Pricecomparison.cloud · GPU cloud

GPU cloud: H100 & A100 hourly price

Training and inference for AI models need datacenter GPUs. We compare on-demand hourly pricing for the two most common: H100 (80 GB, current standard) and A100 (80 GB, cheaper). Specialized GPU clouds cost a fraction of AWS/GCP list prices. Finnish DataCrunch (now Verda) runs in Helsinki.

Prices verified – – services

GPU hours / mo 100 h/mo

Adjust GPU hours with the slider — $/mo columns calculate monthly cost (hours × $/h) · click headers to sort · bar = H100 $/hour (lower is better)

Provider	H100 $/h	H100 $/mo · 100 h ▼	A100 $/h	A100 $/mo · 100 h	Region	Billing	Trustpilot	Notes
Loading…

How to choose?

Vast.ai is the cheapest (marketplace where hosts compete), but price and availability vary — good for flexible, interruptible work. RunPod offers stable on-demand pricing and per-second billing. DataCrunch 🇫🇮 is Finnish and runs in Helsinki (data residency + low latency in the EU); spot prices are very competitive.

Lambda Labs and CoreWeave are larger, more stable players (CoreWeave is often contract-based and fully booked). AWS and GCP are references: list price is several times higher, and large workloads typically run on Savings Plans or contract pricing. Hetzner GPU does not offer H100/A100 but RTX workstation cards as dedicated — cheap, but a different use case. Spot/interruptible is always clearly cheaper than on-demand.