In the ’90s, I thought of Amazon as an online retailer. I bought the stock at $15, sold it at $40, and congratulated myself on my financial acumen. I believed in disintermediation. Cut out the middleman, take some margin, grow fast. I also thought there was a non-trivial chance the whole thing would flame out once the novelty wore off.
Boy, was I wrong.
What I missed was that Amazon was not building a retail business. It was building infrastructure.
Today I think of Amazon as an infrastructure company that happens to sell stuff.
Want to host a website? They’ll provision you compute, object storage, databases, load balancing, and global DNS in minutes. Need elasticity? Autoscaling groups will spin instances up and down faster than you can requisition a server in a traditional enterprise. Need durability? S3 will give you eleven nines and replicate your bits across availability zones without you ever seeing a disk drive.
Want to stream video to millions of people simultaneously? They have a global content delivery network, edge caches, backbone capacity, and traffic engineering that would have made the ARPANET crowd stare in disbelief. They don’t just ship packets; they manage latency, redundancy, and failover at planetary scale.
Want a storefront? The marketplace is a turnkey commerce stack: catalog management, payment processing, fraud detection, recommendation engines, advertising rails, and customer acquisition baked in. You plug in inventory; they supply the demand aggregation and the algorithms.
Need fulfillment? Now we’re in the physical layer. Robotic warehouses, inventory forecasting models, route optimization, last-mile logistics, and a transportation network that increasingly bypasses traditional carriers. It is a distributed system with forklifts.
Underneath all of that is a set of horizontally reusable primitives: distributed computing, large-scale storage, identity and access control, telemetry, machine learning services, and a global private backbone. Retail is just one workload running on top of that stack.
In the ’90s, I thought I was investing in a better bookstore. What I was actually looking at was the early build-out of a global operating system for commerce and computation.
The part I really missed in the ’90s wasn’t just that Amazon would become infrastructure. It was which part of that infrastructure would matter most economically. Today, Amazon Web Services accounts for less than a fifth of total revenue, but it routinely generates well over half of the company’s operating income. In other words, the cloud business is a minority of the top line and the majority of the profit pool.
Retail (North America and International combined) still produces the bulk of revenue, but at far lower margins. The physical and digital commerce layers are enormous in scale, but they are comparatively thin in profitability. The irony is hard to miss: the thing most people think of as “Amazon” is the storefront, while the thing that actually funds the enterprise is the compute fabric underneath it.
Leave a Reply