The term GPU cloud aggregator covers at least four different kinds of service, and the question that separates them is who holds the account at the cloud that runs your machine. This is how I sort the market as of 1 October 2026. It describes kinds of service, not named companies, so check any service against the questions below before you rely on it.
Four kinds of service share the name
| Kind | What you do there | Ask before you pay |
|---|---|---|
| Price tracker or comparison site | Read prices and stock from many clouds, then click through | How old is the price, and does the list show stock |
| Peer-to-peer marketplace | Rent machines that individual hosts list on a shared platform | Who the host is, and what happens when a machine goes offline |
| Deploy-through aggregator | Launch machines at several clouds from one account, one balance and one API | Who is the customer at the cloud, what the fee is, and what happens when a cloud loses a machine |
| Broker or reseller | Ask for capacity or a quote and sign terms | The term, the price per GPU-hour and what is guaranteed in writing |
A price tracker is the only kind that never rents you anything. A comparison page is only as good as its timestamp, so check when a price was last seen. You can compare all providers on GpuPerHour before you launch anywhere. The other three kinds rent you a machine, and they differ in who stands between you and the hardware.
Who holds the provider account decides the trade-offs
A deploy-through aggregator holds the account at each cloud and you hold an account with the aggregator. Four things follow from that. You pay the aggregator, not the cloud. You act through the aggregator's dashboard and API, so the cloud's own console and extra features reach you only where the aggregator passes them on. Your first line of support is the aggregator, because it is the cloud's customer. And any credits or commitments you already hold at a cloud cannot be spent there, because you never use your own account.
What you get in return is the appeal of any multi-cloud GPU marketplace: one signup, one prepaid balance and one API instead of several. DeployGPU's API launches VMs on four providers with one key. Its offers come from the same scrapers that feed GpuPerHour, and every provider DeployGPU launches on is checked every minute.
What it takes away
The catalogue is a subset of each cloud's, and so are the features. On DeployGPU, stopping an instance deletes it and its disk, with no pause and no snapshot. Volumes exist only on Verda and Hyperstack. There is no spot pricing, no reserved pricing you can buy yourself, no invoices (receipts and a CSV export instead) and no SLA. Nothing fails over to another provider when a launch fails or a provider loses a machine.
There is also a fee. Offer prices on DeployGPU include any platform fee, and the deploy page shows the full breakdown before you pay. Card top-ups carry a processing fee of 5% + $0.30. The pricing page has the mechanics.
Where DeployGPU fits and who runs it
DeployGPU is a deploy-through aggregator for four VM providers: QuantaCloud, Massed Compute, Verda and Hyperstack. It also lists Vast.ai containers in the dashboard, labelled Testing Only. All provider accounts belong to DeployGPU. You never connect your own cloud account, and there is no bring-your-own-account option.
Quanta Cloud LLC runs DeployGPU and also runs QuantaCloud, so one of the clouds in the list belongs to the same company as the marketplace. It runs GpuPerHour too. The about page says the same, and the providers page shows how the five differ. This is the catalogue as it stands now, and the GPUs page has a page for each family:
| GPU | Memory | Offers now | GPUs per VM | Providers | From (per GPU-hr) | Details |
|---|---|---|---|---|---|---|
| B200 | 192 GB | 1 | 8x | Verda | $8.03 | Offers |
| H200 | 141 GB | 3 | 1x, 2x, 4x | Verda | $5.44 | Offers |
| H100 | 80 / 94 GB | 8 | 1x, 2x, 4x | Hyperstack, Massed Compute, QuantaCloud | $2.50 | Offers |
| A100 | 80 GB | 13 | 1x, 8x | Hyperstack, Massed Compute, QuantaCloud, Verda | $1.35 | Offers |
| RTX PRO 6000 Blackwell | 96 GB | 5 | 1x | Massed Compute, QuantaCloud, Verda | $2.19 | Offers |
| L40S | 48 GB | 8 | 1x, 2x, 4x | Massed Compute, QuantaCloud | $0.97 | Offers |
| L40 | 48 GB | 13 | 1x, 2x, 4x | Hyperstack, Massed Compute, QuantaCloud | $0.86 | Offers |
| RTX 6000 Ada | 48 GB | 14 | 1x, 2x, 4x, 8x | Massed Compute, QuantaCloud | $0.78 | Offers |
| RTX A6000 | 48 GB | 46 | 1x, 2x, 4x, 8x | Massed Compute, QuantaCloud | $0.48 | Offers |
Offers come in fixed sizes of 1, 2, 4 or 8 GPUs per VM. More than 8 GPUs, or a fixed term, goes through the cluster quote form, which a person answers. That part works like a broker, not like an aggregator.
Five questions to ask any aggregator
I ask these of every service in this category, including DeployGPU, and the right-hand column is DeployGPU's answer, checked 1 October 2026.
| Question | DeployGPU's answer |
|---|---|
| Who holds the account at the cloud? | DeployGPU. There is no bring-your-own-account |
| What is the fee, and where is it? | Inside the offer price, with a breakdown on the deploy page. A card processing fee applies to top-ups |
| What happens when a cloud loses the machine? | A health check every 5 minutes ends the deployment and refunds unused time, with no email and no retry on another provider |
| What survives a stop? | Nothing on the instance disk. Volumes survive on Verda and Hyperstack only |
| What records and guarantees do I get? | Receipts and a CSV export, no invoices, no SLA |
When to use one and when to go direct
| Your situation | Better fit |
|---|---|
| You are still comparing prices | A price tracker |
| Hours to days of on-demand GPU, with no cloud account yet | A deploy-through aggregator |
| You hold credits, a commitment or a private deal at a cloud | Go direct, because an aggregator cannot spend them |
| You need a cloud's own features, such as snapshots or pause | Go direct, or check what the aggregator exposes |
| You need invoices or an SLA | Read the terms first. DeployGPU offers neither |
| More than 8 GPUs, several nodes or a fixed term | A broker or a quote |
My rule is to use an aggregator for short, on-demand work and to go direct once the work is long enough that a cloud's own features and contract terms matter more than one login.
Frequently asked questions
What is a GPU cloud aggregator?
A service that sells machines from several GPU clouds through one account. The kind that deploys for you holds the accounts at each cloud and bills you itself.
Is DeployGPU a GPU cloud aggregator?
Yes, of the deploy-through kind. It launches VMs on QuantaCloud, Massed Compute, Verda and Hyperstack, and lists Vast.ai containers in the dashboard, labelled Testing Only.
Can I use my own cloud account with DeployGPU?
No. All provider accounts belong to DeployGPU, and there is no bring-your-own-account option.
Is a price comparison site the same as an aggregator?
No. A comparison site lists prices and sends you to the provider. It does not launch machines or hold accounts.
Who runs DeployGPU?
Quanta Cloud LLC, which also runs the QuantaCloud GPU cloud that is one of the providers in the list.