A cloud GPU is a powerful graphics chip you rent over the internet by the hour instead of buying one. Here's what that means, why it exists, and what it costs — in plain words.
This post contains affiliate links. If you buy through them, Aliteq may earn a commission — at no extra cost to you. Prices verified at publish time.
Share
A cloud GPU is a powerful graphics chip that lives in a data centre somewhere, which you rent over the internet by the hour instead of buying one for your own computer. That's the whole idea. If you've heard people talk about renting an "H100" or a "4090 in the cloud" and had no idea what they meant, this is the plain-English explanation — what it is, why it exists, and roughly what it costs.
What a cloud GPU actually is
A GPU (graphics processing unit) is the chip that does the heavy maths behind modern AI — running a chatbot, generating an image, training a model. Powerful ones cost thousands of dollars to buy. A cloud GPU is simply one of those chips sitting in a company's data centre that you rent access to remotely: you connect over the internet, run your work on it, and pay only for the time you use. Think of it like renting a car for a trip instead of buying one to leave parked most of the week — you get the capability exactly when you need it, without owning the thing.
Why rent instead of buy?
Three plain reasons. Cost: a top AI card is a four-figure purchase, plus power and cooling; renting is pennies to a couple of dollars an hour, and you pay nothing when it's off. Flexibility: you can grab a huge GPU for one big job and a tiny one for a small task, without owning either. No hassle: no building a PC, no drivers to babysit, no hardware going out of date on your shelf. The trade-off is that it's not yours — you're renting, and a very cheap "spot" rental can be interrupted (more on that below).
What a cloud GPU is good for — and not
Occasional or bursty work
Great for
Yes — pay only when on
Not ideal for
Always-on 24/7 (owning may be cheaper)
Trying big models cheaply
Great for
Yes — rent a big card for an hour
Not ideal for
—
No upfront money
Great for
Yes — nothing to buy
Not ideal for
—
Private/offline-only data
Great for
Depends on provider
Not ideal for
Highly sensitive local-only work
Great for
Not ideal for
Occasional or bursty work
Yes — pay only when on
Always-on 24/7 (owning may be cheaper)
Trying big models cheaply
Yes — rent a big card for an hour
—
No upfront money
Yes — nothing to buy
—
Private/offline-only data
Depends on provider
Highly sensitive local-only work
Referral link
See real cloud-GPU prices, cheapest first
The best way to understand it is to see the live rates — a capable card from about $0.12/hr, an H100 from ~$1.73/hr (Vast.ai, 22 Sep 2026). You only pay while it runs.
Referral link — we may earn a commission at no cost to you. Prices on our compare page are the provider's live figures, cheapest first; this never changes the ranking.
What does it cost?
Cloud GPUs are billed per hour (or per second on some providers), and prices depend on the card. As of 22 Sep 2026, an entry-level 24GB card rents from about $0.12/hour, and a data-centre H100 from around $1.73/hour. You'll see two pricing modes: on-demand (a bit pricier, but the GPU is reliably yours until you stop it) and spot (cheaper, but the provider can reclaim it mid-job if demand spikes — fine for interruptible work if you save often). For a beginner, on-demand is the simpler, safer starting point.
Quick answers
What is a cloud GPU in simple terms?
It's a powerful graphics chip in a data centre that you rent over the internet by the hour, instead of buying one for your own PC. You connect remotely, run your AI work on it, and pay only for the time you use — like renting a car instead of owning one.
How does renting a GPU work?
You make an account with a provider, choose a GPU, and launch an instance — a remote machine with that card. You run your work (a chatbot, image generation, training) on it, then shut it down. You're billed per hour or per second, only while it's running. Many providers offer ready-made templates so you don't have to set everything up from scratch.
Is renting a cloud GPU cheaper than buying one?
For occasional or exploratory use, yes, by a wide margin — you pay cents to a couple of dollars an hour instead of a four-figure purchase. Buying only pays off if you run a GPU heavily and constantly. The break-even is thousands of hours of use, so most people are cheaper off renting.
What can I actually do with a rented cloud GPU?
Run open AI models (chat, coding help), generate images and video, fine-tune a model on your own data, or serve an AI feature in an app. Anything that needs GPU horsepower but that you don't want to buy hardware for. You rent the size of card the job needs and stop paying when you're done.
That's a cloud GPU: rented graphics horsepower, by the hour, no purchase required. Your next step: see the live rates for every card, cheapest first, on our cloud-GPU compare page — it shows the current cheapest option across Vast.ai and Runpod and even the cheapest GPU that runs a given AI model.
