The Bookstore Quietly Became an Ad Company and a Cloud Company. You Still Think It Sells Books.

Search for a pair of running shoes on Amazon and the first four results have a small grey word next to them: Sponsored. You probably don't see it anymore. A brand paid for that placement, and the fee went straight to Amazon — no inventory bought, no box shipped, no return to process. Do that across enough searches and the grey word adds up to $56.2 billion a year in advertising revenue.⁴ That money is not in the cloud, and it is not in the warehouse. It is hiding in plain sight, on the storefront everyone thinks they understand.

The official story is that AWS, the cloud business, subsidizes a loss-making retail operation — and that story is roughly a decade out of date. By 2024 the store was not losing money at all. Amazon stopped being a profit engine bolted to a charity, and quietly became something stranger: a machine with three engines, two of which sit inside the same building.

The subsidy story everyone learned in 2015

For years the headline number told a clean and irresistible tale. AWS, on just 17% of Amazon's revenue, threw off the lion's share of the profit — $39.8 billion of operating income in 2024, against the company's $68.6 billion total.¹³ That is roughly 58 cents of every profit dollar coming from a business that sells server time, not stuff. The natural conclusion: the store loses money, the cloud pays for it, and Jeff Bezos was running the world's most expensive loss leader on purpose. It was a great story. It was even true once — back when North America retail margins were paper-thin and International ran red.

Then the story stopped matching the filing. In 2024 Amazon's North America segment reported $25.0 billion in operating income — on its own, with no cloud money flowing in.¹ And it is not a sleepy number: North America operating income grew 68% in a single year, from $14.9 billion in 2023.¹ The retail operation that the world insists is bleeding is, at the segment level, one of the largest profit pools in American business. Even the company's own analysts had flagged this years earlier — Amazon's North America retail was already running a few billion dollars of operating profit back in 2017, which means the subsidy framing was obsolete long before anyone updated it.⁷

The profit that doesn't have its own line

Here is the part the financial statements actively obscure. Amazon reports three segments — North America, International, and AWS — a structure it locked in back in 2015.⁶ Advertising is not one of them. Its $56.2 billion of 2024 revenue gets a line of its own, but its profit is folded invisibly into the geographic retail segments.⁴ So when you read that North America earned $25.0 billion, you are reading a number that already contains an unknown — and almost certainly large — slug of ad profit. That matters because selling a sponsored slot has almost none of the cost of selling a physical good. There is no warehouse, no truck, no return. The mechanism is the same one that makes a search engine print money: you've already built the audience for another reason, so the marginal cost of showing them one more paid result rounds to zero. Amazon built the audience to sell shoes. It rents that same attention back to the shoe brands, twice.

This is why the cross-subsidy question has the wrong shape. The interesting flow isn't cloud-to-retail. It's that the retail surface area — the search bar, the product page, the checkout — has become the distribution channel for a third business that is more profitable than the goods it sits on top of. The store isn't the thing being subsidized. The store is the thing generating the audience that two other businesses monetize. In Q4 2024 advertising alone pulled in $17.3 billion, up 18%, which the company described as still accelerating off a very large base.⁵

But isn't the store still just bait for the real businesses?

The honest objection is that this is a distinction without a difference: of course the engines aren't truly separate, because the same balance sheet, the same Prime membership, and the same warehouses serve all of them, so calling them three businesses is an accounting fiction. Fair — and Amazon's own disclosure makes it impossible to fully disprove, because the company never publishes a retail-only P&L stripped of ad profit, which means we can't independently confirm that the store stands alone.⁸ But the steelman cuts the other way too. The fact that the businesses share infrastructure is exactly what makes the 'subsidy' word wrong. A subsidy implies one business pays so another can survive. What's actually happening is the reverse: a single asset — hundreds of millions of shopping sessions — is being harvested three times over. That isn't a charity propping up a failure. It's a flywheel where each turn of the store makes the cloud and the ad business cheaper to run, and the company has consistently insisted, against the popular narrative, that nothing is being propped up at all.⁷⁸

So which part of Amazon quietly funds the rest? The lazy answer — AWS — is still the biggest single number, and it's the wrong frame. The store doesn't survive on the cloud's charity. The store is a machine for manufacturing attention, and Amazon learned to sell that attention to the brands competing for it, at margins that don't have to ship anything. The cloud sells servers. The store sells goods. And inside the store, in a grey word next to your search results, a third business sells the only thing in the building that costs nothing to make: the next click. They told you it was a bookstore. It was a tollbooth wearing a storefront.