Amazon's Operating Leverage Story Is Only Halfway Through

Amazon operates three overlapping businesses that share infrastructure but generate very different economics. AWS is the cloud platform: infrastructure as a service, machine learning tooling, and database services. The advertising business monetises Amazon's purchase-intent data, placing it among the three largest digital ad platforms globally. And then there is everything else: the two-sided retail marketplace, Prime subscriptions, fulfillment and logistics, grocery, healthcare, and a device ecosystem.

The confusion about Amazon's value stems from how it reports. Revenue consolidation obscures the margin composition. A dollar of AWS revenue and a dollar of first-party retail revenue look identical on the top line. Below the line, they are not. AWS operates at margins above 35%. First-party retail barely breaks even. The blended 11.2% operating margin structurally understates the quality of the underlying profit engine.

In 2022, Amazon posted a net loss of $2.7 billion. The culprit was a Rivian investment write-down and the aftermath of pandemic-era overbuilding. Headcount surged during COVID and costs followed. The company spent 2023 and 2024 cutting, restructuring, and refocusing on where returns were actually being earned. The operational cleanup was thorough and fast.

The result is a business that added $55 billion in annual operating income between 2022 and 2025. That number deserves more attention than it currently receives.

Gross margin expansion is where the story becomes structural. Amazon's blended gross margin moved from 42% in 2021 to 50.3% in 2025. Nine percentage points on $717 billion in revenue is roughly $64 billion in additional gross profit generated through product mix shift alone. The driver is simple: AWS and advertising are growing faster than first-party retail.

Operating margin at 11.2% still has a long runway ahead. Microsoft's operating margin sits above 45%. Alphabet's is near 32%. Even accounting for Amazon's retail drag, the blended margin of a business that generates the majority of its operating income from software and services should be materially higher than 11%. The structural endpoint is somewhere in the low-to-mid 20s at current business mix.

The consistent pattern of EPS beats tells you the operating leverage is real and that analysts are structurally behind on their models. Amazon beat consensus EPS estimates by 22% in Q2 2024, 25% in Q3, 25% in Q4, 17% in Q1 2025, 26% in Q2 2025, and 27% in Q3 2025. Six consecutive double-digit beats is not randomness. It is margin expansion happening faster than the street models.

Q4 2025 EPS of $1.95 came in exactly at estimate, the first quarter without a meaningful upside surprise in six periods. One quarter of in-line results does not break the thesis, but it is worth watching whether the beat cadence resumes in Q1 2026 when results are due.

AWS is the highest-margin, fastest-growing segment and the primary reason the investment case exists. Cloud infrastructure demand has reaccelerated as enterprise AI deployments require dedicated compute capacity. AWS competes with Azure and Google Cloud for a market that continues to expand as workloads migrate from on-premise infrastructure.

The advertising segment is the segment that most surprises people who have not looked closely. Amazon's purchase-intent data is genuinely differentiated: search on Amazon starts with the intention to buy, which drives higher conversion rates and higher CPMs than general display advertising. The business has grown at double-digit rates for three consecutive years and operates with economics closer to search advertising than traditional display. Every incremental advertising dollar flows through at near-100% gross margin.

North America retail and International retail are the drag on the blended margin, but the direction of travel has changed. North America retail operating income turned solidly positive and continued expanding through 2024 and 2025. The combination of third-party seller growth, Prime membership, and logistics efficiency improvements is gradually closing the gap between retail and software-style economics.

The third-party marketplace take rate is the key metric to watch in retail. Sellers pay for fulfillment, advertising, and logistics at increasing attachment rates. As the take rate rises, the margin profile of the segment improves without requiring additional investment. This flywheel deepens the moat with every incremental seller and buyer who enters the ecosystem.

Revenue grew from $514 billion in 2022 to $717 billion in 2025, a 39% increase. Operating income over the same period went from $12.2 billion to $80 billion, a gain of more than 550%. Revenue grew by less than half. Operating profit grew by more than five times. That ratio defines the margin expansion thesis better than any single metric.

EBITDA hit $165.3 billion in 2025, up from $54.2 billion in 2022. The three-year EBITDA CAGR is above 45%. At a $2.375 trillion market cap, EV/EBITDA of 13.8x represents a significant discount to the software-style multiples these earnings might otherwise attract. The comparable figure for Microsoft is above 20x.

Free cash flow tells a more complicated story. FCF fell to $7.7 billion in 2025, down from $32.9 billion the prior year. The culprit is capex: $131.8 billion invested in 2025, up from $83 billion in 2024. Operating cash flow of $139.5 billion is excellent and continuing to grow. The gap between operating CF and FCF is entirely a function of deliberate investment decisions.

The balance sheet is clean. Cash of $86.8 billion exceeds total debt of $65.6 billion. The net cash position gives management flexibility to sustain the AI infrastructure program without financial stress or credit risk. Total equity of $411 billion reflects the cumulative reinvestment over decades.

AWS growth is reaccelerating. Andy Jassy's April 9 shareholder letter detailed AI and AWS expansion plans that drove the stock's biggest single-day gain in more than five months. The capital commitment backing that confidence is $131.8 billion in capex during 2025, almost entirely directed at data center buildout for AI workloads.

The AI infrastructure argument is straightforward: cloud capacity is a prerequisite for enterprise AI deployment, and Amazon is betting it needs to own that capacity before demand peaks. Data centers take 18 to 36 months to come online. The risk of underbuilding a hyperscale compute platform is arguably worse than the risk of modest overcapacity. Amazon has committed $25 billion in additional data center expansion as of April 9, signalling that the buildout is accelerating, not plateauing.

Amazon's Trainium and Inferentia custom chips are the underappreciated angle. Reducing dependence on Nvidia allows AWS to offer AI training and inference at lower cost to customers while improving segment margins. Jassy explicitly highlighted custom silicon as a strategic priority. If these chips reach competitive performance at scale, AWS margin improvement from AI infrastructure will be faster than consensus currently models.

Healthcare is the longer-dated option. Amazon's April 9 expansion into GLP-1 drug distribution through its partnership with Eli Lilly signals continued build-out of Amazon Pharmacy and healthcare services. This remains early-stage and not yet material to earnings. The optionality exists at a market undergoing structural disruption, and Amazon has repeatedly demonstrated the ability to enter adjacent markets and eventually dominate them.

The 70 analyst buy ratings versus 4 holds and zero sells reflects unusually strong consensus. Price targets cluster around $281, implying roughly 27% upside from current levels. This is not a contrarian thesis. It is the view that the market has not fully caught up to a margin transformation that is already in the financial statements.

The EV/EBITDA multiple of 13.8x is what creates the analytical opportunity. A business growing EBITDA at 45% annually with a clear structural driver of further margin expansion should trade at a premium to that multiple, not at a discount to software peers. The compression in this metric appears to reflect the FCF surface-level concern rather than a view about long-term earnings power.

FCF of $7.7 billion in 2025 looks modest relative to the $2.375 trillion market cap. But FCF yield calculated on the trailing number ignores the explicit investment component of the capex. If 2025 capex of $131.8 billion generates even $30-40 billion in incremental annualised revenue at AWS margins over the next three years, the implied FCF normalisation could be dramatic. The market appears to be pricing the pessimistic read on capex while underweighting the positive read on operating leverage.

Sentiment on Amazon has been elevated in recent sessions, with the 30-day normalised sentiment score averaging above 0.65. That positive tone strengthened sharply on April 9 following the CEO letter, reaching 0.77. Unusually, positive sentiment and strong fundamentals are aligned here. There is no divergence to arbitrage, which means the analytical work has to be done on the financial trajectory, not the market psychology.

AWS holds approximately 33% of the global cloud infrastructure market, ahead of Microsoft Azure at roughly 22% and Google Cloud at 12%. The lead is meaningful and has been maintained for over a decade. Azure's integration with Microsoft Office and Teams gives it an enterprise distribution advantage that Amazon lacks. Google Cloud has the AI research heritage and Gemini integration as a differentiating layer.

Amazon's response is capital and custom silicon. The $25 billion additional data center commitment announced April 9 maintains capacity leadership. Trainium and Inferentia chips reduce the NVIDIA dependency that represents both a cost and a supply chain risk for all three hyperscalers. If Amazon successfully scales custom silicon for AI training workloads at competitive performance, the cost advantage translates directly into AWS margin improvement and competitive pricing flexibility.

The advertising competitive position deserves separate consideration. Amazon competes directly with Google and Meta for digital ad budgets. The purchase-intent moat is structural: product search on Amazon starts the buying process rather than interrupting it, which drives measurably higher conversion rates. Advertiser CPMs on Amazon are higher than most display inventory precisely because the buyer signal is clearer. This advantage is durable as long as Amazon's marketplace remains the default starting point for product search, which it currently is for the majority of US consumers.