Cloud GPU Providers

Runpod is a US neocloud offering NVIDIA and AMD GPUs across their own managed cloud and their community cloud, which gives access to their partners' infrastructure. They also offer serverless GPUs and managed inference endpoints.

Nebius is a Netherlands-based cloud provider with GPU instances, CPU-only instances, and large-scale NVIDIA GPU clusters across Europe and the US, as well as managed inference services. It is a public company, and is a constituent of the Nasdaq-100 index.

Crusoe is a US neocloud that builds data centres offering the latest NVIDIA and AMD GPUs — including the GB200 and MI355X. In addition to GPU instances, Crusoe also offers serverless fine-tuning and managed inference.

Vultr is a US cloud provider with a large global footprint, offering bare metal and virtual machine instances with NVIDIA and AMD GPUs alongside a full stack of compute, storage, networking and managed database services.

Jarvislabs is an India-based NVIDIA GPU cloud aimed at machine learning developers.

Vast.ai is a US-based two-sided GPU marketplace that aggregates capacity from hundreds of hosts. Prices fluctuate based on supply and demand. Details on this page reflect Vast's Secure Cloud offering.
For information on their Community Cloud offering, which lists many more GPU models and locations, see Vast's website or our price comparison page with "include P2P" enabled.

Sesterce is a French neocloud building NVIDIA GPU clusters and serverless inference, with data centres across Europe, North America, and Asia.

Nova is a cost-focused Canadian neocloud, offering consumer and workstation NVIDIA GPUs — including the RTX 5090 and RTX PRO 6000.

Lambda is a US NVIDIA GPU cloud focused on deep-learning workloads, offering on-demand GPU instances and reserved GPU clusters across its data centres in the US and Canada.

Together AI is a US neocloud focused on infrastructure for open-weights models, offering NVIDIA GPU clusters alongside managed inference and fine-tuning APIs.

Verda is a Finnish neocloud running on 100% renewable energy, offering the latest NVIDIA GPUs from data centres across Europe.

Scaleway is a French cloud provider offering EU-based NVIDIA GPU instances and clusters, alongside a full European cloud platform.

CoreWeave is a US-based specialist NVIDIA GPU cloud which counts among its clients some of the leading LLM labs, including OpenAI, Cohere, and Mistral. It offers a broad range of NVIDIA data-center GPUs — from the A100 and H100 to the latest GB200 and GB300 — aimed at large-scale training and inference. It is public company, and a constituent of the Nasdaq-100 index.

Latitude.sh is a global cloud provider offering NVIDIA H100 and RTX PRO 6000 GPUs in bare-metal instances or virtual machines across data centres worldwide. It is headquartered in São Paulo, Brazil, and is part of Megaport, an Australian publicly-traded Network-as-a-Service company.

Hyperstack is a UK neocloud offering on-demand and reserved NVIDIA GPUs from data centres across Europe and North America.

Denvr Dataworks is a North American AI cloud headquartered in Calgary, Alberta, offering on-demand and reserved NVIDIA GPUs and Intel Gaudi accelerators, as well as CPU-only instances.

Leafcloud is Dutch neocloud provider offering NVIDIA GPUs from small Amsterdam-based data centres that repurpose their server-generated-heat for residential heating and hot water.

Voltage Park is a US neocloud offering competitively-priced NVIDIA H100 GPUs, on-demand or with long-term reservations.

Oracle Cloud Infrastructure (OCI) is a hyperscaler offering NVIDIA and AMD GPU clusters.

Seeweb is an Italian cloud provider offering NVIDIA and AMD GPU cloud servers from data centres in Italy, in addition to other cloud services.

LeaderGPU is a Dutch GPU cloud offering dedicated, bare-metal NVIDIA GPU servers hosted in the Netherlands.

Massed Compute is a US neocloud offering on-demand virtual and bare-metal NVIDIA GPUs.

Cirrascale is a US deep-learning cloud offering dedicated nodes with multiple NVIDIA, AMD or Qualcomm accelerators attached. Commitment periods range from one month to one year.

AceCloud is an India-based cloud provider offering on-demand and spot NVIDIA GPUs in data centres in the US and India. AceCloud is part of the Real Time Data Services (RTDS) group of companies.

Hot Aisle is a US neocloud specialising in AMD GPUs, offering bare-metal and virtual MI300X systems. MI355X instances are also available by request.

Hetzner is a cost-focused German hosting provider, with a small range of NVIDIA GPU offerings.

Beyond.pl is a Polish data-centre operator and cloud provider offering NVIDIA B200 GPUs from their data centres in Poznań, Poland.

Amazon Web Services (AWS) is the largest and oldest hyperscaler cloud, offering NVIDIA GPU instances alongside Qualcomm accelerators, and its own custom Trainium and Inferentia accelerators.

Microsoft Azure is a hyperscaler offering NVIDIA GPU virtual machines.

Google Cloud Platform (GCP) is a hyperscaler offering NVIDIA GPU instances alongside its own TPU machine learning accelerators.

DigitalOcean is a US developer-focused cloud offering virtual GPU Droplets powered by NVIDIA and AMD, as well as bare-metal GPU instances.

OVHcloud is French cloud provider, offering NVIDIA GPU instances and bare-metal servers in data centres across Europe and North America, alongside a broad offering of other cloud products.

Atlantic.Net is a long-established US cloud and hosting provider offering dedicated and virtual NVIDIA GPUs.