Also: cloud infrastructure, cloud computing (hosting), public cloud
Cloud hosting is the model in which servers, storage and databases are created on demand through an API from a provider's pooled hardware and billed for the time they exist, rather than rented as a fixed box by the month.
Our take. Elasticity is the feature almost everybody buys and most websites never use. If your traffic chart is a flat line with a weekly wobble in it, a flat-priced instance from a smaller provider is the better default, and the moment to adopt usage billing is when a measured spike forces the question rather than when somebody anticipates one.
Servers you create with a command
The hardware is not the interesting part. A cloud instance is, underneath, a virtual machine built on the same technology as a VPS, running on a computer in a building you will never visit. What changed is everything wrapped around it. Renting a server used to involve a sales conversation, a commitment and a wait. On DigitalOcean, Hetzner Cloud, AWS or Azure it involves one API call, and the machine is running before you have finished reading the confirmation. That single property cascades outwards. If a program can create a server, then a build pipeline can create one, a configuration file kept in version control can describe a whole environment, and the environment can be destroyed at night and rebuilt each morning from that description. Terraform, Pulumi and their relatives exist because infrastructure turned into something you declare instead of something you order. The same property produces the model's signature failure. An undersized fixed server gets slow, which is irritating but has a floor. A misconfigured cloud account gets expensive, and expense has no floor: an autoscaling group with a broken health check will replace instances until somebody notices, a forgotten load balancer bills by the hour for doing nothing, and outbound traffic is metered per gigabyte on the large platforms whether it came from a product launch or from a bot working through your image library. Keeping that under control is an operational discipline, not something finance handles afterwards.
The hyperscalers (AWS, Google Cloud, Microsoft Azure) sell hundreds of services across regions on every continent, with the compliance paperwork large organisations are obliged to ask for, and a pricing structure complicated enough that each of the three publishes its own calculator to help you estimate a bill in advance.
Developer clouds (DigitalOcean, Hetzner Cloud, Vultr, Linode) offer a deliberately short catalogue: instances, block storage, object storage, managed Postgres or MySQL, a load balancer. Prices are flat monthly figures with a traffic allowance included, which removes the metered egress that catches people out elsewhere.
Managed layers on top hide the machine entirely. Application platforms deploy straight from a repository and operate the runtime on your behalf, the PaaS end of this same market, while traditional hosts increasingly run their managed hosting tiers on cloud infrastructure without ever mentioning it.
What stays yours in the raw model is more than newcomers expect: the operating system inside the instance, its patches, its firewall rules, its backups and its monitoring. Buying cloud capacity is buying provisioning speed, not an administrator.
In practice
Two lines in AWS documentation explain most surprise invoices. A NAT gateway, the component that lets private instances reach the internet, is charged for every hour it exists plus a fee for the data it processes, so an experiment abandoned months ago keeps quietly accruing. An Elastic IP address is charged when it is not attached to a running instance, which means shutting a server down to save money starts a different meter. Neither is hidden and neither is a trick; both are published on the vendor's own pricing pages. They simply reward the habit of reading what a resource costs while it is idle, not only what it costs while it works.
The machine is the same thing. What differs is the commercial wrapper: a virtual private server is sold by the month at one fixed size, while a cloud instance is conjured by software and billed only while it exists.
The layer name drawn in architecture diagrams is infrastructure as a service, whereas cloud hosting is the product people actually sign up for, and it usually bundles rather more than that layer strictly contains.
Key takeaways
→The shift is commercial rather than technical: infrastructure became something a program creates and deletes, which is what made it possible to keep environments in version control.
→The failure mode moves from too slow to too expensive, and the expensive version has no natural limit.
→Provisioning speed is what you buy; patching, backups and monitoring inside the instance remain your job unless you pay somebody for those separately.
A rotating selection from the course directory, drawn from the subcategories where this concept is taught rather than picked for it. Details, price and the provider link are on the course page.
Are you building, deploying, or managing AI agents and want to ensure they operate at peak performance? Monitoring and…
Udemy
FAQ
Is it better than a traditional server?
It is better at change. Variable demand, growth nobody can forecast and environments rebuilt from code all favour it. A steady workload on a correctly sized fixed machine is frequently cheaper and always simpler. Ask how much your capacity requirement actually fluctuates before asking which model sounds more modern.
Do I need one of the big three?
For a website, a shop or a typical application, a smaller provider's short catalogue covers everything with far less pricing surface area. The large platforms earn their complexity when you need something specific from them: regions in particular jurisdictions, large-scale data services, or certifications your customers insist on seeing.
Sources
The primary text this definition rests on. Read it before you trust ours.
