Where Is Your GPU Cheapest to Rent Right Now?

The live table refreshes hourly. We call each provider’s public API directly, normalize SKU names to canonical GPU specs, and cache the result for one hour so a high-traffic moment never overwhelms the upstream services. A separate daily snapshot records prices for the 30-day trend chart.

Either the provider does not stock that GPU, or its public API uses an SKU name we have not mapped yet. Unmapped SKUs are logged on every sync so we can extend the mapping. If a card you expect is missing, get in touch.

Secure refers to vetted Tier-3 or Tier-4 datacenter hosts. Community refers to lower-cost hosts on the same provider, often peer-operated. Spot is interruptible compute, the cheapest tier but can be reclaimed by the provider with little notice. On-demand is the standard non-interruptible price.

Per GPU. Where a provider quotes a multi-GPU instance, we divide by the GPU count so prices are directly comparable across single and multi-GPU listings.

Prices vary by region, host reliability, and availability. We surface the cheapest currently rentable per-GPU rate from each provider’s public marketplace or pricing API. The provider page might show a different default region or a non-rentable listing.

Vast.ai exposes many independent host listings per GPU model. We capture two snapshots per group, the absolute minimum and the 25th percentile, so a single outlier listing cannot misrepresent the price. The "min" row is what surfaces as the headline.

They are on the roadmap. Lambda Labs requires an account-tied API key that is gated for new signups, and CoreWeave, Modal, and Colab do not expose clean public pricing APIs. RunPod and Vast.ai together cover the bulk of practical price comparison for self-serve renters, and we will add more sources as access opens up.