Your proposal says Google Cloud. Here's what you'd be renting, and who runs it.

Rented computing, sold by the project

Underneath the product names, a cloud is a rental arrangement. Google rents you slices of its data centres: a machine by the second, storage by the gigabyte, a database somebody else patches. The charge stops when the thing stops.

It's the same category of product as AWS and Microsoft Azure. All three bill for use rather than for a licence, and mostly sell the same ideas under different names. By the usual reporting AWS is largest, Azure second, Google Cloud third, which tells you how many people you can hire and little else.

The organising idea worth knowing is the project. Every machine, database and storage bucket belongs to exactly one, billing attaches to one, and projects are what change hands at handover. So ask early whose organisation yours will sit in. It's a two-minute question at the start of a build and a lawyer's question at the end.

Google built it for itself first, and that shows

Google didn't set out to sell computing. It built infrastructure to index the web and serve video to much of the planet, and the problems it solved were about data and scale rather than about hosting somebody's website. The commercial cloud came out of that machinery later.

Two internal systems have public descendants at the centre of the platform. Borg, which decided what ran on which machine, is the ancestor of Kubernetes, open-sourced in 2014 and later handed to the Cloud Native Computing Foundation. Dremel, for querying enormous datasets quickly, is the ancestor of BigQuery.

That's why the strengths cluster where they do. A requirement that rhymes with something Google solved for itself lands on well-worn ground. A small business website doesn't.

Three places it gets chosen on merit

Set aside the companies that land here because they already pay Google for email. Three real reasons come up.

Data and analytics. BigQuery is a warehouse you load data into and ask ordinary SQL questions of, with no cluster to size and nothing idle between questions. You pay for what's stored and for the work each query does, so an analyst can start without a platform team behind them.

Machine learning. Google designs its own chips for this work and rents them out, and it sells access to its own models alongside open ones through Vertex AI. If the centre of gravity of the project is a model rather than an application, that's a reason rather than a preference.

Kubernetes. It came out of Google and it's the standard way to run containerised software across many machines. Every large cloud runs it well now, so that edge has narrowed, and it stays a big thing to operate. Cloud Run runs the same container without any of it.

Metered pricing, and where the surprise lives

Everything is counted. A machine bills by the second once it's on, storage bills for what's kept, and the line that catches people out is data leaving Google's network for the internet, charged by the gigabyte while data arriving isn't. Video has a different cost shape from forms.

Some discounts arrive without a negotiation. On some machine types, running an instance most of the month lowers the rate automatically, and committing for a year or three lowers it further. Published rates change, so price your design with Google's free calculator rather than a number in an article, this one included.

The comparison worth drawing isn't Google against another cloud. It's metered computing against a flat monthly platform, because the two fail in opposite directions.

Where it stops

It's the wrong shelf for a website. A marketing site with a few forms behind it can run here, and it will cost more attention than it returns. Managed hosting gets it live sooner with nothing to operate.

The hiring pool is smaller, which is the practical edge of that ordering. Plenty of firms know Google Cloud well, and in most markets more people know AWS. You feel that when you want three quotes for one job.

The worry about Google retiring things is half fair. The consumer products it has closed aren't the cloud business, and Google publishes a deprecation policy covering a named set of cloud services with long notice before they change. The narrower question: is anything in your design still labelled preview, because preview carries no such promise on any cloud.

The lock-in is architectural, not contractual. Google said in 2024 it would stop charging for the data transfer involved in leaving, and the other large clouds announced much the same. What doesn't move is the design. It needn't be all or nothing, though: companies use BigQuery on its own while the application runs elsewhere.

Who has to be reachable when it breaks

This decides more of these builds than anything technical, and it isn't Google's part of the problem.

A cloud account is a live system with permissions, budgets, alerts and updates attached. Somebody has to hold the top of that tree, notice the morning something costs four times what it did, and be reachable when a service has a bad day. Managed services shrink that job, never to zero.

For most businesses it isn't a hire. It's one named person and an arrangement, agreed before launch rather than discovered after it.