V
Decentralized peer-to-peer GPU marketplace delivering lowest-cost AI training and inference.
Technical Overview & Architecture
Vast.ai is a peer-to-peer cloud compute marketplace aggregating global GPU capacity from vetted datacenters and independent hosts. Offering on-demand, interruptible (bid/spot), and reserved instances across RTX 4090, RTX 3090, A100, and H100 GPUs, Vast.ai delivers unmatched cost efficiency for batch inference and fine-tuning.
Pricing Breakdown
Transparent tiers and feature allotments for engineering teams.
RTX 4090 (24GB VRAM) Spot/Bid
zsh.20 - zsh.35 / hr
- 24GB GDDR6X VRAM
- 50-70% discount over on-demand
- PyTorch/Jupyter Docker templates
Most Popular
On-Demand A100 / H100
zsh.90 - .40 / hr
- 80GB VRAM
- Vetted datacenter tier
- Root SSH & direct port mapping
Reserved Long-term Hosting
Custom monthly
- Guaranteed instance uptime
- Static pricing
- Dedicated network bandwidth
Compare Vast.ai Against Alternatives
See how Vast.ai stacks up against competitor tools across speed, APIs, and pricing.