Amazon S3 Launches — AWS's Cloud Begins with Storage

Amazon's public cloud began on 14 March 2006, when Amazon Web Services launched Amazon S3 (Simple Storage Service): storage that any developer could rent over the web, with no upfront cost and no minimum fee, at $0.15 per gigabyte per month. Compute followed on 25 August 2006, when EC2 opened as a limited beta. The name Amazon Web Services is older — it dates from July 2002 — and a message-queue service had been in beta since November 2004. But S3 was the first time Amazon sold its infrastructure itself, metered and paid for, to anyone with a credit card.

The press release that announced it was short. Amazon called S3 "storage for the Internet", and described a service that let developers store and retrieve any amount of data from anywhere on the web through a simple interface.

AWS before and after S3

The 2002 API and the 2004 queue service both carried the AWS name, and the S3 press release itself counted AWS as having launched in July 2002, with more than 150,000 developers signed up since. What changed in 2006 was the business: instead of an API over Amazon's own data, S3 rented out raw capacity. The story of the name and of Amazon as a company is on the Amazon page.

What S3 offered on day one

S3 stored objects — files of any kind — in buckets, addressed by a key. The launch specification was deliberately small:

The press release stated the design choice outright: "Amazon S3 is intentionally built with a minimal feature set." The release put the emphasis on simplicity and robustness rather than features. Andy Jassy, then vice president of Amazon Web Services, said in the release that S3 was based on the idea that quality Internet-based storage should be "taken for granted".

Amazon also pitched S3 as access to the same storage infrastructure it used to run its own websites. That was the company's claim; what outsiders could verify was the price list.

Where the data lived

At launch S3 offered no choice of location. When Amazon added European storage on 6 November 2007, its press release referred back to the "U.S. launch" of S3 and said that developers in Europe had been asking, ever since, to keep their data in Europe with lower latency. The European price was a little higher — $0.18 per GB per month.

Japan had to wait longer. On 2 March 2011 AWS opened the Asia Pacific (Tokyo) Region, its fifth region worldwide and its second in Asia Pacific after Singapore. Amazon's release listed S3 and EC2 among the services available there from the first day, with Japanese-language Premium Support. For a Japanese company, that is the date from which data in S3 could stay inside Japan.

How big it was

Twenty years later, on 13 March 2026, AWS published figures for the original system. S3 had launched with roughly one petabyte of capacity on about 400 storage nodes, in 15 racks across three data centres, with 15 Gbps of total bandwidth. The same post gave the 2026 scale: more than 500 trillion objects, more than 200 million requests per second, hundreds of exabytes, 39 regions. The maximum object size had grown from 5 GB to 50 TB, and storage that cost 15 cents a gigabyte was now slightly over 2 cents. All of these are AWS's own figures.

EC2 five months later

S3 was storage only. The second half of what people now mean by "the cloud" arrived on 25 August 2006, when Jeff Barr announced Amazon EC2 on the AWS blog. Each instance was the equivalent of a 1.7 GHz Xeon with 1.75 GB of memory, 160 GB of local disk and 250 Mb/s of network bandwidth, billed at 10 cents per hour. During the limited beta an account could run up to 20 instances.

EC2 was built on top of S3: Barr's post explained that each machine image (AMI) was a packaged-up operating system stored as an S3 object. With the two together, a developer could run a web service on rented machines and rented storage.

Why the date matters

What S3 put on the price list was a unit, not a package: a gigabyte stored for a month and a gigabyte transferred, with no upfront fee and no minimum. Other providers followed with different models: Google offered a platform for whole applications with App Engine in 2008, and Microsoft took Windows Azure to general availability in 2010.

Inside AWS, the metered model kept extending. Data warehousing came with Redshift in 2012, event-driven code without servers with Lambda in 2014, and hosted foundation models with Bedrock in 2023.

For the service models that S3 and EC2 belong to, see IaaS / PaaS / SaaS. The wider sequence is on the cloud computing timeline.