GPU instances for developers
Persistent RTX A6000, A100, L40, and H100 machines, ready to SSH into in minutes. Switch GPUs any time and snapshot your work for later.
~ $ tnr create --gpu a100 --num-gpus 1 --vcpus 8 --template base
✓ Instance created successfully!
Instance ID: 0
~ $ tnr status
ID Status Disk GPU vCPUs RAM Template
0 RUNNING 100GB 1xA100 80GB 8 64GB Base
~ $ tnr connect 0
Connecting to Thunder Instance
✓ SSH key management (412ms)
✓ Establishing SSH connection (1.284s)
✓ Connection established successfully
ubuntu@thunder:~$ nvidia-smi
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.133.20 Driver Version: 570.133.20 CUDA Version: 12.8 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA A100 80GB PCIe On | 00000000:00:05.0 Off | 0 |
| N/A 33C P0 61W / 300W | 1MiB / 81920MiB | 0% Default |
| | | Disabled |
+-----------------------------------------------------------------------------------------+
ubuntu@thunder:~$
Loved by developers
“This pricing is insane.”
“Can’t get enough of this UX.”
Real quotes from ML engineers building on Thunder Compute
Unmatched pricing
Low-cost GPUs for developers
Our on-demand pricing is up to 68% less than AWS
Save while developing
Prices low enough where you can develop on an A100
Scale easily
Change specs at any time with a few clicks
price perGPU-hour
instance 1 · switch GPU
End-to-end flexibility
One environment for development through production
Switch GPUs, add storage, or take a snapshot for later
Stay in VS Code
Connect to a persistent environment in one click
Change specs
One button to change the specs of your instance
Snapshots
Keep the machine
Snapshot the instance’s disk, delete the instance, and start a new one from it later. List caches in .thunderignore to keep snapshots small.
Keep working while it saves
Snapshots run in the background on a running instance
Everything on the disk comes back
OS, packages, code, weights and data. Running processes and memory are not saved.
$5.00/mo
$795.70/mo
$2,336/mo
Snapshot, 100 GB
A100 left running
H100 left running
Storage pricing
Pay for storage, not idle GPUs
Snapshots cost $0.05 per GB each month
No GPU charges
A deleted instance bills nothing while you're away
Billed on disk size
A 100 GB disk costs $5.00 a month to keep
Snapshot vs keep running
Know when to snapshot
Keep an instance for active work. Snapshot it when you step away.
| Snapshot and delete | Keep it running | |
|---|---|---|
| Billing | $0.05 per GB each month | GPU rate, billed per minute |
| GPU | Released | Held for you |
| What's kept | OS, packages, code and data | Disk plus running processes |
| Getting back | Restore, up to 8 min per 100 GB | Connect right away |
| Best for | Nights, weekends, shared setups | Active work and long training runs |
Templates
Launch pre-built workflows
Pre-built templates to launch your AI workflow in one click
Templates include Comfy UI, Ollama, Wan2.1, PyTorch, CUDA Kernels, Forge Neo.
Deploy now
Simple interfaces for any workflow
A VS Code extension, a CLI and an MCP server to manage your instances
~ $ tnr connect 0
⚡️ Connecting to Thunder Instance
✓ Establishing SSH connection (1.284s)
✓ Connection established successfully
ubuntu@thunder:~$ nvidia-smi -L
GPU 0: NVIDIA A100 80GB PCIe
ubuntu@thunder:~$
Drive it from your agent
Claude Code, Cursor or Codex create instances, run commands and take snapshots over MCP
MCP quickstart >Spin up an A100 and check the GPU
●
thunder-compute - create_instance (MCP)(gpu_type: "a100xl")
└
{
"instance_id": "0",
… +4 lines
●
thunder-compute - run_command (MCP)(instance_id: "0", command: "nvidia-smi -L")
└
{
"exit_code": 0,
… +9 lines
●
Instance 0 is up with GPU 0: NVIDIA A100 80GB PCIe.
Launch your first instance
A6000s from $0.35/hr, billed by the minute. Ready to SSH into in minutes.