Vast.ai

GPU marketplace for renting compute at market-driven prices with per-second billing

Vast.ai is a GPU cloud marketplace where providers worldwide list their hardware and users rent it at competitive, market-driven prices. Offers on-demand, interruptible, and auction instance types with per-second billing. Features a serverless inference layer that auto-selects hardware and routes workloads across regions. Supports GPUs from RTX 4090 to H200 and B200. SOC 2 Type II certified. A startup program offers up to $2,500 in free credits.

Pricing: Pay-as-you-go

Hosting Cloud

Pricing Usage Based, from ~$0.06/GPU/hr

HQ 🇺🇸 United States

Founded 2018

Visit website →

Pricing

Posts

Vast.ai operates a GPU compute marketplace where pricing is set by supply and demand rather than fixed rate cards. The platform lists 20,000+ GPUs across 68+ types in 40+ data centers globally. Three deployment models are available: GPU Cloud (on-demand instances), Serverless (inference endpoints), and Clusters (multi-node training). Billing is per-second with a $5 minimum account credit to start, no contracts or sales calls required.

The marketplace model means prices fluctuate. For popular GPUs like the A100 and H100, rates are often lower than fixed-price providers during off-peak demand, but can spike during high-demand periods. Developers who can tolerate price variability or schedule workloads flexibly benefit the most.

Vast.ai provides a REST API, Python SDK, and CLI for programmatic instance management. Pre-configured templates are available for common open-source models. The platform is SOC 2 certified.

Compared to fixed-price GPU providers like RunPod or Lambda, Vast.ai trades pricing predictability for potentially lower costs. Compared to inference API providers like DeepInfra or Together.ai, Vast.ai is lower-level: you get a GPU instance, not a managed API endpoint (though the Serverless product is moving in that direction).