What is cloud computing?
Cloud computing is the rental of computing resources over the internet from a provider that owns and operates large data centers. Instead of buying servers, an organization creates virtual machines, storage buckets, databases, and GPU instances through a web console or API, uses them for minutes or years, and pays for what it consumed. The provider handles buildings, power, hardware replacement, and the physical network.
Cloud services come in layers. Infrastructure as a service (IaaS) provides raw virtual machines, storage, and networking. Platform as a service (PaaS) adds managed databases, container platforms, and message queues so teams do not administer the operating system. Software as a service (SaaS) delivers a finished application, such as a CRM, through a browser. Cloud GPUs are a form of IaaS, and hosted AI models are sold as a service reached through an API.
Why cloud computing matters
For AI work, the cloud matters because it offers hardware that is hard to buy and expensive to leave idle. A team can rent eight GPUs for a two-week training job and return them. It can test three model sizes in an afternoon without a purchase order. For steady production workloads the calculation changes. A GPU that runs every hour of every month costs more to rent over several years than to own, and data that must stay in a particular country or building may not be allowed in a shared cloud region at all. Most organizations end up with a mix, which is why hybrid deployment and multi-cloud are common patterns.
Cloud computing in practice
A wholesale distributor with twelve warehouses moves its reporting to the cloud. Nightly sales and inventory extracts land in cloud object storage, a managed data warehouse builds the tables, and a dashboard service presents them to branch managers each morning. When the planning team later wants a demand forecasting model, it trains on the same data with a cloud GPU instance rented for a weekend, then runs the daily forecast on a small virtual machine. The distributor never buys a server, and the finance team sees the cost of each workload on the monthly bill.
What to watch for
- Track where each dataset is stored and processed, because cloud regions map to legal jurisdictions and data residency rules.
- Watch egress fees, since moving large datasets out of a cloud often costs more than storing them there.
- Set budgets and alerts before launching GPU instances, which bill by the hour whether or not anything is running on them.
- Read the shared responsibility model carefully, because the provider secures the data center and the customer secures everything it configures inside it.