Now available: B200 and H200 on demand, billed per second. See pricing →

GPU infrastructure that
developers love

Run training, inference, batch jobs and sandboxes on T4 to B200 with sub-minute starts, per-second billing, and a developer experience that feels local.

Get StartedRead the docs
terminal
$ ts run train.py --gpu h100 --image pytorch ✓ j-3f9a1c2b7e submitted to app 'train' on NVIDIA H100 80GB NVIDIA H100 80GB HBM3 epoch 1/10 loss 2.314 4.1 it/s epoch 2/10 loss 1.907 4.2 it/s ✓ exit 0 · $0.0412 $ ts sandbox create --gpu a100-80 --volume datasets Creating sandbox m-81de342e82 (NVIDIA A100 80GB) ... ✓ running · ssh: ts sandbox ssh m-81de342e82

Everything you need to ship on GPUs.

A single console and CLI for interactive sandboxes, one-shot runs, persistent volumes and secrets, organised into apps, environments and workspaces.

▶

Runs

ts run script.py --gpu h100. Your file is shipped, the GPU is allocated, logs stream back, and you get an exit code and a cost.

⌘

Sandboxes

Isolated GPU containers with root SSH, CUDA and PyTorch images, idle timeouts and per-host endpoints.

▤

Volumes

Persistent storage mounted at /mnt/name across runs and sandboxes. Browse and upload from the console or CLI.

◈

Secrets

Encrypted environment variables with templates for Hugging Face, AWS, W&B and more. Injected at launch.

Every GPU, on demand.

No reservations, no minimums. Rates include the default vCPU and memory shape.

available
NVIDIA T4
Turing · 16 GB VRAM · up to 8× per host
$1.51 / hour
4 vCPU16 GB RAM$0.000164/s
available
NVIDIA L4
Ada · 24 GB VRAM · up to 8× per host
$1.72 / hour
4 vCPU16 GB RAM$0.000222/s
available
NVIDIA A10G
Ampere · 24 GB VRAM · up to 8× per host
$2.02 / hour
4 vCPU16 GB RAM$0.000306/s
available
NVIDIA L40S
Ada · 48 GB VRAM · up to 8× per host
$3.80 / hour
8 vCPU32 GB RAM$0.000542/s
available
NVIDIA A100 40GB
Ampere · 40 GB VRAM · up to 8× per host
$3.95 / hour
8 vCPU32 GB RAM$0.000583/s
available
NVIDIA A100 80GB
Ampere · 80 GB VRAM · up to 8× per host
$4.73 / hour
8 vCPU48 GB RAM$0.000694/s
available
NVIDIA H100 80GB
Hopper · 80 GB VRAM · up to 8× per host
$7.11 / hour
12 vCPU64 GB RAM$0.001097/s
available
NVIDIA H200 141GB
Hopper · 141 GB VRAM · up to 8× per host
$8.46 / hour
12 vCPU96 GB RAM$0.001261/s
available
NVIDIA B200 180GB
Blackwell · 180 GB VRAM · up to 8× per host
$11.48 / hour
16 vCPU128 GB RAM$0.001736/s

Pay per second. Nothing else.

Prepaid credits in USDT, USDC, BTC, ETH or SOL. Live metering, spend limits, monthly statements, and withdrawals of unused purchased credits.

$

Crypto invoicing

Monthly invoices paid to a unique deposit address. Confirmed on-chain, marked paid automatically.

◔

Live metering

Usage posts every 20 seconds. Burn rate and hours remaining at a glance.

⚑

Spend limits

Monthly limits, pay-as-you-go allowances and alerts. Resources stop before you overrun.

▥

Workspaces

Invite your team with owner, manager and member roles. Separate environments per stage.

Regions

Pick a region or let the scheduler place your workload where capacity is best.

🌐 Auto (lowest latency)🇺🇸 US East🇺🇸 US West🇺🇸 US Central🇪🇺 EU West🇪🇺 EU Central🇯🇵 Asia Pacific Northeast🇸🇬 Asia Pacific Southeast