The Amazon Web Services origin story (part 1)
The AWS origin story begins inside Amazon’s own architectural crisis: remove the “muck” consuming engineers, force every team behind service interfaces, then realize that the resulting Internet operating system could become a product for everyone.
Before Amazon Web Services launched, building a web company meant buying servers, leasing data-center space and guessing demand before the first customer arrived. Amazon faced the same constraints at extraordinary scale.
This narrated history draws on Andy Jassy, Werner Vogels, Jeff Barr and early EC2 engineer Matt Klein to trace the internal changes that made AWS conceivable—from Amazon’s distributed computing manifesto to its first external developer service.
The Q&A below is experimental — extracted from the episode transcript with AI. If you spot a mistake, let us know at jack@scalingdevtools.com.
What problem inside Amazon set the stage for AWS?
Amazon’s fast growth exposed a two-tier architecture built around a monolithic application and shared relational databases. By 1998, those databases were bottlenecks, and senior engineers proposed separating presentation, business logic and data while improving reliability, performance, security and cost. The distributed computing manifesto was an attempt to remove the undifferentiated work Jeff Bezos called “muck.”
What did “muck” mean?
It was the 70–80% of infrastructure work required to create a stable platform before a team could do anything distinctive. Once Amazon recognized how much of its own engineering effort disappeared into that layer, it could infer that every outside developer faced the same burden. A recurring internal cost was therefore also evidence of a much larger product opportunity.
How radically did Amazon reorganize its systems?
A company-wide directive attributed to Bezos required teams to expose data and functionality through network-accessible service interfaces. Direct database access, shared memory and hidden backdoors were forbidden, and the interfaces had to be suitable for external use. The decoupling was difficult because dependencies ran everywhere, but by 2005 it had changed Amazon’s engineering culture into one built around services.
What was Amazon’s first step toward an external platform?
Amazon first repurposed its retail systems for partners such as Target through merchant.com. Around July 2002, Amazon Associates Web Services let developers query Amazon programmatically. Jeff Barr recalled tens of thousands of registered developers and close to ten million calls a day, producing sites Amazon had never imagined—a direct demonstration that outsiders could extend the platform beyond the internal roadmap.
What was the “Internet operating system” insight?
At a 2003 leadership off-site, Andy Jassy and colleagues asked what would happen if developers could build applications from infrastructure services instead of owning the machinery beneath them. The Internet itself would become the operating system. When they listed its necessary components, they concluded that none had been adequately built and that Amazon’s experience operating technology at retail scale gave it something meaningful to contribute.
Was AWS created from spare holiday capacity?
Matt Klein rejects the popular story that Amazon simply rented servers it did not need outside Christmas. His view is that the company identified a real need, was unusually positioned to serve it and executed faster than Google or Microsoft. The earliest product could be rough because Amazon was willing to put it into the world and learn while slower competitors deliberated.
Why did the opportunity fit Amazon?
Bezos later described three tests for entering a business: the market must be large and growing, offer high returns on capital and draw on Amazon’s existing expertise. Infrastructure services met all three. A team of roughly 50–60 people under Andy Jassy began building the Internet operating system that became AWS, turning hard-won internal capabilities into a platform for external developers.