Vast.ai
Agent-Ready AI Infrastructure·vast.ai
Vast.ai is a GPU cloud marketplace that lets developers rent high-performance GPUs (RTX 4090s, H100s, B200s, and more) on demand from data centers and independent hosts worldwide, with per-second billing and no minimum commitment. It provides GPU Cloud for on-demand instances, Serverless for auto-scaling model endpoints, and Clusters for dedicated multi-node GPU training, all provisioned via a console, REST API, CLI, or Python SDK. GPU owners can also list idle hardware on the platform to earn money by hosting compute for renters.
What it's for
Features 18
- 68+ GPU types across 40+ data centers
- CLI
- Dedicated multi-node GPU clusters with InfiniBand networking
- GPU hosting for supplying idle hardware
- GPU rental marketplace
- Interruptible (preemptible) instances
- Live, real-time GPU pricing data
- On-demand GPU instances
- Per-second billing
- Pre-configured one-click model templates
- Pricing calculator
- Private VPN access and audit trails
- Python SDK
- Reserved instances with 1, 3, or 6 month terms
- REST API
- Secure Cloud isolated instances with SSH/CLI/API access
- Serverless GPU endpoints with autoscale to zero
- SOC 2 Type II certified environments
At a glance
Integrations
Complianceself-reported
Resources
Pricing
Pricing is set by real-time supply and demand across the marketplace rather than fixed monthly tiers. Example rates cited: a 4090 for under $5 total, or H100s from as little as $0.90/hour. Savings of up to 80% versus AWS, Azure, or GCP are claimed. A minimum of $5 credit is required to get started. Reserved pricing requires contacting sales.
Most popular tier; positioned for production workloads.
Guaranteed uptime, no interruptions
Best value tier, suited for batch training.
May be reclaimed (preemptible)
Long-term commitment tier; sign-up via Contact Sales.
1, 3, or 6 month terms
Volume discounts available on Reserved instances.