About DataCrunch

DataCrunch is a Finnish provider running GPUs in Nordic data centers powered by wind and hydro. Like the other Nordic entries on this list, cheap clean power and free cooling translate directly into lower hourly rates than equivalent hardware in Virginia or Frankfurt.

It is smaller than Nebius and more self-service than CoreWeave, which makes it a practical middle option: real data centers and current hardware, but you can sign up and deploy without talking to sales.

Pricing

Billed per minute rather than per hour, which genuinely matters for short experiments. An RTX 6000 Ada sits around $1.00/hr, an A100 80GB SXM around $1.60/hr, and H200 nodes around $3.20/hr per GPU.

Spot instances are available at a substantial discount, with the usual condition that they can be reclaimed. For checkpointed training that is an easy win.

Storage is separate, and shared filesystems can be mounted across instances so a dataset does not need copying to every node.

Service Details

Instances are virtual machines running Ubuntu with CUDA images ready to go, reachable by SSH or through hosted JupyterLab. There is an API, a CLI, and a Terraform provider.

They also run serverless inference endpoints for popular open models, billed by request, which covers the case where you want to serve a model without keeping a GPU warm. That is a different cost curve from renting an instance and worth modelling before you commit either way.

Multi-node H200 configurations use InfiniBand. For single-GPU work the per-minute billing and spot pricing are the reasons to look here.

Available in: Finland, Iceland

Learn More

For more information about DataCrunch, visit their website at https://datacrunch.io/

Plan	Price/mo	CPU	RAM	Storage	Notes
RTX 6000 Ada	—	10 vCPU	60GB	150GB NVMe	Around $1.00/hr
A100 80GB SXM	—	22 vCPU	120GB	500GB NVMe	Around $1.60/hr
8x H200 SXM	—	176 vCPU	960GB	4TB NVMe	Around $3.20/hr per GPU