Thunder Compute

Thunder Compute is a cloud GPU provider that offers persistent GPU instances for developers and sandboxes for AI agents. Instances are machines a user connects to and keeps, with a command line tool for creating and connecting, snapshots that save the disk between runs, and a VS Code extension. Sandboxes are isolated environments, with or without a GPU, designed for reinforcement learning environments, evals and post training, and can fan out hundreds of parallel rollouts from a few lines of Python. The service is aimed at machine learning engineers, researchers and startups that want on demand GPUs without long contracts. Available instance GPUs are the RTX A6000, L40, A100 80GB and H100 SXM, in configurations from one to eight GPUs. Instance pricing is billed per minute: RTX A6000 at $0.35 per hour, L40 at $0.79, A100 80GB at $1.09 and H100 SXM at $3.20 per GPU hour, with extra vCPUs at $0.04 per hour. The first 100 GB of disk per GPU is included, snapshots cost $0.05 per GB per month, and there are no data egress fees. Sandboxes are billed per second by resource, including $1.95 per GPU hour for A100 80GB, $2.95 for H100 SXM, $0.051 per vCPU hour and $0.016 per GiB hour of memory, and access is by request. Reserved clusters of 8 to 512 or more GPUs are available on a contract basis through sales.