Vast.ai

Vast.ai runs a marketplace for GPU cloud compute. Independent hosts and data centers list their machines, and customers rent them as Docker container instances. The listings cover more than 68 GPU types, from consumer RTX 3060 cards up to H100, H200, and B200, spread across 40 or more data centers. Besides single instances, Vast.ai sells Serverless endpoints that can scale to zero and multi-node Clusters with InfiniBand. You can provision through the web console, a CLI, a Python SDK, or a REST API. There are no fixed plans. Supply and demand set the prices, and billing is per second with no minimum. On-demand instances are not interrupted. Interruptible instances are advertised as at least 50 percent cheaper but can be preempted. Reserved terms of 1, 3, or 6 months are advertised at up to 50 percent off, with pricing through sales. Independent reviews from mid-2026 put marketplace H100 rates at roughly 0.90 to 1.87 dollars per hour, and RTX 4090s at roughly 0.34 to 0.50 dollars per hour. Storage is billed even while an instance is stopped, and bandwidth costs extra. The platform suits cost-sensitive developers, researchers, and startups doing fine-tuning, batch inference, and experiments. Its main weakness is that reliability is inconsistent. Quality depends on the host, standard marketplace instances have no platform uptime SLA, and reviewers report disconnections and limited support. A Secure Cloud filter limits results to vetted, ISO 27001 data center hosts, and Vast.ai reports SOC 2 Type I compliance.