GPU cloud
Six dedicated NVIDIA cards, 24 to 141 GB.
Hikube GPU cloud provides six dedicated NVIDIA GPUs (L4 24G, L40S 48G, A100 80G, RTX 6000 Pro 96G, H100 80G, H200 141G) for inference, fine-tuning and training, attached to a VM or a Kubernetes node group, in Switzerland.
- Six cards, trial included
- VM or Kubernetes
- Datasets in Switzerland
- ISO 27001
- 3 DC
- SLA 99.99%
Hikube GPU cloud is for workloads that must train or infer in Switzerland. On Kubernetes: GPU Operator and HAMi (optional). HA spare-capacity restart in normal mode excludes GPUs. Terraform integration in preparation. Solution: inference and RAG.
Capabilities
Six cards, trial included
L4, L40S, A100-80, RTX 6000 Pro, H100, H200.
VM or Kubernetes
Same catalogue. GPU Operator and HAMi on the managed cluster.
Datasets in Switzerland
S3 object storage + Vault for job credentials.
Not a worldwide farm
Three Swiss DCs. Not a catalogue of thousands of GPUs.
API control
Hikube API and console. Terraform and Cluster API integration in preparation.
HA and GPUs
HA spare-capacity restart in normal mode excludes GPUs.
NVIDIA GPU catalog (CHF / month)
| Model | Memory | CHF/mo |
|---|---|---|
| L4 | 24G | 165 |
| L40S | 48G | 760 |
| A100-80 | 80G | 1495 |
| RTX 6000 Pro | 96G | 1550 |
| H100 | 80G | 1999 |
| H200 | 141G | 2499 |
What people assume, what is true
What people assumeGPU Operator and HAMi are separate Hikube products.
What is trueTwo optional add-ons of the managed cluster. GPU Operator installs the NVIDIA drivers and the device plugin; HAMi shares one card across several pods and requires GPU Operator. Both are switched on in the console, under CLUSTER ADDONS.
What people assumeThe HA capacity guarantee covers GPUs.
What is trueNo. In normal mode it excludes them. The underlying block storage stays geo-redundant, sync or async; a GPU is a compute attachment, not a volume. Detail: /en/products/block-storage.
What people assumeIt is a turnkey ML PaaS, or a global GPU farm.
What is trueNeither. Six dedicated cards across three Swiss datacentres, attached to a VM or a Kubernetes node group. Inference, fine-tuning, batch jobs, datasets that must not leave the country.
What people assumePublished FLOPS let you compare the cards.
What is trueThe catalogue publishes memory, not marketing IOPS or FLOPS. The SKU follows the model's VRAM, the length of the job and the VPC isolation. All six cards are in the trial: test before fixing a price.
Who answers for what
Hidora SA operates
- The datacentres and the GPU attachment
- The Kubernetes control plane, if you use it
- Support, in French and English
You keep
- Jobs, images and checkpoints
- Switching GPU Operator and HAMi on
- The choice between VM and node group
No CNCF AI conformance: it is not claimed.
In detail
The 14-day trial includes all six cards. An engineer in Geneva sizes the model (L4, L40S, A100-80, RTX 6000 Pro, H100, H200), duration, dataset storage and VPC isolation. No card. Request a trial, or talk to an engineer.
You attach the GPU to an instance or a Kubernetes node group. Provisioning via API and console. Terraform: in preparation. Endpoint detail: docs.hikube.cloud. Support FR/EN.
The catalogue publishes memory: L4 24G, L40S 48G, A100-80 80G, RTX 6000 Pro 96G, H100 80G, H200 141G. We do not publish IOPS or marketing FLOPS. An engineer frames the SKU on model VRAM, job duration and VPC isolation. All six cards are in the 14-day trial: test before you lock a tariff. Prices: pricing page.
Frequently asked questions
Yes. Hikube is Swiss cloud IaaS (VMs, managed Kubernetes, GPU, S3, backup, vault) operated by Hidora SA on three datacenters: Geneva, Gland, Lucerne. Published SLA 99.99%, ISO 27001, engineer support in French and English. It is not shared web hosting. GPUs are in the 14-day trial. Windows Server is a licensed image on instances.
No. Hikube is a sovereign cloud: compute, storage, backups and metadata stay on three independent Swiss datacenters (Geneva, Gland, Lucerne). There is no replication to the EU or the United States.
The operator is Hidora SA, a Swiss company in Lancy (Geneva), with no US parent. Compute, storage, backups and metadata stay in Geneva, Gland and Lucerne. That is not the same legal basis as AWS, Azure or GCP. We are not your counsel: GDPR for EU data in Switzerland relies in particular on adequacy.
Yes, subject to availability. The 14-day trial includes all six cards, H100 included. An engineer scopes the allocation against free capacity; there is no published cap. Production price: 1999 CHF/month.
Yes, optional addons on the managed cluster. HAMi requires GPU Operator. Platform page: the Kubernetes platform.
Yes. Hikube GPU cloud attaches L4, L40S, A100-80, RTX 6000 Pro, H100 or H200 to a VM or a Kubernetes node group, in Switzerland. All six cards are in the 14-day trial.
No. HA spare-capacity restart in normal mode excludes GPUs. All six SKUs stay in the trial.
Through the Hikube API and the console. Terraform and Cluster API integration in preparation. kubectl, Helm and S3 clients stay. Docs: docs.hikube.cloud.
Ready to run on 100% Swiss infrastructure?
14-day trial, no credit card. GPUs included.