Vast.ai is a GPU marketplace where independent hosts list spare compute capacity at prices they set, creating a competitive environment that can deliver rates well below managed cloud providers. Thunder Compute is managed infrastructure with fixed pricing, consistent hardware, and persistent storage included by default.
This comparison covers Vast.ai's current pricing across GPU types, billing structure, storage costs, and the practical trade-offs between a marketplace model and a managed cloud.
Key takeaways:
- Vast.ai H100 on verified datacenter hosts: about $2.21/hr on-demand. Interruptible instances can be cheaper.
- Thunder Compute H100 PCIe: $2.19/hr fixed. A100 80GB: $1.09/hr.
- Vast.ai storage is charged per GB/hr even when instances are stopped. Delete to stop billing.
- Vast.ai bills per second for GPU time. Thunder Compute bills per minute for the full instance.
- Vast.ai minimum deposit: $5 via credit card or crypto. Thunder Compute bills directly, no prepaid credits.
Consider Vast.ai Alternatives for affordable and reliable GPU instances.
Vast.ai GPU Pricing (August 2026)
Vast.ai pricing is dynamic and set by individual hosts. The figures below represent typical marketplace rates on verified datacenter hosts as of August 2026. Interruptible instances are generally 30–50%+ cheaper than on-demand from the same host.
| GPU | Vast.ai On-demand1 | Vast.ai Interruptible1 | Thunder Compute |
|---|---|---|---|
| H100 80GB | $2.21/hr | Varies by host | $2.19/hr (PCIe) |
| A100 80GB | $1.32/hr | Varies by host | $1.09/hr |
| RTX 4090 | $0.39/hr | Varies by host | Not offered |
| RTX A6000 | $0.41/hr | Varies by host | $0.35/hr |
| 1 Vast.ai prices are marketplace rates set by individual hosts and may fluctuate. | |||
Vast.ai Billing and Storage: What the Headline Rate Misses
Vast.ai charges three separate cost streams:
- GPU compute (per second, only while the instance runs)
- Storage (per GB/hr continuously while the instance exists, whether running or stopped)
- Bandwidth (per byte transferred, at rates set by each host).
The storage billing structure catches many first-time users off guard. If you allocate 200GB of disk and stop your instance for a weekend to save compute costs, you continue paying for the 200GB the entire time. At approximately $0.10–$0.15/GB/month, 200GB paused for 10 days costs roughly $6.67–$10 in storage alone. The only way to stop storage charges is to delete the instance entirely.
Thunder Compute includes 100GB per GPU at no extra charge, and storage only accrues when an instance exists. Snapshots for long-term environment preservation cost $0.05/GB/month. There are no bandwidth charges.
Vast.ai Instance Types: On-Demand, Interruptible, and Reserved
Vast.ai offers three rental tiers with meaningfully different price and reliability trade-offs:
- On-demand instances for guaranteed resources and fixed pricing from the selected host. The host cannot reclaim the GPU while you're renting.
- Interruptible instances are 30–50%+ cheaper than on-demand from the same host. They can be paused at any time when a higher-priority job bids for the same GPU.
- Reserved instances offer a middle ground: pre-payment locks in access with more stability than interruptible, typically at a discount versus on-demand.
Vast.ai Host Fees and Payment
Vast.ai does not charge renters a platform fee or markup on host-set prices. You pay the host's listed rate directly, with separate charges for storage and bandwidth also set by the host. This means the total cost of a Vast.ai instance is: (GPU compute × seconds used) + (storage GB × hours allocated) + (bandwidth used × host rate).
Vast.ai requires a minimum $5 deposit to begin renting. Credits are pre-loaded and drawn down as usage accumulates. Accepted payment methods include major credit cards and cryptocurrency. When credits reach zero, running instances are automatically stopped; if no payment method is on file, instances and data are scheduled for deletion.
Thunder Compute bills directly to a credit card on a monthly basis with no prepaid credit requirement.
Host Selection: Verified vs Unverified
Vast.ai's marketplace spans over 17,000 GPUs across 1,400+ providers. Host quality varies significantly, and the cheapest listings are often from unverified or lower-reliability hosts. For jobs where interruption or hardware failure would be costly, filtering by verified hosts and reliability score above 99% is important.
Verified datacenter hosts typically list H100s around $2.21/hr on-demand, which is competitive with managed cloud rates. Unverified hosts may list lower H100 rates, but the risk of mid-run interruption or data loss is substantially higher.
Thunder Compute provides consistent hardware from managed data centers. Every instance uses the same hardware profile, and there is no host-level variance to manage.
Developer Experience
Vast.ai offers SSH access, Jupyter Lab, and container-based deployment. Users select hosts individually, check their bandwidth and reliability scores, and manage their own environment setup per instance. Fine-grained control is available, but it requires more active host management than most managed clouds.
Thunder Compute provides one-click VS Code, Cursor, and Windsurf extensions that connect directly to a running instance. Persistent storage means your installed packages and working directory are ready on every launch. There is no host selection process or per-instance environment management required.
When to Choose Each Provider
Choose Vast.ai if you are:
- Optimizing for the absolute lowest GPU rate
- Comfortable vetting individual hosts.
- Checkpointing frequently.
- Willing to manage the storage billing model carefully.
Choose Thunder Compute for:
- Consistent datacenter GPUs at a predictable managed rate.
- IDE-first development with no host management overhead.
- Workloads that need to run uninterrupted for hours or days.
Last Thoughts on Vast.ai vs Thunder Compute
Vast.ai and Thunder Compute serve different risk tolerances. Vast.ai's marketplace model delivers the lowest possible rates for teams that actively manage host selection and accept the variability that comes with it. Thunder Compute offers lower complexity and consistent performance at a competitive managed rate, with persistent storage and IDE integration included.
FAQ
How much does Vast.ai charge for H100 GPUs?
Vast.ai H100 is listed at approximately $2.21/hr on-demand in the current pricing reference. Interruptible instances can be cheaper, but prices are set by individual hosts and fluctuate with supply. Thunder Compute H100 PCIe is a fixed $2.19/hr on managed infrastructure.
What is Vast.ai's storage pricing?
Vast.ai storage is host-defined, typically $0.10-$0.15/GB/month. Critically, storage is charged per GB/hr whether your instance is running or stopped. Delete instances entirely to stop storage billing. Thunder Compute includes 100GB per GPU at no extra charge.
Does Vast.ai bill per second or per minute?
Vast.ai bills per second for active GPU time. Storage is billed continuously per GB/hr regardless of whether the instance is running. Thunder Compute bills per minute for the full instance.
What is the difference between Vast.ai interruptible and on-demand instances?
On-demand instances have fixed pricing and guaranteed resources. Interruptible instances are 30-50%+ cheaper but can be paused by the host or a higher-bidding renter at any time, with minimal notice. Use interruptible only for checkpoint-friendly workloads.
What is Vast.ai's minimum deposit?
Vast.ai requires a minimum $5 deposit to start renting. Credits are pre-loaded and drawn down as you use compute, storage, and bandwidth. Thunder Compute bills directly to a credit card with per-minute billing and no prepaid credits required.
What payment methods does Vast.ai accept?
Vast.ai accepts major credit cards and cryptocurrency. Thunder Compute accepts major credit cards with no prepaid credit requirement.
Does Vast.ai charge a platform fee or markup?
No. Vast.ai does not charge renters a platform fee or markup on top of host-set prices. You pay the host's listed rate directly, plus separate charges for storage and bandwidth set by the same host.
Which is better for ML teams: Vast.ai or Thunder Compute?
Thunder Compute for teams needing consistent datacenter GPUs, predictable pricing, persistent storage, and VS Code integration. Vast.ai for teams optimizing for the lowest possible rate and comfortable managing host selection, interruptibility, and variable storage billing.