Amazon.com, Inc. AMZN

Key data

The business

Amazon is three businesses stapled together. The online store, first-party goods it buys and resells plus a marketplace where outside sellers pay fees, wrapped in Prime, is the volume machine and runs on thin margins. Amazon Web Services rents servers, storage and databases by the hour and throws off the majority of company operating profit on a fraction of the revenue. Advertising, the fastest-growing and highest-margin piece, sells placement on those same shopping pages. The moat is really two: a logistics network of fulfillment centers and last-mile delivery no rival can cheaply copy, and AWS switching costs that keep enterprises paying once their systems live there. The customer feels the first moat as a package on the doorstep hours after ordering, and rarely thinks about the second at all, which is the point. What ties it together now is AI. AWS is racing to add data-center capacity for training and running models, and that race is reordering where every dollar of cash goes.

The numbers

The last five quarters show a company getting steadily more profitable on steadily larger sales.

Revenue climbed to $200.6B, up 20% year over year in Q2, and operating income outgrew it at 43% as AWS and advertising carried the mix. The Q2 net income of $62.6B is not what it looks like. It runs more than double operating income because of a large non-operating gain, so read the operating line, not the bottom one. The $5.75 diluted EPS blew past the $1.82 the Street modeled almost entirely on that gain.

Across the record the story is margin, not just growth. Revenue compounded about 11% a year from 2021 to 2025, but operating income more than tripled, from $24.9B to $80.0B, lifting operating margin from 5.3% to 11.2%, and to 13.4% in the first half of 2026. That is the retail cost base finally being spread over AWS and advertising dollars.

The cost of the AI race shows up here. Long-term debt jumped from $55.1B in Q3 2025 to $133.0B by Q2 2026, more than doubling in nine months, cash came off its peak, and capital spending, now about 22% of sales and running above the $161B operations threw off, pushed free cash flow negative. Amazon is funding the buildout with debt and its own cash flow at once. AWS revenue growth and operating margin are not split out in this quarter's pulled data, but the consolidated operating profit rose 43% year over year with margin at a company high.

On clean operating earnings of about $94B over the last twelve months, not the reported net income the non-operating gain inflates, the stock sits closer to 26 times than 21, still under its own history but not the outright bargain the headline shows. What this memo believes that the screen does not: the "cheapest in twelve years" tag is half real and half a one-time gain, and the print that settles it is AWS margin holding as capex peaks.

Management

No insider has bought a share in the open market in a year. Forty-six sales totaled $380.2M, and $346.5M of that was Jeff Bezos on August 3. Plan status is not disclosed in the filings pulled here, so whether that sale was pre-scheduled or discretionary cannot be confirmed, though Bezos has sold on a steady cadence for years to fund Blue Origin, the kind of selling that says little about the stock. Amazon pays its people mostly in restricted stock, about 2.5% of revenue, pays no dividend and buys back no stock. Every spare dollar goes back into the business, right now into data centers. Andy Jassy's operating-income results have run ahead of his own guidance for several quarters. The capital question is no longer whether Amazon reinvests, it is whether this reinvestment earns its cost.

How it fails or surprises you

The buildout outruns the demand. Amazon is funding $130B-plus of data-center capacity with new debt while free cash flow sits negative. If AWS demand cools before that capacity fills, returns on the spend fall and the debt stays. The first tell is AWS revenue growth slipping while its operating margin compresses in the same quarter.

The cheap multiple is half a mirage. Reported Q2 net income of $62.6B ran more than double operating income on a non-operating gain. Strip it and "21 times, cheapest in twelve years" becomes closer to 26, no longer a clear bargain. A down-mark of gains in a future quarter would reveal how much of the earnings was ever real.

AWS reaccelerates on AI (right tail). The market is paying for low-20s consolidated growth. If training and inference workloads push AWS growth back into the high 20s while margin holds near record, operating income compounds past 30% and the multiple re-rates on its own twelve-year history. The tell is two straight quarters of accelerating AWS revenue.

Closing thoughts

The market is paying for a low-20s grower that reinvests everything, and it is roughly right about that. The edge, if there is one, sits in AWS: whether the capacity now being poured in shows up as accelerating cloud revenue at a held margin, or as expensive concrete waiting for demand. That single split, AWS growth against AWS margin quarter by quarter, converts the whole position. An ambiguous print, growth up but margin down as new capacity drags, leaves you where you started, and then only the balance sheet's ability to carry $133B of debt matters. The right tail is fatter than the reported multiple suggests, because the market is not paying for an AWS inflection. The left tail is real and specific, a debt-funded overbuild into softening demand, with a flattered headline number hiding how little free cash the business makes today.

The bet is still that the store that ships you almost anything and the cloud arm that rents computing power to half the internet both keep growing while Amazon pours record money into AI data centers, and that the cloud arm earns enough to justify the concrete. What breaks it is that spending landing before the demand does. The one pair of numbers that tells you first, quarter by quarter and untouched by any non-operating mark, is AWS revenue growth and AWS operating margin. If both hold, the buildout is working. If growth fades while margin slips, it is not.