Amazon, eBay, and the Invention of the Take-Rate Economy

Two garages, one invention

Jeff Bezos incorporated Amazon.com in Bellevue, Washington in 1994 and opened it to customers in July 1995, as an online bookstore. Books were a deliberate choice: no physical store could stock the millions of titles in print, so a catalog that had to look infinite in a browser needed no warehouse to match. Bezos named the company after the river carrying the largest volume of water on Earth, and told early investors it would not stay a bookstore for long. It did not. Amazon's own later description of itself, "the Everything Store," became literal within a decade.

Two months after Amazon opened for business, Pierre Omidyar built a hobby auction page onto his personal website in San Jose, California, and called it AuctionWeb. The company Omidyar built around that page became eBay in September 1995, and unlike Amazon it never sold anything itself. From its first day, eBay took a cut of what strangers agreed to trade with each other and left the goods, the risk, and the shipping to them.

Bezos was explicit, in his first shareholder letter after the 1997 IPO, that the company would optimize for long-term market position over short-term profit, a philosophy he titled "It's Still Day 1" and repeated in every letter for the next two decades. That patience is easy to read as a founder's temperament. It is better read as a bet on a specific structure: a company that made money on the flow of transactions across a shelf, rather than on the margin of any single sale, could afford to grow the shelf for years before the take rate on it mattered.

Both companies went public inside the same eighteen months, and Wall Street read them as two different kinds of business. Amazon priced its IPO at 18 dollars a share on May 15, 1997, opening at a market value of about 438 million dollars and closing its first day up roughly 30 percent, at 23.50 dollars, a respectable debut for a company still losing money on every book it shipped. eBay priced its IPO at the same 18 dollars a share on September 24, 1998, and closed its first day around 47 to 53 dollars, a gain near 168 percent. That gap in first-day performance was not an accident of hype; it reflected which company investors understood, correctly, carried the lighter balance sheet, one built to buy and sell goods, the other built from the outset to own nothing and collect a fee on every trade that crossed its shelf.

The take rate, defined

A take rate is the share of the value crossing a platform's shelves that the platform keeps, regardless of who made the item or who bought it. eBay is the cleanest example of the model in its pure form. By 2023 the company reported 132 million annual active buyers moving 73 billion dollars in transaction volume, and it kept 13.81 percent of that volume as revenue, a figure drawn entirely from commission because eBay owns none of what changes hands on its site. That number is the take rate stated plainly: not a markup on cost, not a margin on manufacturing, a toll on the act of exchange itself.

The distinction matters because it describes a genuinely new category, not a variant of an old one. A retailer buys low and sells high, carrying the risk of goods that do not sell. An exchange, in the older sense, such as a stock exchange, charges a fee per trade and never owns a share of what it lists. eBay was built as the second kind of business from its first auction: it never held the goods that moved across its listings, and its revenue was a percentage of other people's transactions from day one.

Amazon reached the same category by a slower route. For its first five years it was a retailer in the ordinary sense, buying books at wholesale and selling them at retail, carrying real inventory and real markdown risk. That changed in 2000, when Amazon opened its Marketplace to outside sellers, letting other merchants list their own goods alongside Amazon's own inventory in exchange for a referral fee on each sale. The move converted Amazon from a pure retailer into a hybrid: part retail margin, part commission on merchandise it never touched, and structurally, from that year forward, it was also collecting a take rate.

Renting the shelf, then renting the server

Amazon's Marketplace fees solved one problem: how to earn revenue from inventory the company did not own. In 2006 Amazon solved a second, less obvious version of the same problem: how to earn revenue from infrastructure it had built for itself and no one else. The company had spent years over-provisioning its own servers to survive the traffic spike of the December holiday season, capacity that sat mostly idle the rest of the year. Amazon Web Services turned that idle capacity into a product, renting compute and storage by the hour to any business that wanted it, and became the first hyperscale public cloud business.

The economic logic connecting eBay's commission, Amazon's referral fee, and AWS's metered billing is the same logic, applied to three different kinds of shelf. eBay charged a cut of a transaction between strangers. Amazon Marketplace charged a cut of a sale between a merchant and a customer. AWS charged rent on computing capacity itself, whether or not the renter ever sold anything on top of it. In each case the company's revenue could grow with someone else's activity, another seller's sale, another developer's server load, without a matching growth in the assets Amazon or eBay had to buy, insure, and depreciate.

This is the moment "platform" stops being a loose description of a website and becomes a load-bearing structure of the balance sheet. By the end of this period, in 2010, Amazon was no longer simply a company that sold things online; it was a company renting out three different kinds of infrastructure, shelf space to sellers, checkout and fulfillment to those sellers' customers, and computing power to anyone at all, and had discovered that rent scaled in ways that owning inventory never could.

What that rental logic would become over the following decade sits mostly outside this article's window, but the direction is worth naming. AWS went on to become one of Amazon's largest profit engines, and the hyperscale cloud category it created now underwrites much of the same AI infrastructure that answer engines run on, a second-order consequence that traces back to a bookseller's decision to rent out its own spare servers rather than let them sit idle.

Why the marketplace model outlived the crash

The doubt showed up in print before it showed up in the stock price. The financial magazine Barron's ran a widely read cover in 1999 asking whether the company deserved the nickname "Amazon.bomb," a skepticism about e-commerce profitability that the crash of the following year seemed, for a while, to confirm. The wider market for internet retail was collapsing around both companies at the same time they were building this model. Between 2000 and 2001, Amazon's stock lost the large majority of its market capitalization even as the company's revenue kept growing, a gap that measured investor doubt rather than business failure: Wall Street had stopped believing that unprofitable e-commerce spending would ever convert into the kind of profit a normal retailer could show.

Pets.com is the era's clearest counter-example. The company launched in November 1998, spent heavily on a Super Bowl advertisement and a balloon in Macy's Thanksgiving Day Parade, and shut down in November 2000, not two years after it opened. Pets.com was a retailer in the old sense and nothing else: it bought pet food and supplies, an expensive, low-margin category to ship, and tried to buy market share with venture money rather than collect a fee on transactions it did not have to fund. When investor patience ran out, there was no commission revenue underneath the business to keep it standing.

eBay, by contrast, carried no inventory risk at all through the crash; a slow quarter meant fewer trades, not a warehouse of goods nobody wanted. Amazon, half retailer and increasingly half landlord, survived on the strength of the half that was becoming a marketplace, and, from 2006, a landlord of computing as well. Its revenue kept climbing through the worst of the crash even as its stock fell. Two decades later, in 2021, Amazon surpassed Walmart to become the largest retailer outside China, helped by its Prime subscription program, which had reached roughly 200 million subscribers worldwide, a scale eBay's pure-commission model never tried to match but that Amazon's hybrid model made possible.

Exporting the rules of the exchange

Being first mattered for a reason beyond revenue. eBay and Amazon built the first digital marketplaces trusted at real scale, trusted enough that a stranger would pay in advance for an item they had never touched, shipped by someone they had never met. That trust did not exist by default; it had to be manufactured, and the two companies manufactured it with tools no government had specified: reputation scores built from buyer and seller feedback, buyer protection guarantees, standardized listing formats, and fulfillment and shipping-time expectations enforced by the platform's own policies rather than by law.

eBay extended that same trust infrastructure in 2002, when it acquired the payments company PayPal and folded a verified way to move money directly into its marketplace, so a buyer and a seller who had never met could complete a transaction without either one handing the other a bank account number. Reputation told a stranger who to trust; PayPal gave that stranger a mechanism to act on the trust without exposing themselves. Both were rules the platform wrote for itself, not rules any regulator required.

Once those tools worked, and worked at the scale of tens of millions of buyers, they became something closer to a protocol than a policy. A seller anywhere in the world who wanted to reach the deepest concentration of paying attention on the internet had, functionally, two choices: build a rival marketplace at comparable scale and trust, an option available to almost nobody, or accept the rules the American platform had already written. A trader listing on eBay in the early 2000s, wherever they were based, operated under an English-language feedback system, a buyer-protection standard modeled on American consumer expectations, and a dispute-resolution process the platform itself designed.

No treaty produced any of this. The World Trade Organization's moratorium on customs duties for electronic transmissions, agreed in 1998, addressed tax administration, not the commercial norms of how a marketplace should behave toward buyers and sellers. The rules that actually governed cross-border online trade, what counted as a trustworthy seller, what a buyer was owed if a package never arrived, were written by two American companies' terms of service, adopted worldwide because that is where the paying customers already were, years ahead of any government catching up.

The pattern set in those terms of service has a direct descendant in the surfaces this publication studies. The businesses an AI answer engine decides to name today are being judged, again, largely by rules a small number of American companies wrote for their own systems, verification signals, structured data, a legible and consistent public record, that read like an update of what eBay's feedback score and Amazon's seller metrics asked for twenty five years earlier: prove yourself in a form the platform, not the customer, can read.

What the record actually supports

None of this claims eBay and Amazon coordinated on a shared plan, or that governments were absent from e-commerce during this period. They were not: the United States and other countries were writing consumer-protection, tax, and antitrust law throughout the late 1990s and 2000s. What the record supports more narrowly is that the operating norms which made cross-border digital commerce function at all, reputation, buyer protection, standardized fulfillment, were set in practice by these two companies before any comparable multilateral rulebook existed, because they had built the first marketplaces used at meaningful scale, and no government-run standard has displaced them since.

The take-rate economics behind this argument rests on established figures: eBay's 13.81 percent 2023 take rate, Amazon's 2000 Marketplace launch and 2006 AWS launch, and the founding and IPO details of both companies are drawn from company disclosure and contemporaneous reporting, not estimate. What is more interpretive is the claim that the marketplace structure itself, rather than differences in product category, funding discipline, or plain luck, is what let Amazon and eBay outlive Pets.com and its peers; the contrast is strong but no controlled comparison isolates the one cause from the others.

Read as a whole, the two 1995 startups did not just survive a crash that killed most of their peers. They invented the category that would go on to define how a very large share of the internet makes money: not by selling things, but by charging rent on the shelf where the selling happens, and by writing, in the process, the operating rules the rest of the world had to learn.