Provider category
Neocloud
Built to sell GPU compute at scale: bare-metal clusters, managed Slurm or Kubernetes, sometimes inference. Usually NVIDIA-centric and financed by GPU-backed debt and long compute contracts.
Source: https://neocloud.fyi/sources/ Checked 2026-09-18
site's own definition
| Name | HQ | What it sells | European footprint |
|---|---|---|---|
| CoreWeave | US | Bare-metal GPU clusters, managed Kubernetes, inference | Barcelona and Bilbao sites leased from Merlin Edged; Crawley and London |
| Nebius | Netherlands | Full-stack AI cloud, bare metal and managed | Finland (Mantsala), Iceland, UK (Ark Longcross), Spain (Getafe 18 MW) |
| Lambda | US | On-demand and reserved GPU cloud, 1-Click Clusters | None verified |
| Crusoe | US | GPU clusters, modular data centres, own power generation | Norway (hydro, 12 to 52 MW), Iceland |
| Together AI | US | Serverless inference, fine-tuning, dedicated clusters | None owned |
| Fluidstack | US (moved from UK) | Bare-metal clusters; builds campuses for named customers | UK presence reduced; withdrew from a 1 GW French project in March 2026 |
| Voltage Park (Lightning AI) | US | On-demand and reserved clusters plus the Lightning platform | None confirmed |
| Vultr | US | VMs, bare-metal GPU, managed Kubernetes | Amsterdam, Frankfurt, London, Paris and others |
| TensorWave | US | GPU cloud on AMD Instinct only | None found |
| Firmus | Australia | GPU cloud and campus development in Asia-Pacific | None |