AI and GPU cloud in Rwanda.
Dedicated NVIDIA and AMD GPU servers for training, fine-tuning and inference, in our Kigali region, at published prices in Rwandan francs. From RWF 3,250 an hour. Your data, your models and your backups stay in Rwanda, under Rwandan law.
Plans and prices, published.
A GPU server works like any other server on the Gewape Cloud Platform: launch it from the console or the API, pay by the hour or reserve it by the month, and keep it next to the rest of your cloud.
| Plan | GPU | CPU · memory | Storage | Per hour | Per month | 12-month plan, per month |
|---|---|---|---|---|---|---|
| gpu.l40.1 | 1 × NVIDIA L40 48 GB | 16 vCPU 96 GB memory | 100 GB boot disk 1.5 TB NVMe storage | RWF 3,250 | RWF 2,051,300 | RWF 1,697,100 |
| gpu.l40.2 | 2 × NVIDIA L40 48 GB | 32 vCPU 192 GB memory | 200 GB boot disk 3 TB NVMe storage | RWF 6,490 | RWF 4,102,500 | RWF 3,394,200 |
Run open models on a GPU that answers only to you.
Use open-weight models such as Llama, Qwen, Mistral, Gemma or DeepSeek with the open-source tools your team already knows: PyTorch, vLLM, Ollama and Hugging Face, on images with NVIDIA CUDA or AMD ROCm installed. A model you pin on your own GPU does not change underneath you. A hosted AI service can update or retire a model on its own schedule; open weights you run yourself make last quarter’s decision reproducible this quarter, which is exactly what an auditor asks for.
Inference
Serve chat assistants, search and document question-answering from open models you choose.
Fine-tuning
Adapt an open model to your language, your documents and your domain.
Computer vision
Image classification, detection and inspection for agriculture, health and logistics.
Speech and documents
Transcription, OCR and extraction over files that are not allowed to leave the country.
The rest of your AI stack, in the same country.
A GPU on its own is not a platform. Your data, applications and networks sit next to it, on one account and one invoice.
Block storage
NVMe volumes for datasets, checkpoints and model weights, with an encrypted option.
Object storage
S3-compatible buckets for training data, model files and backups.
Managed Kubernetes
Run the APIs, queues and apps around your models next to your GPU servers.
Backups and snapshots
Snapshots of your servers and volumes, kept in Rwanda.
Private networking
Private networks, security groups, load balancers, VPN and direct links to your offices.
Key management
Encryption keys, certificates and secrets held in the same region.
One-click apps that pair with a GPU server
Vector search, retrieval, pipelines and dashboards run on ordinary cloud servers beside your GPU, deployed pre-configured in minutes.
From account to running GPU.
Choose a plan
By the hour for experiments, by the month or on a 12-month plan for production. AMD Instinct or more GPUs per server on request.
Verify and fund
A verified account and prepaid credit, the same as every service on the platform. Pay in francs, by mobile money, card or bank transfer.
Launch and run
Your GPU server appears in your console and API, with its storage, backups and networks in Rwanda. Questions get an answer from the team in Kigali, usually within minutes to a few hours.
AI workloads that belong in Rwanda.
Under Rwanda’s data protection law, financial and health data are sensitive personal data. Processing them on a GPU in another country is a cross-border transfer that needs authorisation. Processing them in Kigali is not.
Banks, lenders and insurers
Run models over customer files, loan documents and claims without moving sensitive data out of the country.
Health and the public sector
Process records that must stay in Rwanda, on infrastructure under Rwandan law.
Software companies and AI startups
Train, fine-tune and serve the AI features of your product close to your users in East Africa.
Universities and research
GPU time for training runs, experiments and theses, billed by the hour.
Your models, your data, in writing.
GPU servers, storage and backups stay in Rwanda, in the region you choose for them. Residency goes in the contract.
Your GPUs are dedicated to you and never shared with other customers. We do not access your data or models without your permission.
Disks can be encrypted, with keys held in a managed key service in the same region rather than on the server.
Gewape Cloud Rwanda is registered with the National Cyber Security Authority (NCSA) as a data controller and a data processor, and operates under Rwandan law.
Three questions to ask any AI infrastructure provider.
- In which country are my data, my models and my backups stored and processed?
- Who else can reach the GPU and the disks my workload runs on?
- Can I see the price before I sign?
Our answers are on this page. Ask the same three of anyone else, and ask for them in writing.
Questions about GPU cloud in Rwanda
Can I rent an NVIDIA or AMD GPU server in Rwanda?
Yes. NVIDIA L40 48 GB GPU servers run in our Kigali region, at published prices in Rwandan francs. AMD Instinct GPUs are available on request. You run them like any other cloud server, by the hour, by the month or on a 12-month plan.
How much does a GPU server cost in Rwanda?
From RWF 3,250 an hour on demand, or RWF 1,697,100 a month on a 12-month plan, for one NVIDIA L40 48 GB with 16 vCPU and 96 GB of memory. Prices exclude VAT.
Where is my data processed?
In Kigali, Rwanda. Your GPU server, its storage and its backups stay in Rwanda. Gewape Cloud Rwanda is registered with the National Cyber Security Authority (NCSA) as a data controller and a data processor.
Is the GPU shared with other customers?
No. Every GPU server comes with dedicated GPUs that are yours alone for as long as you run it, with full GPU performance.
Which models can I run?
Any model you are licensed to run. The L40's 48 GB of GPU memory runs open models in the 30-billion-parameter class on a single card and supports FP8. Ready-to-run images come with NVIDIA CUDA or AMD ROCm installed, so PyTorch, vLLM, Ollama and Hugging Face tools work as they do anywhere else.
Can Gewape Cloud see my data or my models?
No. The GPUs are dedicated to you, disks can be encrypted with keys held in the region, and we do not access customer data without your permission.
How are GPU servers billed?
On-demand use is billed by the hour from your prepaid wallet. Monthly and 12-month plans reserve the GPU for you, so it stays yours even when the server is stopped. Launching a GPU server needs a verified account and prepaid credit, as every service on the platform does.
What happens during maintenance?
Planned maintenance is announced at least 72 hours ahead, and the window is agreed with customers on monthly and 12-month plans.
Tell us what you want to run.
The model or workload, how many GPUs, and for how long. We will help you choose the right plan, and if a CPU server is enough for what you need, we will say so.
Run your AI workloads in Rwanda.
Dedicated NVIDIA and AMD GPUs, next to your data, at prices you can see before you sign.