Nat-zero

Provides on-demand ephemeral GPU and CPU virtual machines that serve as drop-in runners for GitHub Actions, solving slow queues, limited resources, and maintenance burden of self-hosted or default runners for model inference, Docker builds, and production-like tests. Serves developers, DevOps engineers, data scientists and ML practitioners in startups, research institutions and enterprises across biotech, automotive, gaming and media. Positioned as a middle option between GitHub-hosted and self-hosted runners with one-line YAML change, pre-installed CUDA stack and per-minute billing.

Key features

• Ephemeral VMs per job
• Cold starts under 60 seconds
• CUDA pre-installed stack
• NVIDIA drivers and cuDNN included
• Container Toolkit pre-installed
• Bundled ingress and egress
• One-line YAML integration
• Rollover billing credits
• Six GPU tiers 16GB to 96GB VRAM
• Spot and on-demand pricing
• Multi-region AWS capacity
• Per-minute billing