Amazon Didn't Plan to Sell Cloud Computing. It Got Good at Plumbing and Noticed.

Around 2000, Amazon tried to do a sensible, boring thing: let other retailers run their stores on Amazon's e-commerce software. The project was called Merchant.com, and it exposed an awkward truth. Amazon's own infrastructure — the plumbing that ran the store — was a tangle. Teams couldn't build on each other's work without tripping over it.⁴ That mess, not a vision, is where the most profitable business Amazon would ever own quietly began. The company that sold you a book was about to discover it had accidentally become very good at something else entirely.

The official story of Amazon is a story of vision: a founder who always meant to build the everything-store, then cleverly rented out his spare servers as cloud computing. It is a great story. It is also wrong on both counts. The expansions weren't a master plan. They were forced infrastructure bets — each one made to solve a problem in front of the company, each one accidentally revealing a market nobody had set out to enter.

It started as a bookstore, and meant it

Amazon was incorporated in 1994 — originally as Cadabra, Inc. — and opened as an online bookseller in July 1995.¹ Books weren't a placeholder for grander ambitions; they were the right first product for selling online, where a near-infinite catalog beats any physical shelf. By the time Amazon went public in 1997, it was still, in its own words, an online bookseller. The 1997 shareholder letter reports 838% revenue growth to $147.8 million and more than 1.5 million customers — and frames the company squarely around books.³ Music, video, and electronics came later, from 1998 onward. The everything-store wasn't a launch plan. It was a destination Amazon walked toward one adjacency at a time.

There's a tidy myth worth clearing on the way through: the famous '$1.50 IPO price.' Amazon actually priced at $18.00 a share when it began trading in May 1997.² The $1.50 figure is a split-adjusted retrospective — Bezos himself used it that way in the 2020 shareholder letter, looking backward through later stock splits.⁸ It's a small thing, but it's the same pattern as the rest of the legend: a clean number that makes the past look more deliberate than it was.

The plumbing nobody meant to sell

Here is the part everyone gets backwards. AWS is not Bezos quietly noticing his servers sat idle at night and deciding to rent them out. The person who wrote the original AWS vision document, Andy Jassy, has said plainly there was no single aha moment and no plan to monetize spare capacity. What happened instead: after the Merchant.com infrastructure mess exposed how fragile Amazon's internal plumbing was, the company spent years forcing its teams to rebuild it as clean, reusable building blocks. At a 2003 executive retreat at Bezos's house, the team looked at what they'd built — and were surprised to realize that running reliable, scalable infrastructure had quietly become one of the things Amazon was best at in the world.⁴ The product wasn't the spare capacity. The product was the competency they'd been forced to develop to keep the store running.

Even the launch date everyone repeats is fuzzier than the legend. The 'Amazon Web Services' brand first appeared in July 2002 — as a developer API platform, not cloud computing at all. The first true infrastructure service, the Simple Queue Service, entered preview in November 2004, more than a year before S3 existed.⁶ The cloud business most people picture launched in March 2006 with S3, with EC2 following in an August 2006 preview.⁵ No clean birthday, no single decision. Just a competency leaking into the open in stages, each one slightly surprising the company that built it.

The one thing that wasn't an accident

If the expansions were emergent rather than planned, what made Amazon — and not some better-organized rival — the one to catch them? The honest answer is a single, deliberate habit: a willingness to spend on infrastructure long before the payoff was visible, and to tolerate the thin or absent profits that follow. Amazon famously ran near break-even for years, and Bezos's 'It's still Day 1' framing was a permission slip to keep investing through the impatience.³ That capital patience is the unglamorous engine. A normally disciplined company would have starved an internal infrastructure rebuild that produced no revenue, killed a queue service in preview that nobody asked for, and never accumulated the slack that turned out to be a market. Amazon funded the plumbing past the point a quarterly mindset could justify — and then monetized infrastructure slack that competitors couldn't have justified building in the first place.

The numbers make the point with a force the narrative can't. In 2024, AWS produced $108 billion in revenue and $39.8 billion in operating income. Amazon's total operating income that year was $68.6 billion on $638 billion of sales — a thin 10.8% margin overall.⁷ Read those figures together and the company turns inside out: the retail giant everyone knows is the low-margin part, and the accidental cloud business carries most of the profit. The bookstore became the everything-store. The everything-store funded the plumbing. The plumbing became the business.

Isn't this just genius rebranded as luck?

The fair objection is that calling AWS 'emergent' undersells real strategic skill — that recognizing your own infrastructure as a product, then committing billions to it, is exactly the kind of vision that deserves the master-plan story. There's truth in that. Spotting the opportunity at the 2003 retreat was a genuine act of insight, not a windfall.⁴ But notice what the objection quietly concedes: the opportunity had to be discovered, because it wasn't designed. Plenty of large companies sat on impressive internal infrastructure in 2003 and sold none of it. What separated Amazon wasn't a prophet at the top — it was a structure that kept funding the plumbing long enough for the opportunity to exist, and a culture willing to look at its own back office and ask whether it was a product. The skill is real. It just isn't foresight. It's the discipline to keep investing in capability before the market for it shows up, and the alertness to recognize the market when it finally does.

Amazon's adjacency engine has no map and no destination. It has a method: solve your own hardest problem so thoroughly that the solution becomes sellable, fund that work past the point patience normally runs out, and stay alert enough to notice when a tool you built for yourself has quietly become a market. Books to everything to cloud looks, in hindsight, like a staircase climbed on purpose. It was really a series of forced steps, each one revealing the next only after Amazon had already paid to take it. The genius was never the plan. It was the refusal to stop building the plumbing — and the willingness to look down and realize the plumbing was the thing worth selling.