Vast.ai vs Thunder Compute (September 2025): Pricing, Hardware, and Which One Fits Your ML Workloads

• Thunder Compute pricing: see the Thunder Compute pricing page.
• Vast.ai pricing fluctuates by host: check the Vast.ai pricing page.

TL;DR

• Thunder Compute: Flat rates (A100 80 GB at $0.78/hr, H100 at $1.47/hr), per second billing, persistent storage, snapshots, VS Code integration in one click.
• Vast.ai: Dynamic marketplace where you bid on individual hosts. Many consumer cards like RTX 4090 are available. A100 and H100 prices vary and may require monitoring.
• Choose Thunder if you want managed infrastructure, predictable billing, and data center GPUs. Choose Vast if you want to hunt for the absolute cheapest listing and do not mind variability in reliability or hardware type.

Hardware: Consumer vs Data Center Cards

Vast.ai is a crowdsourced marketplace. Many hosts offer consumer-grade GPUs (RTX 4090, 3090), plus some data center GPUs. You will need to vet each listing for bandwidth, CPU, and storage limits. See the Vast.ai FAQ.
Thunder Compute provides data center GPUs (A100, H100). Performance is consistent, and the environment is managed for reliability. Visit the Thunder Compute homepage.

Billing and Payment Flow

• Vast.ai: You typically load credits, then pay per second for GPU, storage, and bandwidth. Each host sets those rates. See the Vast.ai billing guide.
• Thunder Compute: Pay per second, no prepaid credits. Storage is $0.15/GB/month for persistent volumes. Snapshots are available and spec changes do not require rebuilding your machine. Details on the Thunder Compute pricing page.

Interruptibility and Reliability

• Vast.ai: Instances can be interruptible based on bidding priority. If you choose an interruptible instance and someone outbids you or the host needs resources, your job can be preempted. See Vast.ai rental types.
• Thunder Compute: On-demand instances are not terminated mid-run. You can snapshot, resize, and continue work without losing progress.

Developer Experience

• Thunder Compute: One click integration with VS Code, persistent storage, spec changes on the fly, minimal interface. Thunder Compute homepage.
• Vast.ai: Fine-grained control but more manual work. You pick hosts, check their speeds, and manage data hygiene across different machines. Vast.ai pricing page.

Which One Should You Pick?

• Pick Thunder Compute if:
  • You need consistent, data center-grade GPUs.
  • You want predictable pricing and billing without host-by-host management.
  • Your team relies on persistent storage, snapshots, and a simple VS Code workflow.
• Pick Vast.ai if:
  • You are optimizing for the absolute lowest possible price and are comfortable monitoring hosts.
  • You are okay with consumer GPUs and potential variability in bandwidth and stability.
  • You can handle potential interruptions or prioritize short-lived training jobs.