2-Amazon Web Services

By 2002, Amazon had a problem that had nothing to do with selling books. It could not ship software fast enough. Every new feature required engineering teams to negotiate with other engineering teams for access to databases, storage, and compute. A project that should have taken a week took a quarter, because most of the quarter was spent on coordination. Jeff Bezos’s response was an internal mandate with a famously blunt ending: every team would expose its data and functionality through service interfaces, all teams would communicate only through those interfaces, and anyone who didn’t comply would be fired. The point was internal velocity. Amazon was not trying to invent an industry. It was trying to stop its own engineers from waiting on each other. But a company that forces every internal capability behind a clean, documented, network-accessible interface has — without quite meaning to — built something that looks a great deal like a public utility. In 2003, Chris Pinkham and Benjamin Black wrote a memo describing Amazon’s infrastructure as a standardized service that could be sold to anyone. Bezos took it further than the memo did. Amazon Simple Storage Service launched in March 2006. Elastic Compute Cloud followed in August. The pitch was almost absurd at the time: rent a server by the hour from a bookstore. Two features of the origin story explain why nobody caught up for years. First, it was battle-tested before it was a product. Most infrastructure companies build a system and then find out whether it survives real load. AWS had already survived Amazon.com — including the retail apocalypse of Q4 — before an external customer ever touched it. The scar tissue came free. Second, the economics were structurally hostile to competitors. Amazon was a low-margin retailer that had learned to run infrastructure ruthlessly cheaply because its retail margins gave it no other option. Selling that capability at prices that looked thin to a traditional software company still looked fat next to running a warehouse. AWS could cut prices repeatedly — it has done so dozens of times — in a way that enterprise hardware vendors, structurally dependent on high margins, could not answer. The scale of the accident is the part worth sitting with. Amazon Web Services grew into a business generating over $100 billion in annual revenue, and for years it produced the majority of Amazon’s total operating income. The internal tooling built so that engineers would stop waiting on each other became more profitable than the store it was built to serve. It also created a category. Essentially every startup founded after 2008 was built on rented infrastructure, which collapsed the cost of starting a software company from millions of dollars in servers to a credit card and an afternoon. The entire modern startup economy is downstream of a memo about internal coordination overhead. Nobody wrote a business plan for this. Somebody wrote an engineering mandate, and the