How Jeff Bezos Deployed Amazon Without AWS

Introduction

Today, spinning up a website is as simple as running "gcloud app deploy" or "terraform apply". But back in the mid-1990s, when Jeff Bezos was launching Amazon, cloud computing didn’t exist. There was no AWS, no Azure, no GCP—only bare-metal servers, co-location facilities, and long nights of manual configuration.

So how did Amazon scale from a tiny online bookstore in Bezos’s garage to a global e-commerce powerhouse—all without the cloud? Let’s step back in time and explore.

1. The World Before Cloud

In the mid-90s:

Hosting meant buying physical servers (usually Dell or HP).
Networking was set up with routers, switches, and ISPs you negotiated with.
Scaling wasn’t auto-scaling—it meant physically ordering more machines.
Deployments involved FTP, tapes, and sometimes sneakers to the data center.

Bezos and his early team had to work in this environment, balancing growth with extremely limited infrastructure options.

2. The Garage Setup

Amazon famously started in Bezos’s garage in Bellevue, Washington. The first setup included:

Sun Microsystems servers for running the website.
A small UPS (uninterruptible power supply) to keep systems running during power hiccups.
A 56kbps modem line for connectivity.
Home-built scripts to restart services when they crashed.

Every order was routed through a simple backend that emailed details for manual fulfillment. No serverless queues, no managed databases—just raw scripts and persistence.

3. Scaling Without AWS

As orders grew, the tiny garage setup couldn’t keep up. Bezos moved to co-location data centers, essentially renting racks in third-party facilities. Scaling meant:

Buying more servers upfront.
Shipping them to the data center.
Manually installing operating systems and configuring networking.

Every surge in traffic—especially during holidays—required guessing demand months ahead, often leading to overprovisioning hardware. Unlike today’s elastic compute, there was no “scale to zero” when traffic dropped.

4. How Deployments Worked

Deployments at Amazon in the 90s were radically different:

Code pushed via FTP or manual scripts.
Downtime was expected—taking the site down briefly was normal.
Logs had to be manually rotated, monitored, and analyzed.
Databases ran on traditional relational systems (like Oracle) without managed backups or replicas.

Imagine debugging production at 2 a.m. when the only monitoring tool you had was "tail -f logfile".

5. Resilience Without the Cloud

To survive:

Amazon invested in redundant servers for failover.
Early engineers wrote load-balancing logic themselves (before ELBs or GCLBs).
They built internal deployment tooling—primitive versions of what we now call CI/CD.

In fact, many of the pain points from this era directly inspired the creation of AWS years later. Bezos realized that the infrastructure Amazon had to build internally was something every company would eventually need.

6. Lessons for Today’s Builders

Looking back, Amazon’s early journey teaches us:

Constraints fuel innovation: Amazon had to invent solutions because nothing existed.
Manual scaling builds discipline: Teams couldn’t “throw servers at the problem,” so efficiency mattered.
Cloud is a privilege: Today, startups deploy globally in minutes with infrastructure-as-code—something unthinkable in the 90s.

Conclusion

Jeff Bezos didn’t have AWS when he launched Amazon—he had a garage, a handful of servers, and a scrappy team willing to do whatever it took. The fact that Amazon scaled under those conditions is nothing short of remarkable.

Ironically, the very struggles Amazon faced in the 90s laid the foundation for AWS, which went on to revolutionize how modern software is built.

So next time you type "gcloud", "aws", or "az" to spin up infrastructure in seconds, remember—Amazon itself was born in a world without the cloud.

Liked this post? Follow me for more explorations of how tech giants were built before the cloud era.