GPU pricing in 2026: H100, A100, L40S cost per hour across 17 cloud vendors

The cheapest published H100 cost per hour we track in June 2026 is the Vast.ai marketplace floor at around $1.49 per GPU-hour for interruptible capacity. The next tier up is RunPod Community Cloud H100 PCIe at $1.99 per GPU-hour and DigitalOcean H100 at $1.99 on a 12-month reservation. Specialist-cloud reserved H100 SXM rates land between $3.29 per GPU-hour (Lambda Labs on-demand per-GPU on 8x, Together AI reserved cluster) and $6.155 per GPU-hour (CoreWeave HGX H100 8x node list rate). Hyperscaler on-demand H100 list rates run $11 to $13 per GPU-hour before negotiated discount. All rates verified June 2026 against the vendor pricing pages linked on each vendor page.

Four pricing models dominate. On-demand per-hour is the default at hyperscalers (AWS, Azure, GCP) and specialist clouds (CoreWeave, Crusoe, Hyperstack). Per-second billing is the Modal, Replicate, and RunPod model and is the right shape for inference and short bursts. Marketplace models (Vast.ai) clear capacity from independent hosts at the cost of interruptibility and variance. Reserved capacity (1 to 36 month) is how every serious training cluster is procured and unlocks the headline low rates. All four are explained on the How vendors price GPU compute page.

No. Every rate on this site is the public list price taken from the vendor pricing page linked on each vendor page. Actual prices vary by region, by GPU SKU and topology, by commitment term, by negotiated discount on enterprise contracts, and by add-on costs (storage, egress, support tier, MLOps platform). The Hidden GPU cloud costs page covers the line items that show up after the headline GPU-hour rate. Use the calculator on the homepage to scope spend, then take it to procurement for a real quote.

Yes, on the Quote-only vendors page. We list each vendor with its contact page and we do not infer rates. As of June 2026 the quote-only vendors in our coverage set are Fly.io GPU (service deprecated through 2025) and OVHcloud GPU (rates not consistently surfaced on the public pricing index). For both, the only honest answer is a sales quote.

Because GPU procurement decisions move millions of dollars and the affiliate-incentive distortion is too large to risk. This site is independent of every vendor listed. No affiliate links. No sponsored placements. The full statement is on the About page.