Clouds

One app across every cloud you use.

Connect the GPU clouds you already have accounts with, list them in the order you prefer, and deploy once. Each replica starts on the first cloud that has room, and each cloud bills you directly.

Request access Read the docs

  priority  cloud          gpu       state
  ────────  ─────────────  ────────  ───────
     1      salad          rtx4090   full
     2      runpod         rtx4090   placing
     3      vast.ai        rtx4090   ready
     4      your machine   rtx3090   ready

Live

Placing real machines today.

These connectors have started and stopped real machines on real accounts.

Salad

Run inference and jobs on SaladCloud's low-cost GPUs, from your own account.

RunPod

Use your RunPod account as serverless GPU capacity.

Vast.ai

Turn Vast.ai's GPU marketplace into serverless capacity.

DigitalOcean

Run GPU Droplets from your DigitalOcean account, with scale to zero.

Yotta

Use Yotta GPUs, from RTX 4090 to H200 and B200, as serverless capacity.

Example hourly prices from each cloud, with no Surfra markup. Salad is the low end of its published band. RunPod, Yotta, and DigitalOcean are list prices. Vast.ai is the cheapest on-demand offer on 16 Oct 2026, and that market moves. Your account's stock and region can differ.

More clouds

Connect these too.

You can connect these now.

AWS
Oracle Cloud
Lambda
Crusoe
Hetzner
Shadeform
Verda
Hyperstack
Vultr
Scaleway
Novita
Paperspace
TensorDock

Your hardware

And the machines you own.

Use your own GPU

How placement works

Failover is the default.

Every app has a priority list. Each row is a cloud and a GPU, such as Salad RTX 4090 or RunPod H100. When a new replica is needed, Surfra tries the rows in order and starts it on the first one with stock.

Per-provider limits let you keep a cheap base on one cloud and push spikes to another. Rows can also be spot machines where the provider offers them.

The cloud key belongs to your team and lives in Surfra encrypted. Read how in Security.

FAQ

Common questions

Which clouds does Surfra support?

Salad, RunPod, Vast.ai, DigitalOcean, and Yotta today, plus AWS, Oracle Cloud, Lambda, Crusoe, Hetzner, Shadeform, Verda, Hyperstack, Vultr, Scaleway, Novita, Paperspace, and TensorDock, and machines you register yourself.

Do I need accounts with every cloud?

No. Connect the ones you want. Placement only offers clouds your team has connected.

Who pays the cloud bill?

You do, directly to each cloud, at its own price. Surfra never resells compute.

Run it on your own clouds.

Surfra is invite only while we test it. Tell us what you want to run and we will send an invite.

Request access