How Amazon Actually Makes Its Money

You buy something on Amazon for $30. Amazon didn't make it. Doesn't own it. In most cases never touched it. Someone else took the manufacturing risk, sourced the inventory, and put up the capital. That someone is now handing roughly $13 of your $30 straight back to Amazon for the privilege of having been visible.

That's the animating tension in a recent breakdown by Tony Talks Business, which works through Amazon's seven revenue lines with a methodical focus on margin. The core finding is worth sitting with: Amazon reported $717 billion in revenue last year and converted $80 billion of it into operating profit. Of that profit, 57% came from a division — Amazon Web Services — that has never shipped a single physical product.

The vans, the boxes, the sortation centers: that's the least profitable part of the operation. The business most people picture when they think "Amazon" is, by design, barely a business at all.

The shop is a loss leader with excellent PR

Amazon's first-party retail — where it actually buys inventory and resells it to you — generated $269 billion in revenue last year. It is the single largest line on the page. It is also, by the standards of everything else in the company, a financial disappointment. The entire North America and international retail operation, $588 billion in combined sales, produced $34 billion in operating income. That's a margin in the low single digits after you've accounted for warehouses, drivers, returns, and the two-day promise.

The video is direct about this: "Nothing about that is a failure. It is the point." Retail exists to generate traffic. Traffic is the raw material that feeds the actually profitable businesses.

The private label strategy fits this logic. Amazon Basics, Amazon Essentials, and roughly 30 other house brands were built using data Amazon already held: which categories were large, undifferentiated, and ripe for a cheaper version. Batteries. Cables. Storage boxes. Amazon didn't need to guess. It could read the sales velocity and price sensitivity on its own platform and manufacture accordingly. The brands were never meant to dominate. They were meant to extract the margin from the least defensible corners of the catalog.

The marketplace math that sellers don't see coming

61% of what sells on Amazon is sold by independent third parties. Amazon doesn't book those as sales — it books what it charges the sellers. Last year, that number was $172 billion.

Here is what that charge looks like on a single $30 transaction, per the video's breakdown:

• Referral fee: 15% for most physical goods. That's $4.50 before anything moves.
• Fulfillment by Amazon (FBA): Around $5.25 per unit for a standard lightweight item.
• Storage: Roughly $0.30, spread across average inventory.
• Advertising: Sellers commonly spend around 10% of revenue on Amazon ads just to maintain visibility. That's another $3.

Total: $13.05 out of $30, and the video calls this "the conservative version." In competitive categories — with returns processing, promotional coupons, and elevated ad spend — the take rate can approach half.

The FBA piece deserves particular attention because of how its optionality works in practice. FBA is technically optional. Practically, it's not, because FBA is what earns a listing the Prime badge, and the Prime badge is what makes a listing competitive against the identical product sitting one slot over. As the video puts it: "A seller who declines FBA to save $5.25 a unit does not simply keep the money. They become slower. Being slower cost them the box, and losing the box cost them the sale entirely rather than a slice of it. That is how an optional service becomes a compulsory one without anybody ever having to make it compulsory."

The Buy Box — the button that defaults to one seller when multiple sellers share a listing — concentrates roughly nine out of ten purchases into whichever seller wins it. Amazon controls who wins it. FBA is among the strongest signals. The structure thus functions as a closed loop: pay the fees to win the box, win the box to make any sales at all.

This is precisely what the Federal Trade Commission and 18 state attorneys general allege in an antitrust case currently scheduled for trial in 2027. Their argument is that Amazon uses its position to prevent sellers from offering lower prices elsewhere and to funnel them into fulfillment services they'd rather avoid. Amazon's position is that sellers chose to participate and grew their businesses in the process. Neither side is going to resolve that dispute here, but the fee arithmetic is not in dispute.

Advertising: the most valuable pixel on the internet

Amazon's advertising business brought in $69 billion last year, making it one of the three largest ad businesses on Earth alongside Google and Meta. What it sells is placement: the top of a search results page, auctioned in real time to whoever bids the most.

The edge Amazon has over every other advertising platform is structural. Google knows you want a coffee grinder. Amazon knows you want a coffee grinder, your card is already saved, and one click separates you from the transaction. That specificity — capturing intent at the precise moment of purchase — commands conversion rates no other platform can replicate. And because Amazon owns both the ad and the checkout, it can prove the connection between the two without inferring anything. Every other major ad platform has spent years trying to demonstrate that their ads drive purchases. Amazon just reads its own records.

There's a tension embedded in this business that the video identifies clearly: "Every ad slot at the top of a search result is a slot that used to be occupied by whatever product was genuinely best. As the ad load rises, the ranking you are looking at drifts further from the one the algorithm would produce on merit." Amazon is monetizing the credibility of its own search results, and at some point, degraded results start eroding the shopping habit that made those results worth anything. Nobody knows where that ceiling is. Amazon, as the video puts it, is finding it empirically.

Prime: a subscription designed to make you forget you're spending

More than 200 million paid members worldwide. $50 billion in subscription revenue, almost entirely from Prime.

The dollars from Prime subscriptions are not actually the point. Survey data cited in the video from consumer research firm CIRP suggests Prime members spend roughly twice what non-members spend on Amazon annually — in the range of $1,100 versus $600. That behavioral gap is the real product.

The mechanism is mundane and effective: once you've paid $139 upfront, shipping feels free, and every purchase you make somewhere else carries an implicit cost. "The friction that used to make you pause before buying an $11 item has been removed permanently in exchange for one annual payment you have already stopped noticing." Prime Video, Amazon Music, live sports rights — none of that needs to make money on its own. It just needs to make cancellation feel like a loss.

In September 2023, Amazon agreed to a settlement with the FTC related to Prime enrollment and cancellation practices. The terms were substantial — real money even by Amazon's scale — though it amounted to roughly three weeks of Prime subscription revenue.

AWS: the company that funds the company

Amazon Web Services generated $129 billion in revenue last year — 18% of the total — and $45.6 billion in operating income, which is 57% of all profit. The entire retail operation, at five times the revenue, produced $34 billion. AWS produced $45.6 billion on a fifth of the sales.

The economics are close to ideal for the seller. Data center infrastructure is expensive to build and cheap to scale. Every additional customer goes on capacity that largely already exists. And leaving is prohibitively difficult: migrating an entire company's infrastructure to a different cloud provider is a multi-year project with real execution risk. The lock-in is contractual too — the cheapest pricing requires committing to spending levels a year or three in advance, so a large share of that $129 billion was contractually secured before the calendar year began.

The AI infrastructure buildout is the bet on top of the bet. Amazon has disclosed that its AI and custom chip businesses have each reached significant scale. It designs its own processors — Trainium and Inferentia — partly to reduce dependence on external chip suppliers. The capital expenditure guiding figure is large enough to constitute a full-scale industrial program.

The structure is the strategy

The shape of it, laid out in full: retail runs near breakeven to generate scale and habit. Scale produces sellers who need to pay for visibility, which is the advertising business. Habit produces subscribers who are locked in, which is Prime. Volume justifies a logistics network so large it became the country's biggest parcel carrier and could then be rented out as a service. All of it sits underneath a cloud computing division that has nothing to do with shopping but funds the whole experiment.

Amazon is, as the video summarizes it, simultaneously "the landlord, the referee, and one of the players" — operating the marketplace, setting its rules, and competing within it. That structure held together through two decades of near-zero reported profit because the cash was real even when the earnings weren't, and the capacity being built kept converting into revenue.

The question worth sitting with now is whether the front door holds. Amazon's retail advantage rests almost entirely on being where product searches start. If AI shopping assistants begin routing purchase intent across the entire internet, comparing prices without any loyalty to a particular storefront, the value of a sponsored placement at the top of an Amazon search page changes materially. Amazon is building its own shopping AI for exactly this reason. Whether it can own the new front door the way it owned the old one is the only structural question that actually matters.