You pay for the seconds between launch and stop, prepaid one hour at a time. That is QuantaCloud's GPU cloud pricing in one sentence: every configuration has an hourly rate, the first hour is charged from your prepaid credit when you launch, and the unused seconds of the current hour come back when you stop. There is no subscription and no term to commit to.
Add credit and launchCurrent hourly prices#
Every configuration on sale right now is listed in the catalog above by GPU memory, with its price per instance-hour and per GPU-hour, its vCPUs, RAM and disk, and its region. Prices and availability move with capacity, and the deploy page shows the rate before you click Deploy.
48 GB per GPU
The RTX A6000 is an Ampere card. The RTX 6000 Ada, L40 and L40S are Ada Lovelace cards with FP8 support.
80 GB per GPU
The A100 80GB comes in SXM4 and PCIe versions, and the H100 is the PCIe card.
96 GB per GPU
The RTX PRO 6000 Blackwell is the only 96 GB card on offer, and the only one with FP4.
141 GB per GPU
The H200 NVL carries the most memory of any on-demand GPU: 141 GB of HBM3e.
The GPU catalog compares the cards side by side and matches jobs to memory tiers.
How billing works#
Billing follows six rules, in the order you meet them:
- You add prepaid credit by card on the Billing page. The minimum deposit is $5, the quick amounts run from $10 to $500, and one deposit can go up to $25,000. Auto top-up is optional.
- Launching an instance charges its first hour, and your balance must cover at least that hour.
- Each further hour is charged when the previous one is used up. The clock starts when you click Deploy, and the provisioning minutes count once the instance is running.
- When you stop, the unused seconds of the current hour are refunded to your balance.
- If you stop before the instance is running, or provisioning fails, the first hour is refunded in full.
- If your balance cannot cover the next hour, the instance is terminated and its disk deleted. QuantaCloud emails you when your balance falls below $2, at most once a day, and auto top-up refills it from a saved card when it drops below a threshold you set between $2 and $1,000.
The net effect is per-second billing with a one-hour prepay: a running instance never costs more than the time between launch and stop, and a launch that never reaches running costs nothing. The rules are the same for every GPU, and the template does not change the price. The billing docs walk through the Billing page itself (deposits, saved cards, auto top-up and the CSV export), and adding credits covers the first deposit.
What the hourly rate includes#
The hourly rate covers the whole VM: its GPUs, vCPUs, RAM and local disk. The disk size is fixed by the configuration and included in the rate, from 256 GB with one RTX A6000 to 5,000 GB with eight A100 SXM4 or L40S GPUs on 2026-09-27. There are no egress or ingress charges for data moving in or out.
What the rate does not buy is storage that outlives the instance. There are no volumes and no snapshots, and stopping an instance deletes its disk, so copy your results off before you stop.
Three worked examples#
All three are our calculations from the rules above, at the prices listed on 2026-09-27, with today's live price next to each.
An RTX A6000 stopped 3 hours 17 minutes after launch
One RTX A6000 cost $0.48 an hour on 2026-09-27 (today: $0.48/GPU-hr).
| Time after launch | What happens | Balance change |
|---|---|---|
| 0:00 | Launch: hour 1 charged | -$0.48 |
| 1:00 | Hour 2 charged | -$0.48 |
| 2:00 | Hour 3 charged | -$0.48 |
| 3:00 | Hour 4 charged | -$0.48 |
| 3:17 | Stop: the 43 unused minutes of hour 4 refunded (43/60 x $0.48) | +$0.344 |
| Net | 3.283 hours x $0.48 | $1.576 |
You are charged $1.92 over the run and get $0.344 back, so the session costs $1.576, about $1.58. That is exactly 3 hours 17 minutes of GPU time at $0.48 an hour.
A launch you stop before it is running
One H200 NVL cost $3.43 an hour on 2026-09-27 (today: the console price). Launch it and $3.43 is charged at once. Stop it while it is still provisioning and all $3.43 comes back, so the net cost is $0.00. A launch whose provisioning fails is refunded the same way.
Eight L40S GPUs for 60 hours
An 8x L40S VM cost $8.72 an hour on 2026-09-27, which is $1.09 per GPU-hour (today: $1.09/GPU-hr). Run it for exactly 60 hours and you pay 60 hourly charges: 60 x $8.72 = $523.20, for 480 GPU-hours.
The part that needs planning is the balance, not the rate. Each hour is charged as the previous one ends, so the run needs $523.20 of credit over its life, or auto top-up. I would fund the full run plus one hour before starting a job like this, $531.92 here ($523.20 + $8.72), because one missed hourly charge terminates the VM and deletes its disk. If a job this size comes back every week, it is also the point where a reserved quote is worth asking for.
Reserved or committed capacity#
Reserved capacity is quoted per brief rather than listed here. We order and build the hardware to your spec, any NVIDIA GPU from a single server to an InfiniBand cluster, and confirm the configuration, lead time and terms in writing. The reserved capacity page has the brief form. Get a reserved quote.
Frequently asked questions#
Is billing per second or per hour?
Both. Charges are made an hour ahead and settled by the second when you stop, so a run of 3 hours 17 minutes costs 3 hours 17 minutes.
Is there a free trial or sign-up credit?
There is no free trial and no sign-up credit: new accounts start with a balance of $0.
Do you issue invoices?
Not today. Each deposit sends a payment receipt by email, and the Billing page exports your transactions as a CSV file.
Is there a minimum spend?
The smallest deposit is $5, and your balance must cover one hour of the configuration you launch: $0.48 for a single RTX A6000 on 2026-09-27 (today: $0.48/GPU-hr).
What happens if my balance runs out mid-job?
The instance is terminated at the first hourly charge your balance cannot pay, and its disk is deleted. There is no grace period. Auto top-up prevents it as long as a saved card works, and it switches itself off after three failed charges.
Do prices change?
Yes. Prices and availability move with capacity, so check the rate on the deploy page before each launch. The tables on this page are live.
My rule for choosing between hourly and reserved: if a job runs for hours or days and then ends, pay by the hour, because you pay only from launch to stop. If you would keep the same GPUs busy most hours of most weeks, send a brief and set the quote against the arithmetic above. If you are new to renting, the walkthrough goes from sign-up to a running GPU.
Add credit and launch