GPU clouds and bare metal
Clouds built mainly around accelerators, from single instances to dedicated clusters and whole servers.
Definition, not a sourced statistic
What it is
A GPU cloud is a provider whose main product is accelerator capacity. It may rent single virtual machines, multi-node clusters joined by a fast network, or bare metal servers where you get the whole physical machine with no shared virtualisation layer. Many add managed Kubernetes or Slurm for scheduling jobs across a cluster.
Who it suits
- Machine learning teams that want accelerator capacity without the wider catalogue of a general cloud.
- Training runs that need many accelerators on one fast network.
- Teams that want root access to whole machines for performance or isolation.
Common contract shapes
- On-demand instances. Self-serve start and stop, billed by time.
- Reserved or committed capacity. A term commitment for supply certainty, a rate, or both.
- Dedicated cluster contract. A set of machines and network set aside for you, usually after a quote, with a lead time.
- Bare metal rental. Whole physical servers rented by time or term, which you manage from the operating system up.
Providers use their own names for these shapes, so compare what is promised, not what it is called. The contract shapes guide explains each one.
What to ask
- Egress and data transfer. Is moving data out charged, and is storage traffic inside the cluster free?
- Service levels. What uptime is promised for nodes and for the network between them, and what happens when a node fails mid-run: replacement, credit or nothing?
- Data residency and access. In which country and which facility will the machines sit, and is that fixed in the contract?
- Security certifications. Which certifications does the provider hold, and do they cover the facility your cluster is in, which may belong to someone else?
- Export controls and sanctions. Has the provider checked export controls on the accelerators for your location and use, and will it ask you for end-use statements?
- Exit. How are disks wiped at the end, how do you get your data out, and can a reservation be cut short or passed to someone else?
- What network joins the machines, for example InfiniBand, and is it shared with other customers?
- Who owns the hardware: the provider, a partner data centre or a financier?
The full list is on the printable buyer's checklist. For rules on chips and where they can go, see export controls and compute thresholds.
On this site
Entries listed here because their own summary or checked facts on this site fit this type. An entry can fit more than one type, and a provider missing here may still offer it. This is a list, not a ranking, and in name order. We link to providers for prices.
Providers
- Cirrascale Cloud Services: A cloud provider offering private AI training and inference on several accelerator families, including AMD, NVIDIA, Qualcomm and Tenstorrent.
- Civo: A cloud provider built around managed Kubernetes, with GPU compute, private cloud and sovereign regions.
- CoreWeave: A cloud built around AI workloads, offering GPU and CPU compute, bare metal servers, storage and managed Kubernetes.
- Crusoe Cloud: An AI-focused cloud offering NVIDIA and AMD GPUs with high-speed networking, managed Kubernetes and managed Slurm.
- Cudo Compute: A provider that designs, deploys and operates GPU clusters and AI infrastructure. Its focus is dedicated clusters rather than single instances.
- Denvr Dataworks: An AI cloud that describes itself as Canadian, offering on-demand and reserved GPU compute, managed inference and shared NVMe storage, with data centres in Canada and the US.
- DigitalOcean GPU Droplets: GPU virtual machines from a developer-oriented cloud, with NVIDIA and AMD options from single-GPU to eight-GPU configurations.
- E2E Networks: An India-based cloud provider offering rentable NVIDIA GPUs, virtual machines, managed Kubernetes and AI model endpoints.
- Firmus Technologies: A company (Firmus Technologies Pty Ltd) that designs and operates modular AI infrastructure it calls AI Factories, and offers GPU cloud services and custom deployments from them.
- Gcore: A provider combining cloud, edge network, security and AI services. Its GPU cloud covers several NVIDIA GPU generations, with or without InfiniBand networking.
- Genesis Cloud: A cloud provider offering GPU and CPU virtual machines with attachable storage volumes. It describes its infrastructure as run on renewable energy.
- Hetzner: A German hosting company offering virtual cloud servers, load balancers and networking, plus separate dedicated servers. GPU hardware is listed under its dedicated server range rather than the cloud product.
- Hot Aisle: A cloud offering AMD Instinct GPU compute as virtual machines, dedicated nodes and bare-metal clusters.
- Hyperstack (NexGen Cloud): An on-demand GPU cloud run under the NexGen Cloud group. It provides GPU virtual machines, object storage and managed Kubernetes.
- Jarvislabs: An India-based GPU cloud offering virtual machines, clusters, model APIs and managed inference endpoints.
- Lambda: A GPU cloud with single on-demand instances, pre-configured multi-node clusters and larger dedicated deployments.
- Latitude.sh: A bare metal and GPU cloud provider that owns and operates servers and network systems in its own locations, with Kubernetes, virtual machines and storage.
- Massed Compute: A GPU rental service with on-demand NVIDIA instances, bare metal servers and InfiniBand clusters, reachable by REST API.
- Mistral Compute: An AI cloud from Mistral AI that describes European-hosted GPU capacity with orchestration and model-building tools.
- Nebius: An AI cloud offering GPU compute with InfiniBand networking, plus managed inference and MLOps tooling.
- Nscale: A full-stack AI cloud offering inference endpoints, managed Slurm, Kubernetes, virtual machines and on-demand GPU compute, run from its own and partner data centres.
- Oracle Cloud Infrastructure: Oracle's cloud compute service, including bare metal servers and GPU instances with both NVIDIA and AMD accelerators.
- OVHcloud Public Cloud GPU: GPU instances on the public cloud of a large European provider, covering AI inference and training as well as rendering.
- Scaleway: A European cloud provider offering GPU instances for training, inference and graphics work.
- Sesterce: A provider that calls itself a European AI factory company (a company claim), with sites in France, a GPU cloud, a cluster control plane and serverless inference.
- Taiga Cloud (Northern Data Group): A European GPU cloud from Northern Data Group offering bare metal and GPU as a service on NVIDIA H100 and H200 hardware.
- TensorWave: A cloud provider built around AMD Instinct accelerators. It offers bare metal GPU access with managed Kubernetes and Slurm options.
- Verda: A cloud platform for AI workloads, offering GPU instances, InfiniBand clusters and serverless GPU containers. Storage options include block storage and a shared filesystem.
- Voltage Park: A GPU cloud offering NVIDIA H100, B200 and GB300 capacity, either on demand or under longer reserved contracts.
- Vultr: A general cloud provider with virtual machines, Kubernetes, storage and a range of GPU instances, spread over many regions.
Related
- Glossary: GPU cloud (neocloud), Bare metal, InfiniBand, On-demand, Reserved, SLA (service level agreement).
- Guide: How to get GPUs this week
- Guide: What is a capacity block, and the other contract shapes
- Provider directory
- Compare offer types
Other offer types: Hyperscale cloud instances · Marketplaces and aggregators · Decentralised and community networks · Regional and sovereign clouds · Managed inference and serverless GPU · Chip-maker and accelerator clouds · National and institutional HPC, and free or funded programmes · Colocation, wholesale, powered shell and build-to-suit · On-premises hardware, and leasing or financing it · Capacity brokers, reserved contracts, secondary and spot markets · Edge and telecom sites, and converted sites
Definitions and questions written in our own words, checked on 3 October 2026 and due for review by 3 January 2027. Sourced facts stay on the provider, programme and policy pages this page links to. This is general information to help you ask better questions. It is not legal, tax or procurement advice. For export controls, sanctions and contracts, check the primary rules and talk to a qualified adviser. Something wrong? Tell us.