Scaleway GPU Instances provide on-demand GPU compute for AI training, fine-tuning, inference, video, and rendering workloads. Teams can select among GPU families and provision instances through the console, CLI, or Terraform.
Scaleway GPU Instances are cloud virtual machines with GPU acceleration for parallel workloads. They provide on-demand compute for AI model training, fine-tuning, and inference, as well as video, graphics, visual computing, and rendering. The product page lists NVIDIA B300-SXM, H100-SXM, H100 PCIe, L4, and L40S families so teams can choose hardware suited to different workload profiles.
Instances can be launched through the Scaleway console, CLI, or Terraform and billed per minute on a pay-as-you-go basis. Scaleway describes European data residency and data transfer without egress charges. A cost estimator supports configuring an instance, volume, and flexible IPs; the supplied information does not state specific GPU rates.
Choose among B300-SXM, H100-SXM, H100 PCIe, L4, and L40S. The page describes B300-SXM for high-VRAM complex reasoning, H100-SXM for multi-GPU scaling, H100 PCIe for versatile compute and fine-tuning, L4 for efficient low-latency acceleration, and L40S for combined AI, graphics, and media workloads.
Create GPU instances through the console, CLI, or Terraform. The FAQ describes per-minute pay-as-you-go billing, with resources available to provision and delete as workloads change.
GPU OS 13 is based on Ubuntu Noble and includes the NVIDIA driver, Docker, and NVIDIA Container Toolkit. The image is designed to work with NVIDIA NGC container images.
GPU instances can be used with Scaleway Kapsule managed Kubernetes. The NVIDIA GPU Operator handles GPU setup for the cluster.
Scaleway states that moving data into or out of the infrastructure carries no egress charge. Its estimator lets users configure an instance, volume size, and flexible IPv4 addresses to estimate costs.
The product page says training datasets, models, and user data are hosted in Europe, and states that GPU Instances are HDS compliant with a 99.5% SLA.
Use GPU compute for model training and fine-tuning, including workloads that need the H100 PCIe's stated versatility or the B300-SXM's high-VRAM capabilities.
Deploy models for inference, including LLM inference and model-serving workloads. The L4 is positioned for high-throughput, low-latency acceleration, while L40S is described for generative AI and LLM inference.
Attach GPU instances to a Kapsule managed Kubernetes environment and use the NVIDIA GPU Operator to manage GPU setup for a new pool.
Use L4 for video and graphics acceleration, or L40S for media workloads, 3D graphics, and rendering alongside AI tasks.
Start from GPU OS 13, which includes Docker and NVIDIA Container Toolkit, and deploy containerized workloads with GPU access enabled using Docker's --gpus flag.
The FAQ describes a pay-as-you-go model in which GPU usage is billed per minute. The supplied page does not give specific rates; use the pricing information or estimator for configuration-dependent costs.
Create a Scaleway console account with Owner status or the necessary IAM permissions, then follow the console steps to create a GPU Instance. Instances can also be launched through the CLI or Terraform.
GPU OS 13 is based on Ubuntu Noble and comes with the NVIDIA driver, Docker, and NVIDIA Container Toolkit. It is intended for GPU-accelerated applications and is designed to work with NVIDIA NGC container images.
Yes. The page supports use with Kapsule managed Kubernetes, where the NVIDIA GPU Operator manages GPU setup. For Docker containers, use NVIDIA Container Toolkit and launch the container with the --gpus flag.
Scaleway says all GPU Instances are HDS compliant and have a 99.5% SLA. The product page also says datasets, models, and user data are hosted in Europe.
cirrascale.com
Dedicated AI cloud compute, storage, networking, and deployment services for training and inference.
www.tensordock.com
TensorDock is a marketplace-based cloud infrastructure platform for on-demand GPU and CPU servers. It supports machine learning, rendering, cloud gaming, scientific computing, and other workloads that need configurable virtual machines.
verda.com
Verda is a full-stack AI cloud for running AI workloads on GPU infrastructure. It provides on-demand NVIDIA GPU instances and self-service multi-node clusters with managed Slurm or Kubernetes.
lepton.ai
NVIDIA DGX Cloud Lepton connects developers to a global network of GPU compute through a cloud-based offering. The available source positions it for developers who need access to accelerated computing resources.
