In 2006 Amazon launched S3 for storage and EC2 for computing, letting anyone rent capacity by the hour with a credit card. The idea that infrastructure could be provisioned by code rather than by ordering and racking machines spread quickly. Startups no longer needed capital to buy servers, and large companies began moving workloads out of their own data centres.
At the same time, content delivery networks matured. Cloudflare was founded in 2009 and eventually made CDN and DDoS protection available at a free tier, which changed what small sites could expect.
It is easy to read that as a tidy story of progress. From the support desk it looked messier: customers who had always bought a shared hosting plan suddenly had a third option, and a fair number of them chose it for the wrong reasons. This chapter covers what actually changed in those six or so years, and what stayed exactly as it was.
Before: servers were things you bought
To see why 2006 mattered, remember the process it replaced. If you needed a dedicated server, you picked a model, signed a contract of a month or more, waited for the provider to install an operating system, and received a root password by email. A day or two was quick. If you needed ten servers for a launch, you ordered ten, paid for ten, and kept paying for ten after the launch had faded.
Shared hosting sat at the cheap end of that world. Hundreds of sites lived on one machine, managed through a control panel, and you never saw the server at all. The trade was simple: low cost, little control, and a neighbour who could slow you down.
What the new model actually offered
S3 arrived in March 2006 and EC2 followed later that year, first as a limited beta. The pitch was three things at once. You paid for what you used, by the hour. You could create and destroy machines through an API, which meant scripts could do the work that used to need a person with a screwdriver. And the machines were virtual, several customers sharing one physical host, separated by a hypervisor such as Xen, and later KVM.
The VPS market had already sold that idea. What was new was the speed and the billing. A virtual private server in 2005 was often a monthly plan with a ticket queue. A cloud instance in 2007 was a function call.
Other providers followed the pattern from different angles. Heroku, from 2007, let developers push code and skip the server entirely. Google App Engine appeared in 2008 with a similar promise, and Microsoft's Azure became generally available in 2010.
A rough timeline
None of those dates is the moment everyone switched. Adoption was slow at first. Early on, cloud instances had no persistent local disk by default, so a stopped machine could lose its data, and a lot of sysadmins reasonably saw that as a toy. It took a few years, and a few well-publicised outages, for people to learn the habits that cloud hosting demands: keep state somewhere durable, expect machines to disappear, and have more than one of everything.
CDNs and the small site
Content delivery networks existed before Cloudflare, and large publishers had used them for years. The change from 2009 onward was who could afford one. A CDN keeps copies of your static files, such as images, stylesheets and scripts, on servers close to your visitors, so a reader in Sofia fetches a picture from nearby rather than from a machine in Texas.
Putting a CDN in front of a site works by changing where the domain's DNS records point. Visitors connect to the CDN, which answers from its cache where it can and fetches from your origin server where it cannot. The same position in the path lets it absorb floods of junk traffic before they reach you. Offering that on a free plan meant a hobby blog could have protection that once needed an enterprise contract.
What did not change
Hosting still came down to the same questions: where is the code running, who patches it, how do you get your data back out, and what happens when it breaks at three in the morning. Cloud providers took over the hardware layer, but responsibility for the application, the updates and the backups stayed with the customer. This is the part that surprised newcomers, who assumed that renting from a very large company meant someone else was watching their site.
Shared hosting did not vanish either. For a site with a few hundred visitors a day, a managed shared plan remained, and remains, a perfectly good answer: cheaper, simpler, and patched by people whose job it is. The cloud made it normal to treat servers as disposable and to buy exactly as much as you needed.
The glossary covers terms like VPS and hypervisor if any of this was new.