Neocloud, Explained: The Cloud That Sells Exactly One Thing

The clouds you have heard of, Amazon Web Services, Microsoft Azure, Google Cloud, are supermarkets. That is the whole idea of a hyperscaler: one place, every aisle. Databases, storage, video plumbing, virtual desktops, tools for problems you did not know were problems. Hundreds of services under one roof, with compliance paperwork and a loyalty program.

A neocloud looked at all those aisles and built none of them. It is the guy selling one incredible thing out of a garage. You do not go to him for groceries. You go because he has the thing, the thing is exceptional, and the entire operation exists in service of the thing.

In practice, the garage is a data center full of GPUs arranged specifically for AI work: chips clustered close together, fast connections between them, everything organized around one question, which is how quickly a very large model can train or run on this hardware. Rent by the hour, bring your own ambitions.

That focus is the pitch. The supermarket rents out GPUs too, but the supermarket also has to keep every other aisle stocked. The garage does one thing, all day, for exactly one kind of customer: people building AI.

Honesty section, as always.

First, "neocloud" is a label popularized largely by the companies wearing it and the analysts tracking them. The "neo" is doing marketing work. Underneath the prefix sits a plain business: buy hardware, rent hardware out, hope the demand outlasts the loan.

Second, the tidy supermarket-versus-garage picture is already smudging at the edges. The supermarkets rent out plenty of AI chips themselves, and the bigger garages keep bolting on extra services as they grow. Tech categories are snapshots, not species. This one is at its crispest right now, which is exactly why the word is suddenly everywhere.

Third, the genuinely open question: there are more than a hundred garages selling roughly the same incredible thing. What happens to them if the shortage eases, or when the chips they bought this year start looking slow next to whatever ships in three years? Nobody honestly knows, including them.

And fourth, the part I am deliberately not explaining today: filling a garage with some of the most expensive hardware on earth takes staggering amounts of money, and where that money comes from, and where it goes after, is one of the strangest stories in tech right now. That one gets its own post next week. Bring a whiteboard.