Cloud TPU

GPU resources for Kubernetes workloads with NVIDIA GeForce RTX 4080 Super (16 GB) or L40S (48 GB) cards, billed at a fixed monthly fee with free set-up on six-month commitments; data kept in Aruba's European data centres.

NVIDIA GPU instances and GPU-backed Kubernetes nodes (A100, L40S, H100, H200, B200, B300) in the UK, billed hourly on demand or with 6–36 month commitments

NVIDIA GPU capacity delivered on dedicated Bare Metal servers (L40S and H100 configurations) with MIG partitioning, usable from IaaS, OpenShift and Managed Kubernetes, for sovereign-FR AI training and inference

Virtual machines with single or 8-GPU NVIDIA and AMD configurations for AI training and inference, available on-demand, as spot capacity or on contract

Dedicated single-tenant 8-GPU servers for large-scale AI training and inference, provisioned on contract

NVIDIA GPU instances with 1 to 8 cards (A30, V100, A40, A5000, RTX 3080 Ti, RTX Pro 6000 and B300) for AI/ML training, inference and HPC, billed per second in selected zones

GPU compute instances on NVIDIA H100, H200, and A100 hardware for AI training and inference workloads, available as bare metal or virtualised, with hourly billing and reserved capacity options

End-to-end AI development platform with AutoML, data labelling, distributed training on Ascend and GPU clusters, and one-click deployment to cloud or edge

Virtual machines with dedicated NVIDIA H200 GPUs (PCIe passthrough) in fixed S, M, L and XL templates for AI training, inference and HPC, available in Frankfurt (de/fra/2) and Worcester (gb/bhx).

Deploys built-in catalogue or custom AI models to dedicated GPU-backed inference endpoints called via API key; available to pre-approved customers

NVIDIA A30, A100 (80 GB), H100 (80 GB) and RTX 6000 Blackwell (96 GB) GPU VMs in Amsterdam for AI, ML and rendering, billed hourly with the VM included and discounts for commitments; deployable via OpenStack or Gardener-managed Kubernetes.

Sales-led, reserved bare-metal NVIDIA GPU capacity (e.g. GB300 NVL72 racks and B200 hosts) ring-fenced per project and region on InfiniBand/NVLink fabrics, delivered via Nscale Kubernetes Service, Managed Slurm or bare metal for large-scale training and inference.

Managed Slurm batch scheduler running on Kubernetes, providing HPC-style scheduled queues for large-scale AI training and mixed workloads on Nscale GPU capacity.

Flexible GPUs (fGPUs) allocated and attached to VMs (up to two of the same model), with NVIDIA models from K2, P6, P100 and V100 to A100, L40, H100 and H200 depending on Region, for AI/ML training and inference

Public Cloud GPU instances with NVIDIA GPUs, including H200, H100, L40S, L4, V100S and Quadro RTX 5000, plus NVMe storage and up to 25 Gbps of public and private networking. Used for AI training and inference, video processing and rendering.

Virtual machines with NVIDIA L4, L40S, H100 (PCIe and SXM) and B300-SXM GPUs for AI training, inference and rendering, billed per minute of uptime in Paris and Warsaw zones

Cloud servers with NVIDIA GPUs for AI/ML training and inference, rendering and media workloads

On-demand NVIDIA GPU instances (H100, H200, A100, L40S, A40, A16) with hourly billing for AI training and inference workloads, with reserved capacity available via the Vultr sales team