Scaleway Pricing & Developer Hub

Scaleway European Cloud Architecture: Sovereign GPUs & Data Privacy Economics

Scaleway is France's leading sovereign cloud provider, operating Tier-3 and Tier-4 datacenters (notably DC5 in Paris, powered by 100% renewable energy and adiabatical cooling with zero water waste). For European engineering teams subject to strict GDPR, NIS2, and EU AI Act regulations, Scaleway provides a legally insulated alternative to US hyperscalers. The platform offers dedicated GPU instances—ranging from entry-level Nvidia L4 (24GB) at €0.68/hour to 8x Nvidia H100 SXM5 nodes at €2.80/hour per GPU—with zero exposure to the US CLOUD Act.

Scaleway Production GPU & Compute Rate Card (EUR & USD Equivalent)

Scaleway Terraform GPU Instance Deployment

terraform {
  required_providers {
    scaleway = { source = "scaleway/scaleway", version = ">= 2.30.0" }
  }
}

resource "scaleway_instance_ip" "public_ip" {}

resource "scaleway_instance_server" "ai_node" {
  type        = "RENDER-S" # 1x Nvidia L4 24GB
  image       = "ubuntu_noble"
  name        = "production-vllm-node"
  zone        = "fr-par-2"
  ip_id       = scaleway_instance_ip.public_ip.id
  enable_ipv6 = true

  root_volume {
    size_in_gb = 100
    volume_type = "b_ssd"
  }
}

Scaleway SRE Production Guardrails

• Block Storage IOPS Quotas: Scaleway offers standard and high-speed Block Storage (`b_ssd`). When hosting large LLM weights (70B model requires 75GB+), ensure volumes are provisioned with at least 5,000 IOPS to avoid slow model weight load times during node restarts.
• Private Network Interconnect (VPC): Scaleway provides free, unmetered VPC connections between Compute instances and Managed Kubernetes (Kapsule). Routing database traffic across Scaleway private networks eliminates public exposure and guarantees zero bandwidth fees.

Engineering Q&A: Scaleway Pricing & Compliance

Scaleway Kubernetes (Kapsule) Auto-Healing & Spot Eviction Protocol

Running distributed inference workloads on Scaleway Managed Kubernetes (Kapsule) requires configuring node auto-healing with custom health probes. While Scaleway bare-metal and GPU instances provide high reliability, spot GPU nodes can be preempted with a 120-second warning signal. By integrating Kubernetes Cluster Autoscaler with custom pod termination lifecycles (`preStop` hooks), inference pods automatically drain in-flight WebSocket connections and checkpoint model states before node eviction occurs.

Advanced SRE Checklist: Scaleway Deployment

• Enforce Block Storage NVMe Over-Provisioning: Scaleway b_ssd volumes dynamically scale IOPS with volume capacity. Sizing a root or cache disk at 250GB guarantees 5,000 baseline IOPS, preventing model weight loading bottlenecks during cold restarts.
• Private Network Subnet Segmentation: Deploy GPU nodes and managed databases across isolated Scaleway Private Networks (VPC). Restrict public internet exposure strictly to Scaleway Load Balancer endpoints running SSL termination.

Scaleway Elastic Metal Provisioning & Cloud-Init Bootstrap

For workloads requiring direct bare-metal hardware access without virtualization hypervisors, Scaleway Elastic Metal servers boot in under 15 minutes. Using standard cloud-init automation templates, engineering teams can pre-install Nvidia CUDA 12.4 drivers, Docker CE, and Kubernetes kubelet agents automatically upon initial server spin-up. Elastic Metal instances communicate over private 10 Gbps physical network interfaces, providing dedicated physical hardware isolation for high-security enterprise banking, sovereign healthcare, and defense sector deployments.