Vast.ai
Vast.ai is a two-sided GPU compute marketplace that connects developers with 1,400+ independent hosts offering underutilized GPU hardware across 500+ locations worldwide. By aggregating supply from independent data centers and individual providers, Vast.ai delivers GPU compute at 40–80% less than traditional hyperscalers like AWS, Azure, and GCP.
The platform supports three deployment models: GPU Cloud (individual instances), Serverless (autoscaling inference endpoints), and Clusters (multi-node training). Users can spin up high-performance instances in seconds using Docker containers, choosing from 35+ GPU types ranging from consumer RTX cards to enterprise B200s. Pricing is set dynamically by the marketplace based on supply and demand.
Founded in 2018 by Jake Cannell, the company manages 20,000+ GPUs and achieved 310% growth in 2024. It holds SOC 2 Type 2 certification and is particularly popular among AI researchers, ML engineers, and indie developers who need affordable compute without long-term commitments.
GPU Hardware
| GPU Models | B200 H200 H100 SXM H100 NVL A100 SXM A100 PCIe RTX 5090 RTX 5080 RTX 5070 Ti RTX 6000 Pro RTX 6000 Ada RTX 4500 Ada RTX A6000 RTX A5000 RTX A4000 L40S L40 A40 A10 RTX 4090 RTX 4080 RTX 4070 Ti RTX 4070 RTX 4060 Ti RTX 4060 RTX 3090 Ti RTX 3090 RTX 3080 Ti RTX 3080 RTX 3070 Ti RTX 3070 Tesla V100 Tesla T4 A2 GTX 1080 |
| Max VRAM | 192 GB |
| Max GPUs per Instance | 8 |
| Interconnect | NVLink, InfiniBand |
| Multi-Node Training | Yes |
Pricing
| Starting Price | $0.06/hr |
| Billing Granularity | Per-second |
| Spot/Preemptible | Yes |
| Reserved Discounts | Up to 50% (1-6 month reserved) |
| Free Credits | Small test credit on signup |
| Egress Fees | Varies by host ($/TB) |
| Storage | Varies by host ($/GB/hr, charged while instance exists) |
Marketplace-driven pricing: hosts set their own rates based on supply and demand. Three tiers available — On-Demand (guaranteed uptime), Interruptible (50%+ cheaper via bidding), and Reserved (1/3/6-month terms). Budget: RTX 4060 from $0.06/hr. Mid-range: RTX 4090 from $0.29/hr, A100 from $0.67/hr. High-end: H100 from $1.55/hr, H200 from $1.97/hr, B200 from $2.67/hr. Note: storage is charged even when instances are stopped, and bandwidth fees apply per TB. $5 minimum deposit to start.
Infrastructure
| Regions | 500+ locations, 40+ data centers |
| Uptime SLA | No formal SLA (host reliability scores visible) |
| Serverless / Autoscaling | Yes |
| Private Networking / VPC | Yes |
Developer Experience
| Pre-installed Frameworks | PyTorch TensorFlow CUDA vLLM ComfyUI |
| Docker Support | Yes |
| SSH Access | Yes |
| Jupyter Notebooks | Yes |
| API / CLI | Yes |
| Setup Time | Seconds |
| Kubernetes Support | No |
| Custom Images / Templates | Yes |
| Persistent Storage | Yes |
Business Terms
| Min Commitment | None |
| Compliance | SOC 2 Type 2 HIPAA GDPR CCPA |
| Best For | AI training inference fine-tuning Stable Diffusion batch processing research LLM serving generative AI |
| Support Channels | Live Chat (24/7) Discord Email Documentation |
| Payment Methods | Credit Card Crypto (Coinbase Crypto.com) |
How does it compare?
Compare Vast.ai against other cloud GPU providers.
Frequently Asked Questions
What type of workloads is Vast.ai ideal for?
Vast.ai caters primarily to users who need: AI training, inference, fine-tuning, Stable Diffusion, batch processing, research, LLM serving, generative AI
As a GPU Marketplace provider, Vast.ai differentiates itself through its GPU selection and pricing model. Starting at $0.06/hr, the platform provides accessible compute for researchers, startups, and enterprises building AI applications.
GPU models available: B200, H200, H100 SXM, H100 NVL, A100 SXM, A100 PCIe, RTX 5090, RTX 5080, RTX 5070 Ti, RTX 6000 Pro, RTX 6000 Ada, RTX 4500 Ada, RTX A6000, RTX A5000, RTX A4000, L40S, L40, A40, A10, RTX 4090, RTX 4080, RTX 4070 Ti, RTX 4070, RTX 4060 Ti, RTX 4060, RTX 3090 Ti, RTX 3090, RTX 3080 Ti, RTX 3080, RTX 3070 Ti, RTX 3070, Tesla V100, Tesla T4, A2, GTX 1080.
Compare Vast.ai features with your project requirements at Vast.ai official website.
What is Vast.ai Trustpilot rating and total review count?
Vast.ai holds a Trustpilot rating of 4.1/5.0 with 237 reviews as of June 30, 2026. The company has been operating since 2018.
Trustpilot reviews provide insight into the day-to-day experience of using a cloud GPU provider, including billing transparency, instance availability, and technical support quality. Read the full reviews on Trustpilot.
Check the latest features and pricing at Vast.ai official website.
Can I use custom ML frameworks on Vast.ai?
The ML framework ecosystem at Vast.ai:
PyTorch, TensorFlow, CUDA, vLLM, ComfyUI
These frameworks come pre-configured and optimized for the available GPU hardware. Custom images (Yes) allow you to deploy any framework or library not included by default, including nightly builds, custom forks, or specialized inference engines like vLLM or TensorRT.
Persistent storage (Yes) ensures your datasets and model weights survive instance restarts.
Check which ML frameworks come pre-installed on Vast.ai official website.
What developer tools are available at Vast.ai?
Here is the developer experience at Vast.ai:
Setup time: Seconds — this is how quickly you can provision and access a GPU instance after initiating the request.
Available tools:
- Docker containers: Yes
- Direct SSH access: Yes
- Jupyter notebooks: Yes
- Programmatic API/CLI: Yes
- Custom Docker images: Yes
This combination of tools makes Vast.ai suitable for both exploratory research (Jupyter) and production MLOps pipelines (API + Docker + SSH).
See the full setup documentation and API reference on Vast.ai official website.
Does Vast.ai support scale-to-zero GPU deployments?
Serverless availability at Vast.ai: Yes
With serverless GPU, you deploy a model container and the platform handles autoscaling, load balancing, and cold starts automatically. You pay only when your endpoint is processing requests — there are no charges during idle time. This can reduce costs by 80-95% compared to always-on dedicated instances for bursty inference workloads.
Vast.ai on-demand pricing starts from $0.06/hr (Per-second billing).
View serverless deployment options and cold-start benchmarks on Vast.ai official website.
What is Vast.ai uptime SLA guarantee?
Vast.ai operates out of United States with GPU infrastructure available in:
500+ locations, 40+ data centers
The platform provides an uptime guarantee of No formal SLA (host reliability scores visible). For teams running production-grade inference or training pipelines, the combination of regional availability and SLA commitments should factor into provider selection.
Private networking / VPC: Yes
For region-specific pricing and compliance details, visit Vast.ai official website.
Can I run distributed training across multiple GPUs at Vast.ai?
Here is how Vast.ai handles multi-GPU workloads:
GPU interconnect: NVLink, InfiniBand
Maximum GPUs: 8 per instance
Multi-node support: Yes
When scaling beyond a single GPU, the interconnect technology determines how efficiently GPUs communicate during distributed training operations like all-reduce and gradient synchronization. Vast.ai offers NVLink, InfiniBand connectivity, which is an important factor when comparing multi-GPU providers for large-scale AI workloads.
Check available multi-node cluster configurations on Vast.ai official website.
Are spot instances available at Vast.ai for cost savings?
Does Vast.ai offer spot instances? Yes
Spot or preemptible instances are one of the most effective ways to reduce cloud GPU costs. They work by offering unused GPU capacity at a discount, with the trade-off that the provider can reclaim the instance with short notice. This is well-suited for training jobs that save checkpoints frequently, as well as batch inference workloads.
Base pricing at Vast.ai: $0.06/hr (Per-second billing).
For spot instance terms and savings estimates, visit Vast.ai official website.
How much does Vast.ai charge for outbound data transfer?
Vast.ai handles data transfer fees as follows: Varies by host ($/TB)
This matters particularly for teams running distributed training across providers or serving models via API endpoints that return large payloads. Zero or low egress fees can significantly reduce overall costs for production inference deployments.
Available storage: Varies by host ($/GB/hr, charged while instance exists).
Check the full bandwidth and data transfer pricing on Vast.ai official website.
How can I get free GPU credits at Vast.ai?
Free credit availability at Vast.ai: Small test credit on signup
Vast.ai GPU instances start from $0.06/hr, so any promotional credits can stretch across multiple short training runs or inference tests.
Claim your free GPU credits at Vast.ai official website.
What is the maximum VRAM available on Vast.ai GPU instances?
The GPU fleet at Vast.ai includes both data center and workstation-class accelerators:
B200, H200, H100 SXM, H100 NVL, A100 SXM, A100 PCIe, RTX 5090, RTX 5080, RTX 5070 Ti, RTX 6000 Pro, RTX 6000 Ada, RTX 4500 Ada, RTX A6000, RTX A5000, RTX A4000, L40S, L40, A40, A10, RTX 4090, RTX 4080, RTX 4070 Ti, RTX 4070, RTX 4060 Ti, RTX 4060, RTX 3090 Ti, RTX 3090, RTX 3080 Ti, RTX 3080, RTX 3070 Ti, RTX 3070, Tesla V100, Tesla T4, A2, GTX 1080
Maximum VRAM per GPU: 192 GB
Maximum GPUs per instance: 8
Interconnect: NVLink, InfiniBand
This hardware selection covers use cases from cost-effective inference on consumer GPUs to large-scale distributed training on enterprise accelerators.
For detailed GPU specs, VRAM configurations, and multi-GPU options, check Vast.ai official website.
What are the pricing plans and billing options at Vast.ai?
GPU compute at Vast.ai is billed on a Per-second basis with rates beginning at $0.06/hr for the most affordable GPU option. This billing granularity is particularly useful for short training runs, experimentation, and inference tasks where you may only need a GPU for minutes at a time.
Does Vast.ai offer spot instances? Yes
Are reserved discounts available? Up to 50% (1-6 month reserved)
Payment methods: Credit Card, Crypto (Coinbase, Crypto.com).
View the complete GPU pricing calculator at Vast.ai official website.
User Feedback
There are no public trader reviews for this firm yet. If you have traded with them, be the first to leave a short, honest review and help other traders.
Share Your Experience
Short, honest feedback helps other prop traders understand what it is really like to work with this firm.