Korva is a scroll of the internet’s best ideas.
Compute without meetings
It took a long time to get Amazon.com onto AWS. What the effort drove was the understanding that Amazon couldn't keep having so many meetings, and needed compute that people could plug into through a purely API surface, with no one to talk to. If that worked for the internal retail teams, it could work for anyone. Retail turned out to be so big that Amazon had to start with outside customers, and AWS served external customers before it served internal ones.
Logistics follows the pattern
Logistics went the opposite direction. Amazon used outside providers like UPS, FedEx, and USPS, then built its own delivery network, and now offers it to third parties. The pattern repeats market after market: Amazon is its own first best customer, and that gives a new product the scale to get off the ground before it is sold to everyone else.
Chips take the same path
Chips follow the same path. The early versions of Graviton and Trainium were terrible. Inside managed services like the Redshift database service, customers aren't told which processor runs underneath, so Amazon can put its weak early chips there. That gave it the volume to iterate and improve them until they could be sold externally, and Trainium now runs Anthropic and other AI products.
Why do managed services hide the processor?
In a managed database service, the customer buys results such as query speed, not a particular machine. That lets the provider swap in new hardware without the customer rewriting anything. Renting a virtual machine is different, because the customer picks the instance type. AWS first launched its Arm-based Graviton processors in 2018.
Why does an API scale better than meetings?
A person can handle a handful of requests a day, while software answers requests at any hour without waiting for anyone. A well-defined interface also lets every team build on top of a service without negotiating with its owners. A widely told account of Amazon's history holds that around 2002, Jeff Bezos required every team to expose its data and functions through service interfaces.
How did Amazon build its delivery network?
Amazon began by relying on carriers such as UPS and FedEx, then added its own warehouses, vans, and drivers as its volume grew. It started leasing cargo aircraft in 2016, which gave it capacity for its own packages. Today the network also carries packages for outside sellers.
More from Invest Like the Best with Patrick O'Shaughnessy
