in its mechanics — tariff friction in the retail business is real, and CEO Andy Jassy explicitly warned in February 2026 that tariffs could pressure margins through the year. But the argument applies primarily to a segment generating approximately $5 to $10 billion in annual operating income and is being conflated with the valuation of a company generating $80 billion in total operating income, the majority of which comes from businesses that have no tariff exposure at all.

Amazon's operating income in FY2025 was $79.975 billion. AWS contributed approximately $50 billion of that figure — 35% operating margin on $128 billion in revenue from cloud infrastructure contracts with enterprises that pay for compute, storage, and database services regardless of where physical goods are manufactured. The advertising segment contributed approximately $25 to $28 billion — margins approaching or exceeding 40% on advertising sold against the purchase intent of 260 million Prime members, none of which involves imported goods. The retail North America and International segments, together representing $520 billion in revenue, contributed the remaining $5 to $8 billion in operating income at margins below 2%. It is the low-margin segment that faces tariff pressure. The high-margin segments that determine the majority of intrinsic value are unaffected. The stock is priced as though the earnings impairment is across all three businesses simultaneously.

The cloud infrastructure market is the most consequential economic infrastructure being built in the current era. Corporations, governments, healthcare systems, and research institutions are migrating their computing workloads from owned physical hardware to rented cloud infrastructure at a pace that has been sustained for fifteen years and is accelerating rather than decelerating. The market exceeded $800 billion in 2026 and is on a trajectory to $1.9 trillion by 2030 at an 18.7% compound annual growth rate. The structural driver is not fashion but economics: cloud computing eliminates capital expenditure on server hardware, allows capacity to scale elastically with demand, and provides access to specialized infrastructure — GPU clusters for AI, purpose-built database engines, global content delivery networks — that no individual enterprise can efficiently maintain independently. The customer who has moved their applications to cloud infrastructure does not move them back; the migration cost was paid to get in, and the institutional disruption of leaving is larger than the disruption of staying. This makes the cloud infrastructure business a subscription business with switching costs, even though it is priced by consumption rather than seat.

The digital advertising market is structurally distinct. Advertising is a market in which the highest-valued inventory is the inventory closest to a purchase decision. A pharmaceutical advertisement appearing during a television broadcast reaches a general audience; a pharmaceutical advertisement appearing in a search result for a specific drug reaches a person actively researching that drug. Amazon's advertising inventory is unique: it appears on a platform used by 260 million paying subscribers specifically to purchase goods, and it is measurable at the transaction level in a way that television, display, and most search advertising is not. When an advertiser bids on an Amazon keyword, they know exactly how many products sold as a result. This closed-loop attribution — intent to purchase, served advertisement, completed transaction — is the most valuable targeting environment in digital advertising, and it has allowed Amazon's advertising revenue to grow from $9.5 billion in FY2019 to $68 billion in FY2025 without any meaningful degradation in advertising effectiveness or pricing power.

Amazon's three-segment architecture was not planned as such; it evolved from the interaction of three distinct strategic decisions. The first was the 2006 decision to externalize Amazon's internal computing infrastructure as a public cloud, which became AWS. The second was the recognition that Prime membership — built initially as a free shipping service — had created a captive audience whose purchasing data was the most valuable first-party targeting database in commercial advertising. The third was the gradual shift from direct retail (Amazon as seller) to a marketplace model (Amazon as platform) in which third-party sellers pay fulfillment fees, advertising fees, and subscription fees in exchange for access to that customer base. Each decision compounded the others: Prime members buy more, which makes advertising more valuable, which attracts more sellers, which improves Prime selection, which attracts more Prime members. The flywheel is not a metaphor; it is a description of how the units of each business are the inputs to the others.

The moat at Amazon is best understood by examining what an enterprise would need to do to stop using AWS, what a brand would need to do to replicate Amazon's advertising targeting, and what a consumer would need to do to replace Prime. For AWS, the answer is migration: moving existing workloads off AWS requires re-architecting applications built on AWS-specific services, re-establishing DevOps tooling configured for AWS APIs, retraining engineering teams, and re-negotiating contractual commitments. S3, AWS's object storage product, has become the de facto standard for cloud-native data storage; the $0.09 per gigabyte data egress fees that apply to moving data out of S3 have created a concept called "data gravity" — data stored in S3 tends to stay in S3 because the cost of moving it exceeds the value of doing so at any scale. AWS offers over 250 distinct services that interconnect within its infrastructure; migrating to a competitor means not just moving the primary application but rebuilding the integrations to every ancillary service. For advertising, no other platform can offer purchase-intent targeting at Amazon's scale with closed-loop transaction attribution. Google's Shopping ads are the closest analog, but Amazon's transaction data — what a specific customer has purchased, returned, searched for, and browsed over years of Prime membership — is not reproducible by a search company. For Prime, replacing it means replacing one-day delivery, Prime Video (315 million average viewers), Prime Music, and the cumulative convenience of an account that knows your purchase history: that substitution requires adopting multiple competing services simultaneously.

	AWS	Microsoft Azure	Google Cloud
Market share (2026)	~32%	~23%	~11%
Q4 2025 revenue ($B)	$35.6	~$20–22	~$11.3
YoY growth (Q4 2025)	+24%	~+35%	~+28%
Operating margin	~35%	Not disclosed (buried in segment)	~10–15%
Forward backlog	$244B (+40% YoY)	Not disclosed	Not disclosed

The competitive table contains the most important fact about AWS's position: the $244 billion backlog growing 40% year over year. A backlog is contracted revenue — enterprises that have signed agreements to spend specific amounts on AWS over coming periods. At $244 billion, the AWS backlog represents nearly two years of current revenue already committed. The 40% year-over-year growth in backlog, occurring in the same period that Azure was growing at 35%, indicates that enterprises are committing to AWS infrastructure spending at an accelerating rate even as competitive pressure has increased. Azure's superior growth rate (35% vs. 24%) deserves honest acknowledgment: Microsoft has leveraged its deep integration with enterprise Office 365, Teams, and Dynamics deployments to create a credible alternative to AWS for workloads with natural Microsoft connectivity. Azure has gained market share points over several years. But AWS's absolute revenue growth — adding approximately $27 billion in annual revenue in FY2025 on a $101 billion base — exceeds Azure's absolute additions on a smaller base. The moat has not been breached; it has been partially encircled, which is a different and manageable condition.

FY2025 delivered total revenue of $716.9 billion, growing 12.4% year over year, with operating income of $79.975 billion — up 16.6% from $68.6 billion in FY2024 — and GAAP net income of $77.67 billion, equivalent to $7.29 per diluted share. Operating cash flow was $139.5 billion, up 20% year over year. Free cash flow, however, was approximately $7.7 billion — because Amazon spent $131.8 billion in capital expenditure in FY2025, up from $83 billion in FY2024, and has guided $200 billion in capex for FY2026. This divergence between GAAP operating income and free cash flow is the central financial fact about Amazon in 2026: the company is investing at a historically unprecedented rate in data center capacity, GPU infrastructure, and AI computing capability, which depresses near-term FCF while building the physical infrastructure that will support AWS revenue for the next decade.

The capex surge deserves analysis rather than fear. Amazon spent $53 billion in capex in FY2023 and generated $32 billion in FCF. The incremental $147 billion between FY2023 and FY2026E guidance is being deployed in data centers and AI infrastructure. Jassy stated in February 2026 that he is "confident" the investment will see a "strong return on invested capital," though the timing of return is not specified. The AWS backlog growing 40% — contracted enterprise commitments to pay for the infrastructure being built — provides evidence that the demand exists to absorb the capacity. The prior capex cycle (FY2019 to FY2021), in which Amazon invested heavily in logistics and fulfillment, ultimately produced the one-day delivery network serving 185 million U.S. Prime members and making 8 billion same-day or next-day deliveries in FY2025. The capex-to-return cycle in infrastructure has historically validated itself at Amazon.

GAAP and adjusted figures at Amazon diverge primarily because of stock-based compensation, which was approximately $21 to $24 billion in FY2025 — approximately 3.3% of revenue. This SBC is already deducted in arriving at the GAAP operating income of $79.975 billion; there is no adjusted figure that needs to be unwound to arrive at an honest earnings number. The operating income figure is GAAP-inclusive of SBC and is therefore the right baseline for valuation. The net income of $77.67 billion against operating income of $79.975 billion reflects modest net non-operating income after interest income, interest expense, and investment gains and losses — some of which (investment gains on Anthropic, Rivian equity positions) are non-recurring. The normalized pre-tax earnings base is approximately $82 billion ($79.975 billion operating income plus approximately $5.5 billion in interest income, less $3.5 billion in interest expense, excluding non-recurring investment gains).

Andy Jassy took the CEO position from Jeff Bezos in July 2021 and inherited a business losing money at the operating level — FY2022 operating income was $12.2 billion, down from $24.9 billion in FY2021, as pandemic-era over-investment in logistics capacity overwhelmed revenue growth. The transformation from $12.2 billion in operating income in FY2022 to $79.975 billion in FY2025 represents the most consequential operational turnaround in Jassy's tenure: he shed excess logistics capacity, reduced headcount by tens of thousands of roles, restructured the retail operating model around profitability rather than market share, and simultaneously allowed AWS to accelerate its AI infrastructure buildout. The operating margin expanded from 1.7% in FY2022 to 11.2% in FY2025 across a $717 billion revenue base, which is not a small-company agility story — it is a large-company execution story validated by numbers that are difficult to argue with. Jassy owns approximately $100 million in Amazon equity and has received stock awards tied to performance rather than tenure, creating meaningful alignment with long-term shareholders.

The growth runway is most legibly captured in the trajectory of the two businesses that determine Amazon's long-term intrinsic value.

Year	AWS Revenue ($B)	AWS YoY Growth	Advertising Revenue ($B)	Ad YoY Growth	Total Oper. Income ($B)
FY2021	$62.2	+37%	$31.2	+58%	$24.9
FY2022	$80.1	+29%	$37.7	+21%	$12.2
FY2023	$90.8	+13%	$46.9	+24%	$36.9
FY2024	$107.6	+19%	$56.2	+20%	$68.6
FY2025	~$128	+19%	~$68	+21%	$80.0
FY2026E	~$155	~+21%	~$80+	~+18%	~$88–95

The FY2022 total operating income collapse — to $12.2 billion from $24.9 billion — shows what happens when Amazon's retail segment overinvests in capacity that revenue doesn't fill. AWS grew 29% that year, advertising grew 21%, and the company still lost money on an operating basis because of the retail restructuring. This is the right way to read the table: the year that tested whether the thesis was real, AWS and advertising both sustained strong growth while total operating income fell off a cliff due to issues entirely outside the high-margin businesses. The subsequent recovery to $36.9 billion in FY2023 and $80 billion in FY2025 is the retail restructuring resolving, not a reversal of the underlying drivers. The two high-margin engines have grown continuously and consistently regardless of what the retail segment was doing.

AWS has grown from $62 billion to approximately $128 billion in four years while accelerating — the 24% Q4 2025 growth rate is higher than the 13% trough in FY2023 and approaching the 29% growth of FY2022 on a base that has doubled. The acceleration is driven by enterprise AI workloads: training large language models requires massive GPU clusters that only cloud providers can deliver at scale, and inferencing those models in production applications requires persistent, low-latency cloud infrastructure. Jassy stated in March 2026 that he expects AWS to reach at least $600 billion in annual revenue by 2036 — more than four times the current annualized run rate — driven by AI workload adoption that is still in its earliest commercial deployment phase. The $244 billion backlog growing 40% year over year provides the most credible forward evidence that this trajectory is supported by actual enterprise commitments rather than speculative projection.

Amazon's advertising segment has captured approximately 7 to 8 percent of the global digital advertising market, estimated at $800 billion to $1 trillion by 2026. The 260 million Prime members worldwide — of whom 185 million are in the United States — represent approximately 55% of the U.S. household population. Each of these members is a purchaser with a transaction history, not a browser with a search query. Advertising inventory served against this audience generates measurably higher return on ad spend than most alternatives, creating pricing power that has allowed Amazon advertising revenue to grow at 20 to 22 percent annually even as the digital advertising market has faced uncertainty from signal loss, privacy regulation, and AI creative disruption. The international advertising opportunity — Prime has recently introduced ads-supported video tiers in 16 countries and has 315 million average viewers on Prime Video — represents a significant additional runway in markets where Amazon's advertising infrastructure is a fraction of its U.S. scale.

At $205 per share, Amazon trades at a market capitalization of approximately $2.18 trillion. With net debt of approximately negative $50 billion (excess of cash and equivalents over long-term debt), the enterprise value is approximately $2.13 trillion. On GAAP operating income of $79.975 billion, the EV/operating income multiple is approximately 26.7 times. On normalized pre-tax earnings of approximately $82 billion (operating income plus structural interest income, excluding non-recurring investment gains) and diluted shares of 10.65 billion, normalized pre-tax EPS is approximately $7.70. At $205, the multiple is 26.6 times. The 15 times threshold implies a compelling entry price of approximately $116.

What makes the current multiple uncomfortable despite Amazon's exceptional business quality is the combination of two simultaneous factors that reduce near-term visibility. First, the $200 billion capex guidance for FY2026 — up 52% from FY2025's already elevated $131.8 billion — creates genuine uncertainty about FCF for two to three years. Not because the investment is wrong (the AWS backlog suggests the demand exists to absorb it) but because the magnitude is unusual even by Amazon's standards and the ROI timeline is unspecified. Second, the tariff risk to retail margins — real but concentrated — creates headline earnings volatility that is difficult to quantify until import tariff rates stabilize. The two uncertainties compound each other: investors cannot precisely model FY2026 earnings because the capex decision affects depreciation, the tariff exposure affects gross margins, and both effects are still developing. A business whose earnings are temporarily imprecise is riskier to buy at 26 times than one whose earnings are predictable to within a reasonable range.

The intelligent bear argues that Azure's superior growth rate — growing at 35% versus AWS's 24%, for nine consecutive quarters above 30% — signals a structural shift in enterprise cloud preference driven by Microsoft's integrated enterprise software stack, and that AWS's $244 billion backlog is a trailing indicator of commitments made before enterprises recognized the AI stack would favor Azure's integration with Copilot, M365, and Teams. This is the most credible version of the bear case. The counter is that AWS's absolute revenue additions in FY2025 exceeded Azure's — AWS grew approximately $20 billion in revenue versus Azure's estimated $15 to $18 billion — and the backlog's 40% year-over-year growth in the same period Microsoft was most aggressively marketing Copilot suggests the two are winning different segments of enterprise AI demand rather than competing for the same workloads. Azure wins where existing Microsoft software integrations determine infrastructure choice; AWS wins where teams are selecting infrastructure on technical merit alone.

The thesis changes in two scenarios. If normalized pre-tax EPS reaches $10 within two years — achievable if operating income grows 15% annually from $80 billion — the stock at $205 drops to approximately 20.5 times forward pre-tax earnings, approaching but still above the threshold. For the conviction record to flip from watchlist to portfolio, either the stock needs to fall to approximately $116 or operating income needs to grow to approximately $136 billion, implying a period of time and execution risk that remains real. The tariff uncertainty resolves the question faster in either direction: if tariffs are contained or reversed, retail margins stabilize and operating income grows faster; if tariffs escalate and cause demand destruction in consumer electronics and third-party marketplace goods, the retail drag on total operating income grows.

Amazon at $205 is approximately 23 percent above where it becomes compelling on the most rigorous normalized earnings standard. That is not an extraordinary premium to wait out — operating income growing 15 percent annually closes the gap in roughly 18 months — but it is a premium nonetheless, and the capex surge and tariff uncertainty mean the next 12 to 18 months carry more earnings volatility than usual. The business itself — three compounding engines sharing data, infrastructure, and a flywheel relationship that has not weakened in 15 years of competitive attack — is not in question. The price is.