Amazon Didn't Plan to Own the Stack. It Got Cornered Into It, One Bottleneck at a Time.

In its first years, Amazon never wanted to touch a single book. The plan was elegant and asset-light: take the order online, hand it to a publisher, let someone else pack the box. That lasted only as long as someone else did the job well — and they didn't. Third-party fulfillment was weak, customer service worse, and Amazon was reactively dragged into building warehouses it had hoped to never own.⁸ Two decades later that same reluctant instinct — fix the thing nobody else will fix, then sell the fix — produced the most profitable infrastructure business in the world.

The official story is that Amazon is a master vertical integrator who saw the whole board years in advance: warehouses, then cloud, then private label, each a deliberate move toward owning the stack. Almost none of that is how it happened. Every layer was built to unstick an internal bottleneck, and only afterward turned outward and sold. The grand design is a story Amazon's success lets us tell backwards.

The accident that became the profit engine

Take the part everyone gets most wrong. The legend says AWS was born when Amazon noticed it had spare servers and decided to rent them out. AWS's own CEO has flatly killed that story — the spare-capacity tale, he said, 'won't die, but it isn't true.'⁵ What actually happened is far less cinematic and far more instructive. Around 2000, Amazon tried to build Merchant.com, a platform to run other retailers' storefronts, and discovered it couldn't do it efficiently. The engineers kept hitting the same wall: every project that should have taken three months was spending three months just standing up the database, compute, and storage layer underneath it.⁶ The problem wasn't a shortage of servers. It was the absence of standardized, reusable infrastructure that internal teams could call on demand.

So Amazon forced every team to build in decoupled pieces, talking to each other only through APIs — and, almost without noticing, became a services company internally.⁶ The leap was reconceiving that internal plumbing as a product other companies would pay for. But — and this is the detail that destroys the spare-capacity myth — the internal services couldn't simply be flipped on for outsiders. The external products were built from scratch.⁵ The first true infrastructure service, SQS, didn't launch until November 2004; S3, the landmark, arrived in March 2006, with EC2 about five months after.⁷ What looked like renting out leftovers was years of deliberate construction aimed at a bottleneck Amazon had hit first because it was bigger than anyone else.

Why the boring layer makes all the money

Here is the thesis, plainly: Amazon's vertical integration is not a plan executed but a habit compounded — solve your own bottleneck, then monetize the solution — and the durability rests not on vision but on the brute economics of infrastructure too expensive to copy. The proof is in where the money lives. In 2024, AWS produced $39.8 billion of operating income on $107.6 billion of revenue. The North America retail business — the original company, the warehouses, the storefront, nearly four times the revenue at $387.5 billion — produced $25.0 billion.¹ The store everyone shops at is not the business that pays for everything. The plumbing is.

The compounding is the whole point. Ten years before, AWS earned $4.6 billion when all of Amazon earned $89 billion; by 2024 AWS alone earned $108 billion and the company earned $638 billion.² Total operating income jumped 86% in a single year, to $68.6 billion.¹ Each layer built to serve Amazon ended up serving the world, and each external customer paid down the fixed cost of infrastructure that Amazon would have built anyway. That is the self-reinforcing trick: the warehouse that ships your order ships a seller's order through Fulfillment by Amazon; the servers that run the storefront run a startup's app. The internal bottleneck and the external product share one bill.

Isn't a happy accident still just luck?

The fair objection is that 'compounding accident' is too generous to Amazon and too convenient for this argument: plenty of companies hit internal bottlenecks and build internal tools; almost none turn them into the most profitable infrastructure business on earth. Doesn't that gap require vision after all? Partly — the leap from internal plumbing to external platform was a genuine act of imagination, and it happened around 2003, not by inertia. But notice what the necessity bought that vision alone couldn't: scale and timing. Amazon hit the infrastructure wall first because it was already operating at a size no one else had reached, which meant it built the standardized services first, sold them first, and let early customers fund the depth of the stack before competitors knew there was a market.⁵ The accident wasn't the insight. The accident was being forced to do the expensive, unglamorous work years ahead of everyone else — and that head start is the moat, not the cleverness.

Too embedded to cleanly unwind

The same integration that prints money is now the heart of the government's case against it. In September 2023, the FTC and 17 state attorneys general sued Amazon for illegally maintaining monopoly power.³ Read past the headlines and notice what the lead theories actually are — not, despite the popular telling, primarily about Amazon favoring its own private-label brands in search. The complaint's two core theories are anti-discounting practices that allegedly stop sellers from offering lower prices on rival platforms, and the conditioning of Prime eligibility on the use of Fulfillment by Amazon.³ In other words: the regulators are aiming squarely at the seams of the vertical stack — the points where one Amazon layer makes you use another. The self-preferencing claim, that Amazon 'degraded the quality of the search results,' is supporting context, not the spear.⁴

And here is why the integration that invited the lawsuit also complicates it. When fulfillment, the marketplace, and the loyalty program are wired into one another, there is no clean line to cut. Unwinding the bundle means redesigning how millions of sellers reach customers — which is precisely the kind of remedy courts move slowly toward. The case is scheduled for trial in October 2026; no liability has been found.⁴ The structure is the target and the defense at once.

Amazon never sat down and drew the stack. It reached for warehouses because publishers fumbled the boxes, built reusable infrastructure because its engineers were drowning in plumbing, and discovered each time that the fix was worth more sold than kept. The strategy that looks like foresight is really a reflex compounded across two decades — fix the bottleneck, then bill the world for the fix. The genius wasn't seeing the whole board. It was being big enough to hit every wall first, and disciplined enough to turn each wall into a tollgate. You can copy the idea in an afternoon. The twenty years and the billions it took to build the road underneath it are the part no one can borrow.