◉ Compute Instances — GPU and CPU, under the leading cloud’s price

Know what it costs.
Then start it.

The hardest part of renting capacity is usually finding out what it will cost before you have already spent it. Here the hourly rate is on the screen next to the machine, and the estimate updates as you choose.

Billed by the hour. No commitment, no reservation, no minimum term.

Upfront
The rate, before you commit
Hourly
Billing, with no minimum term
Minutes
From chosen to answering
Yours
To stop the moment you are done

What runs on it

◆Model training and fine-tuning◆Batch inference at volume◆Rendering and simulation◆Data processing pipelines◆Build and test fleets◆Research workloads with a deadline◆Anything that needs more machine than you own◆Model training and fine-tuning◆Batch inference at volume◆Rendering and simulation◆Data processing pipelines◆Build and test fleets◆Research workloads with a deadline◆Anything that needs more machine than you own
How it is different

Four things that should
never have been hard.

Renting a machine is a simple transaction that most platforms make complicated. These are the four places it usually goes wrong.

◈

The price is visible first

Every machine shows its hourly rate beside it, and the estimate for your chosen duration updates as you pick. You are never finding out afterwards.

⟡

No reservation to negotiate

Start one for an hour. Start forty for an afternoon. There is no term to commit to and nothing to call anyone about.

⏻

Stopping actually stops

Terminate an instance and the meter stops with it. Not at the end of an hour you have already been committed to, and not at the end of the month.

≡

One bill with everything else

Capacity appears on the same invoice as your databases, deployments and domains, in the same currency, from the same company.

The console

What is running, what it costs, and the button that stops it.

The two questions anyone actually has about rented capacity are how much is running and how much has it cost. Both are on the first screen, not behind a report.

  • Everything running, in one list
    Across regions and machine types, with what each one is costing per hour right now.
  • Quota before you hit it
    What your account is allowed and what it is using, so a refused deployment is never a surprise.
  • Stop is one action
    No wizard, no confirmation chain, no waiting for a state transition to allow it.
  • Spend as it accrues
    The running total, updated as the hours accumulate rather than reconciled at the end of the month.
Running nowLive
gpu-a100-80 · us-east$2.41/hr
cpu-32 · eu-west$0.39/hr
Instances2 of 5
This month$184.20
Committed termNone
What changes

Capacity you can reason about.

Most cloud spend is not wasted on the wrong machine. It is wasted on machines nobody could see, running past the point anyone needed them, on a term somebody signed a year ago.

Capacity elsewhere
  • ✕A price you work out from a calculator
  • ✕Reserved terms negotiated in advance
  • ✕Instances nobody can find to stop
  • ✕Spend reconciled at month end
  • ✕A separate console per workload type
  • ✕A sales conversation before a GPU
Compute
  • ✓The hourly rate, next to the machine
  • ✓Hourly, with nothing to commit to
  • ✓One list of everything running
  • ✓Spend visible as it accrues
  • ✓CPU and GPU in the same place
  • ✓Sign up and start one
What it removes

The gap between deciding
and running.

Minutes
From chosen to answering
Subject to the capacity you selected.
Zero
Reserved terms to sign
Hourly, with no minimum.
One
List of everything running
Across every region and type.
Live
Spend, as it accrues
Not reconciled at month end.
The details that matter

Rented capacity, without the tenancy agreement.

The estimate is the one you are charged against

The figure shown while you choose is derived from the same rate the meter bills at. It is not an indicative number from a separate pricing page that drifted out of date six months ago.

GPU and CPU, one flow

The same three steps whichever you need. No separate product to learn.

Quota is enforced before you start

You are told what the account allows rather than finding out from a failure.

Regions are a choice, not a contract

Pick where it runs per instance. Nothing is pinned by an account-level decision.

Nothing provisions before you confirm

The machine, the duration and the estimate are all on the screen at the moment you commit, and nothing starts until you do.

“
Our research team stopped asking finance for permission, because they could see the number before they pressed the button. That changed how fast we try things.
Director of Machine Learning · Consumer technology group
Before you start one

What people ask.

How is it billed?+

By the hour, at the rate shown beside the machine when you chose it. There is no minimum term, no reservation and no commitment to negotiate.

What happens when I stop an instance?+

The meter stops with it. You are not held to the remainder of an hour, a day or a term you did not choose.

Do I need to talk to someone to get a GPU?+

No. Sign up, pick the machine, and start it. There is no sales conversation between you and capacity.

How do I know what I am spending?+

The console shows what is running, what each one costs per hour, and the running total as it accrues — not a figure reconciled at the end of the month.

Can I limit what my team can start?+

Yes. Each account has a quota covering instance count and total resources, enforced at the moment of deployment rather than discovered afterwards.

Is capacity on a free tier?+

No. Compute is billed from the first hour. The free allowances apply to databases, storage and repositories; capacity is paid for because it costs from the moment it starts.

Pick the machine.
See the number. Start.

Capacity by the hour, priced before you commit to it.

Hourly billing. Stop whenever you are done.