Log inGet started

GPUs for agents

GPU instances and sandboxes for builders on the AI frontier

Get started
Products

Start from your terminal or from your agent

Instances for hands-on development. Sandboxes for RL environments and evals at scale. Both on the same GPUs.

Instances

A GPU instance you connect to and keep. Snapshot it when you are done.

~ zsh
$ tnr create --gpu a100
✓ Instance 0 created
$ tnr connect 0
✓ Connected to instance 0
ubuntu@instance-0:~$ nvidia-smi -L
GPU 0: NVIDIA A100 80GB PCIe
A100 80GB from $1.09/hrExplore instances →

Sandboxes

Sandboxes for RL environments and evals, with or without a GPU. Fan out hundreds of parallel rollouts from a few lines of Python.

agent.py
import thunder_sandbox as thunder
sandbox = thunder.Sandbox.create(
    cpu=4, memory=32, storage=50,
    gpu_type=thunder.GPUType.H100, gpu_count=1,
    timeout=900)
sandbox.wait_until_ready()
process = sandbox.exec("nvidia-smi")
From $1.95/GPU-hrExplore sandboxes →
Pricing

Stop wasting runway on GPUs

Switching to Thunder Compute takes minutes and saves months of runway.

Get startedSee pricing
price perGPU-hour
  • Thunder Compute$3.20
  • Lambda$3.29
  • Runpod$3.49
  • Paperspace$5.95
  • CoreWeave$6.16
  • Amazon Web Services$6.88
  • Azure$6.98
  • Oracle Cloud$10.00
  • Google Cloud$10.98
On-demand list prices, checked Sep 30, 2026
LOVED BY DEVELOPERS

“Can't get enough of this UX.”
“We are saving so much on GPU Sandboxes.”

Real quotes from ML engineers and RL researchers building on Thunder Compute

I ship faster and our CEO loves the savings. So far, I’ve convinced nine of our MLEs to switch.

ML Engineer

Series B startup

The amount of value is insane. We are bootstrapped and run 2-3 finetuning jobs around the clock.

Co-founder

Data Science Consultancy

The VS Code extension helps me iterate faster. Anything that makes paper deadlines less stressful is a win.

PhD in Machine Learning

U.C. Berkeley

The snapshot flow means I can pause experiments without thinking about cloud housekeeping all day.

Applied AI Lead

Healthcare AI startup

Moving from queueing jobs to spinning up GPUs on demand changed how quickly our team tests ideas.

Research Engineer

Independent AI lab

The pricing finally lets us run bigger sweeps without turning every experiment into a budget meeting.

ML Platform Engineer

Seed-stage robotics company

Research

From the lab

Research and engineering from the team building Thunder Compute.

View all writing

Thunder Compute Raises $13 Million Series A Round to Put Idle GPUs Back to Work

August 19, 2026

How Thunder Compute works (GPU-over-TCP)

June 24, 2026

GPU Virtualization: Approaches and Tradeoffs

June 24, 2026

Get started today

Cloud GPU instances and sandboxes for AI workloads. Switch now and save up to 68% vs AWS.

Instances

A GPU instance you connect to and keep, from VS Code, the CLI or your browser.

From $0.35/hrBilled by the minute
Sign up now

Sandboxes

For RL environments, evals and agents. Fan out hundreds in parallel from a few lines of Python.

From $1.95/GPU-hrBilled per second
Get access