It has been 20 years since Amazon Web Services, or AWS, jumped into the cloud computing market with S3 (Simple Storage Service) on March 14, 2006.

At the time, Amazon announced S3 in a notice of just one paragraph. A blog post written by Jeff Barr, an Amazon chief evangelist, was also only a few paragraphs long. It was a time when nobody expected the service to shake up the IT industry.

According to AWS, S3 was designed based on five principles: built-in security, 99.999999999 percent durability, availability, performance and elasticity. Those principles remain unchanged 20 years later.

But its scale has grown to an unimaginable level. At launch, S3 had only about 1 petabyte of capacity across 15 racks and 400 storage nodes. It now stores hundreds of exabytes of data across 39 AWS regions and 123 availability zones and processes more than 200 million requests per second.

The number of stored objects exceeds 500 trillion. The maximum size of a single storable object has increased 10,000-fold, to 50 terabytes from 5 gigabytes. The price per gigabyte has fallen 85 percent, to about 2 cents from 15 cents. AWS said customers using S3 Intelligent-Tiering have cumulatively saved more than $6 billion in storage costs.

It has also evolved technically. AWS is rewriting S3 core code in Rust over 8 years. Rust blocks entire types of memory errors at the compilation stage. It also uses automated verification techniques based on mathematical proofs to check consistency each time code changes are made.

The key to S3 durability is a microservices system that continuously checks all data at the byte level. If signs of data corruption are detected, an automated recovery system is triggered.

Backward compatibility is another keyword AWS emphasises. Code written for S3 in 2006 still runs today without modification. The infrastructure has been replaced over multiple generations and the request processing code has been completely rewritten, but the data stored 20 years ago and the APIs remain intact, the company said.

AWS aims to grow S3 beyond simple storage into a foundation for all data and AI workloads. The recently launched S3 Tables is a managed table service based on Apache Iceberg, and S3 Vectors is vector storage for semantic search and retrieval-augmented generation (RAG). AWS said S3 Vectors accumulated more than 250,000 indexes and more than 40 billion vectors within 5 months of launch.