ComputeFinder

Compute providers and access routes

Compare 86 recorded ways to get compute, from hyperscalers to decentralised networks and public programmes.

Compare GPU models, locations and price bases

Start with the decision you need to make

No provider pays to be listed or to be placed higher. Selected GPU comparisons include dated price information where the source use permits it. For the wider directory, follow the official pricing links.

A listing does not establish current availability, suitability, price or procurement approval. Missing providers are a coverage gap, not evidence that no other route exists.

Hyperscalers

Hyperscaler

Amazon Web Services (EC2)

The virtual server service of a very large general-purpose cloud. It covers a wide catalogue of instance types, including accelerated options for machine learning.

Recorded workload fit: Teams that already run on AWS or need compute alongside a broad set of managed cloud services.

Recorded access route: Self-serve account signup, with sales contact available for larger commitments.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Hyperscaler

Microsoft Azure

Microsoft's cloud virtual machine service, with GPU-enabled and HPC-oriented machine types alongside general-purpose ones.

Recorded workload fit: Organisations already using Microsoft tooling or needing GPU capacity inside an enterprise cloud account.

Recorded access route: Self-serve account signup, with enterprise agreements and sales contact for larger needs.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Hyperscaler

Google Cloud (Compute Engine)

Google's virtual machine service. Besides GPUs, it is the place to reach Google's own Cloud TPU accelerators for machine learning.

Recorded workload fit: Teams that want TPU access or already build on Google Cloud data and AI services.

Recorded access route: General Compute Engine access includes self-serve signup, but A5X has a separate route. Checked 14 November 2026: Google directs A5X users to their account team to reserve capacity and obtain instance or cluster instructions. A5X is bare metal, not Spot or Flex-start. Confirm region, quota, delivery and terms directly; documentation is not proof of spare capacity or general availability.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Hyperscaler

Oracle Cloud Infrastructure

Oracle's cloud compute service, including bare metal servers and GPU instances with both NVIDIA and AMD accelerators.

Recorded workload fit: Teams wanting bare metal GPU access from a large cloud, or already running Oracle workloads.

Recorded access route: Self-serve account signup, with sales contact for larger clusters.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Hyperscaler

IBM Cloud

IBM's public cloud, offering GPU and AI accelerator options including NVIDIA GPUs, Intel Gaudi 3 and AMD Instinct MI300X, with bare metal and OpenShift integration.

Recorded workload fit: Enterprises already using IBM, Red Hat OpenShift or watsonx that want GPU capacity in the same environment.

Recorded access route: Through the IBM Cloud catalogue. The GPU page does not itself show pricing, and notes spot pricing on select GPUs.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Hyperscaler

Huawei Cloud

A global cloud platform with virtual servers, GPU-accelerated servers, the ModelArts machine learning platform and the Pangu family of large models.

Recorded workload fit: Organisations wanting a cloud with regions in China and other countries, with GPU servers and model services.

Recorded access route: Self-serve international site; availability and terms vary by region.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Hyperscaler

Alibaba Cloud

A global cloud platform with elastic virtual servers including GPU instances, and the Model Studio AI service.

Recorded workload fit: Organisations wanting a cloud with regions in China and other countries, including GPU instances for AI training.

Recorded access route: Self-serve through its international site, with several purchasing models listed on the ECS page.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Hyperscaler

NAVER Cloud Platform

A South Korean cloud platform offering infrastructure, platform and software services for corporations, with GPU servers and the CLOVA AI services.

Recorded workload fit: Organisations wanting cloud and AI services from a Korean provider, including public sector and regulated sectors.

Recorded access route: Self-serve console; the site is mainly in Korean.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Hyperscaler

Tencent Cloud

Tencent Cloud is a general-purpose cloud route with virtual machines, GPU compute, dedicated hosts and bare-metal products.

Recorded workload fit: Teams comparing an international cloud account with general-purpose and accelerated compute; confirm registration, identity, pricing, regional product availability and allocation directly.

Recorded access route: Signup documentation lists email, Google and WeChat routes. Registration location cannot later be changed and affects tax, phone and identity handling; identity documents must match that jurisdiction, and applicants using mainland-China identity documents are directed to Tencent Cloud China. Supported customer countries, successful identity verification, payment acceptance, quota, current capacity, product availability by region and successful provisioning were not verified.

Pricing not assessed

Record checked 23 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU clouds

GPU cloud

CoreWeave

A cloud built around AI workloads, offering GPU and CPU compute, bare metal servers, storage and managed Kubernetes.

Recorded workload fit: AI labs and companies running large training or serving workloads who want a specialist GPU cloud.

Recorded access route: Account signup and sales contact; larger capacity is typically arranged directly with the provider.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Lambda

A GPU cloud with single on-demand instances, pre-configured multi-node clusters and larger dedicated deployments.

Recorded workload fit: Machine learning teams that want to start with a single GPU instance and grow to clusters.

Recorded access route: Self-serve signup for on-demand instances; sales contact for clusters and dedicated capacity.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Crusoe Cloud

An AI-focused cloud offering NVIDIA and AMD GPUs with high-speed networking, managed Kubernetes and managed Slurm.

Recorded workload fit: Teams running distributed training or inference who want managed orchestration on dedicated GPU infrastructure.

Recorded access route: Console signup for self-serve use; sales contact for larger deployments.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Nebius

An AI cloud offering GPU compute with InfiniBand networking, plus managed inference and MLOps tooling.

Recorded workload fit: AI teams, including those in Europe, that want GPU clusters and managed inference from one provider.

Recorded access route: Self-serve signup, with sales contact for reserved or large-scale capacity.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Vultr

A general cloud provider with virtual machines, Kubernetes, storage and a range of GPU instances, spread over many regions.

Recorded workload fit: Developers who want GPU capacity alongside ordinary cloud services in a choice of locations.

Recorded access route: Self-serve account signup.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

DigitalOcean GPU Droplets

GPU virtual machines from a developer-oriented cloud, with NVIDIA and AMD options from single-GPU to eight-GPU configurations.

Recorded workload fit: Developers and small teams who want straightforward GPU servers within a familiar cloud console.

Recorded access route: Self-serve account signup; provision from the console or API.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Together AI

A platform for running open models through an inference API, with fine-tuning and dedicated GPU clusters for larger jobs.

Recorded workload fit: Teams that want to call open models by API first, and move to dedicated GPUs or fine-tuning later.

Recorded access route: Self-serve API signup; sales contact for reserved capacity and large clusters.

Token-based pricing; check which token and billing unit applies

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Scaleway

A European cloud provider offering GPU instances for training, inference and graphics work.

Recorded workload fit: Teams that want GPU compute hosted in Europe under European jurisdiction.

Recorded access route: Self-serve account signup.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

OVHcloud Public Cloud GPU

GPU instances on the public cloud of a large European provider, covering AI inference and training as well as rendering.

Recorded workload fit: Teams wanting GPU instances from a European provider without committing to a large cluster.

Recorded access route: Self-serve account signup; instances are managed through a control panel, API or CLI.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

E2E Networks

An India-based cloud provider offering rentable NVIDIA GPUs, virtual machines, managed Kubernetes and AI model endpoints.

Recorded workload fit: Teams in India, or those needing Indian-hosted compute, who want GPU access from a domestic provider.

Recorded access route: Self-serve account signup; enterprise contact available.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Genesis Cloud

A cloud provider offering GPU and CPU virtual machines with attachable storage volumes. It describes its infrastructure as run on renewable energy.

Recorded workload fit: Developers and small teams who want self-serve GPU instances with a public API.

Recorded access route: Self-serve signup, with a web dashboard, a public API and SSH access to instances.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Gcore

A provider combining cloud, edge network, security and AI services. Its GPU cloud covers several NVIDIA GPU generations, with or without InfiniBand networking.

Recorded workload fit: Teams that want GPU instances alongside CDN, edge and security services from one vendor.

Recorded access route: Self-serve signup, with a web control panel, API and Terraform; sales contact for the newest large configurations.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Cudo Compute

A provider that designs, deploys and operates GPU clusters and AI infrastructure. Its focus is dedicated clusters rather than single instances.

Recorded workload fit: Organisations planning dedicated GPU clusters who are comfortable with a quote-based process.

Recorded access route: Contact the company through its website; the pricing page shows GPU products as quote on request.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Hyperstack (NexGen Cloud)

An on-demand GPU cloud run under the NexGen Cloud group. It provides GPU virtual machines, object storage and managed Kubernetes.

Recorded workload fit: Teams wanting self-serve GPU virtual machines with API and Terraform tooling.

Recorded access route: Self-serve through a web console, with an API (Go and Python SDKs) and a Terraform provider.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

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.

Recorded workload fit: AI teams wanting self-serve GPU capacity with European hosting and stated compliance certifications.

Recorded access route: Self-serve signup through its web console.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

TensorWave

A cloud provider built around AMD Instinct accelerators. It offers bare metal GPU access with managed Kubernetes and Slurm options.

Recorded workload fit: Teams who want to run workloads on AMD GPUs rather than NVIDIA hardware.

Recorded access route: Request a proof of concept or speak to the sales team through its website.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Voltage Park

A GPU cloud offering NVIDIA H100, B200 and GB300 capacity, either on demand or under longer reserved contracts.

Recorded workload fit: Teams that want either quick self-serve GPU access or a reserved block of capacity.

Recorded access route: Self-serve through its dashboard for on-demand use, and a contact form for reserved capacity.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

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.

Recorded workload fit: Developers and small teams who want general-purpose European cloud servers or dedicated GPU servers.

Recorded access route: Self-serve signup, with a web console, REST API, command line tool and Terraform support.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

STACKIT

A European cloud platform with compute, storage, Kubernetes, managed databases and AI services. It positions itself around data hosted in Europe and digital sovereignty.

Recorded workload fit: European organisations that need cloud services hosted and operated within Europe.

Recorded access route: Accessed through the STACKIT Portal; colocation and consulting are priced individually.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

IONOS Cloud

A cloud platform offering virtual machines, bare metal, managed Kubernetes, storage and databases. It also lists GPU servers and a hosted AI model service.

Recorded workload fit: Teams wanting a general cloud with data centres in Europe and the US, with the option of GPU servers.

Recorded access route: Self-serve through the Cloud Panel, Data Center Designer, API and SDKs, or a Terraform provider.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Exoscale

A European cloud provider offering compute instances, managed Kubernetes, storage, databases and GPU servers across several European zones.

Recorded workload fit: Teams that want European cloud zones with stated security certifications.

Recorded access route: Self-serve through a web portal, API, command line tool and Terraform provider.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

UpCloud

A Finnish cloud provider offering cloud servers, GPU servers, managed databases and Kubernetes, with data centres in Europe, North America and Asia Pacific.

Recorded workload fit: Developers and businesses wanting a general cloud with locations across several continents and GPU options.

Recorded access route: Self-serve through a control panel, API, command line tools and a Terraform provider.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Yotta (Shakti Cloud)

An Indian data centre operator whose Shakti Cloud offers GPU virtual machines, bare metal and serverless GPU compute hosted in India.

Recorded workload fit: Organisations that need GPU capacity hosted within India.

Recorded access route: Pricing is obtained by contacting the company through its website.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Sakura Internet

A Japanese hosting and cloud company offering Sakura Cloud, rental and dedicated servers, and GPU services under the Koukaryoku name. Services are hosted in Japan.

Recorded workload fit: Organisations that need GPU or cloud capacity hosted in Japan under Japanese law.

Recorded access route: Self-serve through a web control panel, REST API and command line tools; service pages are mainly in Japanese.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Core42

A UAE-based provider of sovereign AI infrastructure for governments and enterprises. It offers a sovereign AI cloud, a sovereign public cloud and a private cloud.

Recorded workload fit: Governments and large organisations needing in-country or sovereign AI infrastructure in the UAE.

Recorded access route: Engagement is through its website and sales process; no public price list was found.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Fluidstack

A company building and operating large AI data centres in the USA. Its homepage describes gigawatt-scale sites, closed-loop cooling and off-grid solar power.

Recorded workload fit: Large AI developers seeking dedicated, purpose-built capacity through direct engagement.

Recorded access route: No self-serve signup or public price list was visible on the homepage; engagement appears to be by direct contact.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

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.

Recorded workload fit: Teams wanting managed training or inference on GPU capacity in Europe or the US, or reserved GPUs through sales.

Recorded access route: A web console is available, and GPUs can be reserved by contacting sales. No public pricing page is linked.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Civo

A cloud provider built around managed Kubernetes, with GPU compute, private cloud and sovereign regions.

Recorded workload fit: Developers and teams wanting a Kubernetes-first cloud with GPU options and regional sovereignty.

Recorded access route: Self-service signup through a web dashboard, with a guided demo and a startup programme also offered.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Akamai Cloud (Linode)

A distributed public cloud built into Akamai's edge network, with compute, storage, Kubernetes, serverless and GPU instances for AI inference.

Recorded workload fit: Teams running inference close to users, or wanting a general cloud with GPU options and a developer API.

Recorded access route: Free account signup, a web Cloud Manager, APIs and SDKs. Some newer GPUs are available by request.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

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.

Recorded workload fit: Teams wanting dedicated bare metal or GPU servers with infrastructure-as-code tooling.

Recorded access route: Provisioning through Terraform, a CLI, and Go and Node SDKs, with API documentation.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Modal

A serverless platform for running AI inference, training, sandboxes and functions on GPUs and CPUs spread across multiple clouds.

Recorded workload fit: Developers who want to run code on GPUs without managing servers or quota requests.

Recorded access route: Self-serve signup and an in-browser playground. The pricing page describes usage-based billing while code runs.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Fireworks AI

A training and inference platform for open models, with serverless, on-demand and reserved deployment tiers.

Recorded workload fit: Teams serving or fine-tuning open models through an API, from experiments to reserved capacity.

Recorded access route: Web platform, an OpenAI and Anthropic compatible API, a CLI and an SDK. Serverless is billed per token.

Token-based pricing; check which token and billing unit applies

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

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.

Recorded workload fit: Teams needing GPU capacity with Canadian data residency, or on-premises AI infrastructure services.

Recorded access route: Self-service console, REST API, CLI, and Python and Terraform bindings.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

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.

Recorded workload fit: European teams wanting GPU access by CLI or API, or reserved capacity through sales.

Recorded access route: Cloud portal, CLI and API. The site names on-demand, reserved and spot billing modes, with sales contact for large reservations.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Khazna Data Centers

A Middle East data centre operator that designs, builds and operates facilities. It sells data centre capacity, not a self-serve GPU cloud.

Recorded workload fit: Organisations looking for hyperscale data centre capacity in the UAE and, as it expands, Saudi Arabia.

Recorded access route: Engagement is through the company directly.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

HUMAIN

A company that describes itself as a full-stack AI ecosystem, covering sovereign data centres, cloud infrastructure, generative AI models and applications.

Recorded workload fit: Organisations seeking sovereign AI infrastructure and services through direct engagement.

Recorded access route: The page read gave no self-serve route; engage the company directly.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Cloudflare Workers AI

A serverless service that runs open-source machine learning models on GPUs across Cloudflare's network, with no servers to manage.

Recorded workload fit: Developers adding inference to web apps who prefer serverless calls to renting GPU instances.

Recorded access route: Self-serve through Workers bindings, Pages or the Cloudflare API; usage-based.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

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.

Recorded workload fit: Organisations in Asia-Pacific wanting GPU clusters or bare metal and willing to enquire for terms.

Recorded access route: The site offers an enquiry route; no public price list was found on the homepage.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Mistral Compute

An AI cloud from Mistral AI that describes European-hosted GPU capacity with orchestration and model-building tools.

Recorded workload fit: Teams that want to keep data in the EU and use bare metal, managed clusters or a private AI studio.

Recorded access route: No public price list was found on the product page; access is through the company.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Baseten

An inference platform for deploying open-source and custom models, with managed cloud, self-hosted and hybrid options.

Recorded workload fit: Teams serving models in production who want dedicated deployments or model APIs.

Recorded access route: Self-serve basic plan; higher plans and self-hosting are arranged with the company. Billing is by compute time or by token.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Koyeb

A serverless platform for deploying containers and AI workloads on GPUs, CPUs and accelerators across distributed locations.

Recorded workload fit: Developers who want serverless deployment with autoscaling, including scale to zero.

Recorded access route: Self-serve; the pricing page describes pay-per-second billing. Corporate status: the homepage says Koyeb is joining Mistral AI (company statement).

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Cirrascale Cloud Services

A cloud provider offering private AI training and inference on several accelerator families, including AMD, NVIDIA, Qualcomm and Tenstorrent.

Recorded workload fit: Teams wanting dedicated multi-vendor accelerator servers on monthly terms.

Recorded access route: Pricing is per server per month with optional longer terms, as described on the pricing page.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

SambaNova

An AI hardware and inference company that runs its own RDU chips, sold as racks and offered as a cloud inference service.

Recorded workload fit: Teams wanting OpenAI-compatible inference APIs on non-GPU accelerators.

Recorded access route: Cloud access through SambaCloud; the pricing page text was not readable in full, so check it directly.

Token-based pricing; check which token and billing unit applies

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Cerebras

A maker of wafer-scale AI systems that also sells cloud inference, dedicated capacity and on-premises systems.

Recorded workload fit: Teams wanting inference on non-GPU hardware, dedicated or on-premises setups.

Recorded access route: Current access conditions are not recorded here. Check the linked official pricing page before choosing a route; no eligibility or availability is guaranteed.

Token-based pricing; check which token and billing unit applies

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Groq (GroqCloud)

An inference provider built on its own Language Processing Unit chips, offering GroqCloud for developers.

Recorded workload fit: Developers wanting usage-billed inference APIs for open models, including speech models.

Recorded access route: Current access conditions are not recorded here. Check the linked official models page and the provider's own site before choosing a route; no eligibility or availability is guaranteed.

Token-based pricing; check which token and billing unit applies

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Hot Aisle

A cloud offering AMD Instinct GPU compute as virtual machines, dedicated nodes and bare-metal clusters.

Recorded workload fit: Teams wanting AMD GPUs on usage-based billing.

Recorded access route: Launch through a terminal UI, API or CLI; billing is metered by the minute.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Massed Compute

A GPU rental service with on-demand NVIDIA instances, bare metal servers and InfiniBand clusters, reachable by REST API.

Recorded workload fit: Developers and teams wanting on-demand NVIDIA GPUs without contracts.

Recorded access route: Self-serve on demand; bare metal and commitments by contacting the company.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Jarvislabs

An India-based GPU cloud offering virtual machines, clusters, model APIs and managed inference endpoints.

Recorded workload fit: Developers in India wanting per-minute GPU billing with no commitment.

Recorded access route: Self-serve with a CLI and Python SDK. The pricing page describes on-demand, spot and reserved options.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

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.

Recorded workload fit: European teams wanting sovereign GPU capacity with managed Kubernetes or SLURM.

Recorded access route: The page mentions a cloud portal with per-minute billing; no pricing link was found on the page, so treat terms as by enquiry. The page says Northern Data is now part of Quake AI and that accounts, services and billing are unchanged (company statement).

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Applied Digital

A company that designs, builds and operates large data centre campuses for AI training and inference.

Recorded workload fit: Large tenants seeking leased AI data centre capacity at campus scale.

Recorded access route: Closest allowed type is gpu_cloud: this is a data centre developer and operator, not a self-serve cloud. Capacity is arranged by lease with the company.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

TeraWulf

An energy and data centre company hosting high-performance computing and AI workloads under a brand called WULF Compute.

Recorded workload fit: Large tenants seeking powered, high-density AI hosting.

Recorded access route: Closest allowed type is gpu_cloud: this is a hosting and power operator, not a self-serve cloud. Terms are by arrangement with the company.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Digital Realty

A colocation and interconnection operator with data centres in many metros, including high-density AI deployments.

Recorded workload fit: Enterprises wanting colocation from single racks to private suites, with connectivity services.

Recorded access route: Closest allowed type is gpu_cloud: this is colocation, not a GPU rental service. Space is contracted with the company.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

GMO Internet (GMO GPU Cloud)

GMO GPU Cloud is an organisational compute route with managed HPC and bare-metal options.

Recorded workload fit: Organisations checking an inquiry-led managed HPC or bare-metal route; confirm suitability with the provider.

Recorded access route: The official contact route excludes individual customers, and the contracting route is inquiry-led. Published managed-HPC pricing does not establish eligibility or capacity. Customer geography, contractual eligibility beyond corporate status, current capacity and successful provisioning are not verified.

Publishes a price list

Record checked 17 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

CloudGPU

A paid GPU rental platform with deployment templates and documented REST and MCP routes. CloudGPU says it uses third-party GPU suppliers and is in public beta without an uptime SLA.

Recorded workload fit: Developers investigating experimental GPU workloads who can check location, billing and data-preservation requirements before committing.

Recorded access route: Read the official agent guide before connecting an agent or creating an account. Check whether the chosen configuration is billed by the minute or per started hour. Back up results before destroying an instance; CloudGPU's terms say local data is deleted. Do not use the ordinary rental or Job route for a hard location requirement: CloudGPU's guide says its regional data is for planning and does not enable region-bound purchases. Read the provider's current eligibility and acceptable-use terms before signing up.

Publishes a price list

Record checked 17 November 2026. Inspect sources and limits; confirm current terms and availability with the provider.

GPU cloud

Neysa

Neysa's Tytan is a dedicated GPU cluster route in India, arranged with the provider rather than presented here as an instant self-serve instance.

Recorded workload fit: Teams investigating a longer-term dedicated cluster in India, rather than a short on-demand experiment.

Recorded access route: Tytan's product page lists 6 to 36 month reserved terms, with bare metal, managed Kubernetes or Slurm. Ask Neysa for the exact accelerator and configuration, spare capacity, minimum commitment, start date and complete quote before ordering. Its 18 November release describes an operational Mumbai MI350X deployment with capacity already contracted; this does not establish capacity available to new customers. The product page lists MI350 separately: do not assume its generic configuration describes that MI350X deployment. Customer eligibility, available allocation and successful provisioning are unverified.

Pricing not assessed

Record checked 19 November 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Marketplaces

Marketplace

Vast.ai

A marketplace where hosts, from individuals to data centres, list GPUs and renters choose among them.

Recorded workload fit: Cost-aware users, researchers and hobbyists comfortable choosing between hosts of varying type and reliability.

Recorded access route: Self-serve signup with prepaid credit; hosts can also sign up to list machines.

Prices set by hosts or bids

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Marketplace

RunPod

A GPU platform offering pods, serverless GPU endpoints and multi-node clusters from a shared catalogue in many regions.

Recorded workload fit: Developers and small teams who want pods for development and serverless endpoints for inference.

Recorded access route: Self-serve account signup.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Marketplace

Shadeform

A single console and billing relationship for provisioning GPUs from many separate cloud providers.

Recorded workload fit: Teams that want to compare and switch between several GPU clouds without managing separate accounts.

Recorded access route: Self-serve account signup.

Publishes a price list

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Marketplace

Replicate

A platform and API for running, fine-tuning and deploying machine learning models, including community-published open models.

Recorded workload fit: Developers who want to call hosted models through an API, or package their own with the open-source Cog tool.

Recorded access route: Self-serve signup, with Python, Node.js and HTTP access and a web playground. The pricing page says some models are billed by hardware and time, others by input and output.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Marketplace

Prime Intellect

An AI development platform combining a post-training lab, inference API and a compute offering that sources GPUs from multiple cloud providers.

Recorded workload fit: Researchers and teams wanting to rent GPUs from several providers through one dashboard, or run reinforcement learning environments.

Recorded access route: Web dashboard, CLI and Python library. Rates are shown on the site by GPU type; no separate pricing page was linked.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Marketplace

Hyperbolic

A GPU access platform drawing on a network of compute suppliers. It offers on-demand GPUs, reserved clusters and private cloud, plus an OpenAI-compatible API for model serving.

Recorded workload fit: Teams wanting self-serve GPU rental without a sales process, or longer reserved capacity.

Recorded access route: Self-serve dashboard for on-demand instances; reserved clusters and private cloud are arranged separately.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Marketplace

Hugging Face Inference Providers

A unified API that routes requests for hosted models to a range of partner inference providers. Dedicated Inference Endpoints are a separate service.

Recorded workload fit: Developers who want one API across several inference vendors without changing code.

Recorded access route: Self-serve with a Hugging Face access token, through Python and JavaScript clients or an OpenAI-compatible chat API.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Marketplace

SF Compute

SF Compute is a marketplace route for GPU virtual machines, bare metal and managed Slurm.

Recorded workload fit: Teams prepared to complete provider onboarding and arrange payment before a marketplace order can be filled; confirm current allocation and suitability directly.

Recorded access route: Account application and successful onboarding are required. An order is filled only when its payment conditions and matching compute are available; an instance without a matching allocation waits. Customer eligibility, current regions and capacity, allocation time and successful provisioning were not verified. Listing unused compute for resale is subject to prior company authorisation.

Prices set by hosts or bids

Record checked 23 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Decentralised networks

Decentralised network

Akash Network

An open compute marketplace where independent providers bid to run containerised workloads.

Recorded workload fit: Developers comfortable with containers who want to source compute through a bidding process.

Recorded access route: Self-serve through the network's tooling; payment and wallet requirements are described on the site.

Prices set by hosts or bids

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Decentralised network

io.net

A network that pools GPU capacity from many sources and offers it as on-demand clusters for AI workloads.

Recorded workload fit: AI teams exploring distributed GPU supply as an alternative to a single cloud provider.

Recorded access route: Self-serve signup through the platform; check the site for current payment options.

Token-based pricing; check which token and billing unit applies

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Decentralised network

Render Network

A distributed GPU network aimed mainly at 3D rendering and generative imaging, where node operators contribute spare GPU capacity.

Recorded workload fit: 3D artists and creative studios needing rendering capacity, and GPU owners looking to contribute capacity.

Recorded access route: Self-serve for creators; node operators join through the network's own onboarding.

Token-based pricing; check which token and billing unit applies

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Decentralised network

Aethir

A distributed GPU cloud where providers contribute GPU servers, with one product for AI workloads and another for cloud gaming.

Recorded workload fit: AI and gaming teams interested in distributed GPU supply, and operators who want to contribute hardware.

Recorded access route: Contact or sign up through the site; the available route depends on the product.

Quote required

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Decentralised network

Salad

A platform that orchestrates GPUs across idle consumer computers worldwide, alongside dedicated data centre options. Workloads are run as containers.

Recorded workload fit: Teams running interruptible, containerised GPU workloads that can tolerate consumer-grade hardware.

Recorded access route: Self-serve through a web portal and API, with Kubernetes integration available.

Publishes a price list

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Decentralised network

Bittensor

An open blockchain network in which independent subnets produce digital commodities such as compute, inference and storage. Contributors are paid in the network's TAO token.

Recorded workload fit: Developers and miners who want to build on, or contribute to, a token-incentivised machine intelligence network.

Recorded access route: Accessed through a Python SDK and the btcli command line tool, with a wallet and TAO staking for mining and validation.

Token-based pricing; check which token and billing unit applies

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Decentralised network

Golem Network

A decentralised marketplace where people sell spare computing capacity and requestors rent it, for work such as rendering and AI models.

Recorded workload fit: Developers who want to submit tasks to community-provided compute, and owners of idle hardware who want to offer it.

Recorded access route: Requestors integrate through a JavaScript API or a Ray interface; payment between parties uses the GLM token.

Token-based pricing; check which token and billing unit applies

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Decentralised network

Flux (RunOnFlux)

A decentralised cloud made of user-operated nodes. It offers general app deployment, AI and edge services, with applications run as Docker containers.

Recorded workload fit: Developers who want to deploy containerised apps on community-run infrastructure, and people who want to operate nodes.

Recorded access route: Deploy from Docker images or Git repositories through FluxCloud; the page read did not state a pricing route, so check the provider.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Decentralised network

Nosana

An open-source GPU cloud with a dashboard for renting GPUs and a route for owners to offer idle GPUs to the network.

Recorded workload fit: AI teams needing GPU rental for inference, fine-tuning or rendering, and GPU owners wanting to share capacity.

Recorded access route: Deploy through a dashboard with a searchable GPU catalogue; GPU credits or custom arrangements are available.

Token-based pricing; check which token and billing unit applies

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Decentralised network

Ritual

A research lab building infrastructure for long-running autonomous AI agents, covering native compute, privacy, verification and coordination.

Recorded workload fit: Developers building autonomous agent systems who want to test on its live network.

Recorded access route: Documentation and a live test network are linked from the site; no pricing was shown on the page read.

Quote required

Record checked 3 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Public programmes

Public programme

UK AI Research Resource (AIRR)

Publicly funded AI supercomputing capacity in the UK, offered to researchers and organisations through application routes.

Recorded workload fit: UK researchers, academics and UK-registered organisations needing large-scale AI compute.

Recorded access route: Apply through one of the published routes, following the UKRI guidance for that route.

Allocated by application, not bought by the hour

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Public programme

EuroHPC AI Factories

A European network of AI-oriented supercomputing sites offering computing resources and support services.

Recorded workload fit: European startups, SMEs and scientific users developing AI.

Recorded access route: Apply through the access modes and calls published by EuroHPC, with a helpdesk and FAQs for guidance.

Allocated by application, not bought by the hour

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Public programme

US NAIRR Pilot

A National Science Foundation-led pilot connecting US researchers and educators with computing, data and training resources.

Recorded workload fit: US-based researchers, educators and start-up projects working on AI.

Recorded access route: Apply through the opportunity that fits: research, education, start-up or deeper partnerships.

Allocated by application, not bought by the hour

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

Public programme

Canadian Sovereign AI Compute Strategy

A Canadian government strategy to expand domestic AI compute, including a fund to help businesses buy compute.

Recorded workload fit: Canadian businesses in eligible sectors looking for support with compute costs.

Recorded access route: Apply through the AI Compute Access Fund; eligibility details are on the fund page linked from the strategy.

Allocated by application, not bought by the hour

Record checked 2 October 2026. Inspect sources and limits; confirm current terms and availability with the provider.

How we label evidence
  • Official source the body that made the rule or the data says so
  • Two sources agree independent sources match
  • One source, not confirmed a single report or claim
  • Watching a signal, not shown as a fact