Vast.ai is a GPU marketplace with some of the lowest cloud GPU prices, but host quality, uptime and pricing vary widely between listings. Its hosts range from individual hobbyists to large datacenters, and it lists a wide range of hardware.
Takeaways
- Vast.ai has some of the lowest headline GPU prices, but host quality, uptime and pricing vary between listings.
- Vast.ai prices move about 7% a day in a typical week, based on daily checks of verified US/CA hosts.
- Only 4 GPUs are reliably available on Vast.ai: RTX 5090, RTX 4090, RTX PRO 6000 and A100 40GB.
- Datacenter GPUs like the H100 and H200 usually come from only a few Vast.ai hosts.
- Thunder Compute has the cheapest A100 80GB among dedicated alternatives, at $1.09/hr billed per minute.
On-demand GPUs
Starting at $0.35/hr, billed per minute.
Vast.ai - GPU Rental Marketplace

Vast.ai is a decentralized GPU marketplace without fixed pricing or standardized infrastructure. Instead, you browse listings by price, hardware and availability.
When Vast.ai Shines
Vast.ai is best for flexible, cost-sensitive workloads that can tolerate unpredictable performance.
- Bursty inference, experiments and hobby training on consumer cards
- Low headline prices, if you shop around and accept variability
When Vast.ai Struggles
Vast.ai's variability affects real-world workloads. Bandwidth is billed separately, and storage charges continue while an instance is stopped, so total cost depends on the host and usage.
- Mixed reliability and performance across heterogeneous hardware
- Prices and availability that can change hourly
- Security and compliance needs that may require vetted datacenter hosts
- Substantial egress and ingress rates at some hosts
- Storage billed separately
Vast.ai Pricing and Availability
Thunder Compute has checked Vast.ai's marketplace daily since Aug 5, 2026. For each GPU model, we record:
- Verified hosts in the US and Canada. Unverified hosts and hosts in other regions are excluded, so the global market is larger than these counts show.
- The median price/GPU-hour across those hosts, which is the figure plotted in the charts. We also record the lowest and highest listing.
- The number of distinct hosts listing the GPU that day. A day with no qualifying host counts as unavailable.
- Prices include compute and 100GB of storage. Each host bills bandwidth separately, and bandwidth isn't included.
Unless noted, the Vast.ai charts use the last 30 daily checks.
Daily Price Changes
Vast.ai GPU prices move about 7% from one day to the next in a typical week. Between August and November 2026, that weekly figure ranged from 4% to 15%. Each GPU is weighted by how many hosts list it.
Which GPUs Can You Reliably Find on Vast.ai?
Only 4 GPUs are reliably available on Vast.ai: RTX 5090, RTX 4090, RTX PRO 6000, A100 40GB. Another 7, including the H100, H200 and A100 80GB, are limited: you can usually find one, but from only a few hosts. The rest (L40S, L40, T4, RTX A5000, A40) were listed on fewer than 15 of the last 30 days.
Prices Trends of Reliably Available GPUs on Vast.ai
Even reliably available GPUs swing in price. Over the last 30 checks, the A100 40GB was the steadiest, with its highest daily median 29% above its lowest. The RTX 4090 varied the most, from $0.48 to $1.03 per hour.
Vast.ai Competitors
For teams that need more predictable performance and transparent pricing, several Vast.ai alternatives offer dedicated datacenter GPUs with consistent infrastructure.
| Provider | A100 80 GB ($/hr) | H100 80 GB ($/hr) | Notable Features & Storage |
|---|---|---|---|
| Vast.ai 1 | $1.28 | $2.40 | Marketplace pricing with variable host quality, availability, storage, and bandwidth charges. |
| Thunder Compute | $1.09 | $3.20 | On-demand NVIDIA GPUs with no ingress/egress fees. Hibernation and spot pricing for cost control. |
| Hyperstack | $1.35 | $2.50 | Prices based on SXM variants; focuses on sustainable computing and reserved options. |
| Crusoe Cloud | $2.00 | $3.90 | Prices based on SXM variants; focuses on sustainable computing and reserved options. |
| CoreWeave | $2.70 | $6.16 | NVLink/HGX specialized cloud; public price card normalized from 8-GPU node pricing. |
| Paperspace | $3.18 | $5.95 | DigitalOcean ecosystem; high-availability on-demand instances for Gradient and ML. |
| Lambda | $3.19 | $3.79 | Standard on-demand pricing; specialized for ML and research instances. |
| 1 Vast.ai A100 80 GB and H100 80 GB prices are current verified US/CA marketplace listings, priced as compute plus 100 GB storage; bandwidth is billed separately. Last reviewed on November 13, 2026. | |||
NoteNote on consumer cards vs datacenter GPUs
Marketplaces commonly list consumer GPUs because hobbyists supply much of their capacity, so cards like the RTX 4090 and 3090 are abundant. A tested datacenter A100 80GB or H100 80GB is usually the safer choice for predictable training throughput. Vast.ai's documentation explains the difference between its community and datacenter servers.
1. Thunder Compute

Thunder Compute is a low-cost, reliable option for developers who need flexibility. Per-minute billing means you pay only for the compute you use. The native VS Code integration suits prototyping and iterative development.
- A100 80 GB at $1.09/hr, billed per minute
- H100 80 GB at $3.20/hr
- Persistent instance storage, snapshots, and spec changes without rebuilds
- One-click VS Code integration and a minimal interface that is easy to learn
- No ingress or egress fees
If you are coming from a marketplace, you can expect fewer surprises in performance and uptime, while still paying less than most datacenter clouds.
2. Hyperstack

Hyperstack is a full-stack AI cloud and NVIDIA Cloud Partner built for production workloads, from single-GPU training to distributed inference across clusters. Hyperstack On-Demand offers a wide GPU lineup with transparent pricing and no hidden costs.
- NVIDIA A100, H100, H100 PCIe, H200 SXM, RTX PRO 6000 SE and B300 GPUs
- No ingress or egress fees
- GPU VM hibernation to cut idle costs, plus spot VMs at 20% below on-demand prices
- AI Studio for serverless inference with popular open-source models
- 1-click deployment, supporting Ubuntu, AlmaLinux and Debian images
- On-demand Kubernetes and object storage
Hyperstack also offers Secure Private Cloud, an isolated environment with reserved capacity for compliance-sensitive work.
3. Crusoe Cloud

Crusoe Cloud aligns high-performance computing with environmental sustainability by powering its data centers using otherwise wasted or stranded energy. They offer a clean conscience alongside competitive rates tailored for robust SXM configurations, appealing directly to eco-conscious enterprises and research teams.
- A100 80 GB: $2.00/hr
- H100 80 GB: $3.90/hr
- Notable Features & Storage: Rates are based on SXM variants. Crusoe Cloud focuses on climate-aligned computing and reserved capacity.
4. Lambda

Lambda is a long-standing GPU cloud for machine learning, with infrastructure built for deep learning. Its transparent on-demand pricing has no hidden fees, which suits training and scaling research models.
- A100 80 GB: $3.19/hr
- H100 80 GB: $3.79/hr
- Notable Features & Storage: Straightforward on-demand pricing. Instances are built for ML training and research.
5. CoreWeave

CoreWeave is a large-scale specialized cloud for multi-GPU workloads that need maximum throughput. Its infrastructure is built around NVLink and HGX topologies, serving large enterprise AI deployments and rendering jobs.
- A100 80 GB: $2.70/hr
- H100 80 GB: $6.16/hr
- Notable Features & Storage: Built for NVLink/HGX topologies. CoreWeave's public rates are normalized from 8-GPU node pricing.
6. Paperspace

Paperspace, part of DigitalOcean, offers high-availability cloud GPUs with a simple developer experience. Its integration with the Gradient ML platform helps teams move projects from experimentation to production pipelines.
- A100 80 GB: $3.18/hr
- H100 80 GB: $5.95/hr
- Notable Features & Storage: High-availability on-demand instances for Gradient and ML pipelines.
Last Thoughts on Vast.ai Alternatives
Vast.ai trades reliability for price: it suits experiments, hobby workloads and flexible inference, but its prices move daily and most datacenter GPUs come from only a few hosts. For production training, fine-tuning or any workload where uptime matters, a dedicated provider like Thunder Compute offers predictable performance and pricing.
FAQ
What is the cheapest alternative to Vast.ai for A100 GPUs?
Thunder Compute offers the cheapest A100 80GB among Vast.ai alternatives at $1.09/hr, billed per minute. Hyperstack charges $1.35/hr, Crusoe Cloud $2.00/hr, CoreWeave $2.70/hr and Lambda $3.19/hr.
What are the main disadvantages of using GPU marketplaces like Vast.ai?
GPU marketplaces like Vast.ai have three main drawbacks: inconsistent reliability and performance across hosts, prices and availability that change daily, and security or compliance gaps that may require vetted datacenter hosts.
How much do Vast.ai prices change from day to day?
In a typical week, the median Vast.ai GPU price moves about 7% from one day to the next. Between August and November 2026, the weekly figure ranged from 4% to 15%. Fixed-price clouds change their prices only a few times a quarter.
Which GPUs are reliably available on Vast.ai?
Four GPUs averaged more than 5 verified US/CA hosts a day in the 30 days to November 3, 2026: RTX 5090, RTX 4090, RTX PRO 6000 and A100 40GB. The H100, H200 and A100 80GB were usually listed, but by only a few hosts.
Does Vast.ai pricing include storage and bandwidth?
No. Vast.ai bills storage and bandwidth separately from compute, and some hosts charge substantial egress and ingress rates. Thunder Compute's tracked Vast.ai prices include compute plus 100GB of storage, but not bandwidth.