Cloud GPU platforms provide on-demand access to Graphics Processing Units (GPUs) through virtualized infrastructure optimized for AI and machine learning workloads. These services offer specialized GPU hardware, automated scaling, and ML-specific development environments, enabling organizations to access high-performance computing resources without physical hardware investment.
43 verified AI-first sites in Cloud GPU.
Last reviewed .
Cloud platform for running AI and ML workloads. Features seamless deployment, scaling, and management of AI applications. Includes GPU access, model training, and inference optimization. Supports ML development workflow.
Pricing: Usage-based pricing. Enterprise: Custom solutions.
Specialized cloud infrastructure platform. Features GPU cloud computing, AI infrastructure, and deployment solutions. Includes automated scaling and enterprise support. Supports AI workloads.
Pricing: Pay-per-use. Enterprise: Custom solutions.
GPU cloud computing platform. Features on-demand GPU access, serverless deployment, and resource management. Includes cost optimization and scaling tools. Supports AI development.
Pricing: Pay-as-you-go. Enterprise plans available.
Cloud infrastructure platform partnering with top AI labs, governments, and enterprises. Selected by Anthropic for custom data centers in NY and Texas. Features Atlas OS bare-metal provisioning, Lighthouse monitoring and optimization, and dedicated GPU clusters. SOC 2 Type 2, ISO27001, GDPR certified. Single-tenant by default with 15-minute response SLAs.
Pricing: Enterprise pricing. Contact for solutions.
GPU cloud computing marketplace. Features AI training infrastructure, cloud GPU rentals, and deployment tools. Includes cost optimization and resource management. Supports ML development.
Pricing: Pay-per-use. Custom enterprise solutions.
AI-native neo-cloud for GPU-intensive workloads. NVIDIA H100, H200, and Blackwell-class capacity with clusters built for large-scale training and inference.
Pricing: On-demand and reserved-see Nebius.
On-demand GPU cloud for ML training and inference with containers and notebooks. Now part of Voltage Park's neo-cloud stack, extending elastic access alongside Voltage Park's owned H100/Blackwell fleet.
Pricing: Pay-as-you-go GPU hours-see TensorDock / Voltage Park.
Development cloud platform. Features GPU-powered notebooks, development environments, and collaboration tools. Includes automated provisioning, usage analytics, and cost management. Supports deep learning workflows.
Pricing: Pay-as-you-go. Team plans available.
Australian sovereign GPU cloud for AI training and inference. On-demand H100/H200 clusters in Melbourne and Sydney, plus private-cloud and supercluster options for government and research workloads.
Pricing: On-demand and reserved GPU capacity; contact Sharon AI.
Self-serve GPU rental as Docker containers or full VPS. Consumer and datacenter cards from RTX 3060 through H100/B200, with Jupyter, PyTorch, and other templates.
Pricing: Pay-as-you-go from about $0.05/hour; reservations via sales. See SimplePod.
Distributed GPU cloud (SaladCloud) aggregating tens of thousands of consumer and prosumer GPUs for batch inference, training jobs, and container workloads. Pay-as-you-go from very low hourly rates with free credits for new users.
Pricing: Usage-based from roughly $0.02/hour; free credits for new users-see Salad.
Serverless GPU cloud platform for AI and ML workloads. Features automatic scaling, pay-per-use pricing, and simplified deployment. Includes support for PyTorch, TensorFlow, and JAX frameworks. Offers managed infrastructure with zero configuration required. Supports GPU-accelerated inference and training workloads. Includes comprehensive monitoring and logging tools. Particularly valuable for developers needing scalable GPU compute without infrastructure management overhead.
Pricing: Pay-per-use pricing based on compute time. Free tier available for testing.
Cloud GPU infrastructure platform designed for AI and machine learning workloads. Features NVIDIA GPU instances, automated provisioning, and global availability. Includes container orchestration, workspace management, and collaborative development tools. Offers high-performance GPU computing with competitive pricing. Supports various AI frameworks and deployment workflows. Includes comprehensive monitoring and resource optimization features. Particularly valuable for teams requiring reliable GPU infrastructure for AI development and training.
Pricing: Pay-as-you-go pricing. Enterprise plans with custom configurations available.
Serverless ML infrastructure platform for deploying and scaling machine learning models. Features automatic scaling, GPU acceleration, and simple deployment workflow. Includes model versioning, monitoring, and cost optimization tools. Offers support for various ML frameworks and pre-trained models. Features pay-per-use pricing with no infrastructure management required. Includes API endpoints and webhook support. Supports custom model deployment and fine-tuning. Particularly valuable for ML engineers seeking serverless model deployment with minimal configuration.
Pricing: Pay-as-you-go pricing. Free tier available. Enterprise plans with custom pricing.
Agent-native AI infrastructure (FlexAI) with an OpenAI-compatible inference API across 20+ open models, Agent SDK for tool-using loops, dedicated GPU endpoints, and private AI cloud (VPC, on-prem, air-gapped). Path from serverless tokens to reserved NVIDIA/AMD fleets for production agents.
Pricing: Free trial credits; serverless per-token/media rates; dedicated GPU-hour and private cloud quotes-see FlexAI.
Omnicloud platform aggregating multi-cloud GPUs, CPUs, and storage with algorithmic pricing. Features spot bidding for async workloads, short-term reservations (hours to months), and batch job SDK/API. Includes WEKA storage and InfiniBand interconnect. Supports training, fine-tuning, inference, and batch processing. Provides transparent market-based pricing updated weekly based on supply and demand. Offers access to thousands of NVIDIA GPUs across multiple datacenters. Available in USA, Canada, and Europe. Particularly valuable for ML teams requiring flexible GPU access without long-term commitments.
Pricing: Algorithmic pricing based on supply/demand. A100: $0.18-$0.36/hr. H100: $11.98-$14.56/hr (8 GPUs). Storage: $0.08/GB/month. Batch inference: $0.10-$0.50 per 1M tokens.
Silicon-agnostic AI inference platform. Automatically routes workloads across compute providers to optimize cost, latency, and performance. Supports NVIDIA and alternative accelerators. $5 free credits. Unified control plane for heterogeneous GPU infrastructure.
Pricing: Usage-based. $5 free credits.
NVIDIA compute marketplace connecting developers to global GPU capacity from cloud partners. Evolved from Lepton AI acquisition; offers on-demand and reserved Blackwell and NVIDIA-architecture GPUs with NIM, NeMo, and agent workflow integrations.
Pricing: Usage-based marketplace pricing; early access signup on site.
Cloud GPU and AI infrastructure provider offering high-performance compute clusters for model training and inference. Provides on-demand and reserved capacity, managed infrastructure options, and enterprise-grade platform services for large-scale AI workloads.
Pricing: On-demand and reserved pricing; see site for current rates.
AI cloud platform focused on AMD GPU infrastructure for training and inference workloads. Features high-memory accelerator options, optimized networking and storage paths, and infrastructure tailored for production AI and HPC-style pipelines.
Pricing: Usage-based and enterprise pricing; see site for current options.
Cloud GPU marketplace and infrastructure platform providing on-demand NVIDIA and AMD instances for AI training, inference, and research. Supports flexible VM configurations, scalable compute deployment, and cost-focused GPU access across regions.
Pricing: Pay-as-you-go and reserved options; see site for current pricing.
GPU cloud and inference service for affordable access to open-weight LLMs. API and elastic compute for frontier open models without long-term hardware commitments.
Pricing: Usage-based pricing; see hyperbolic.ai for GPU and API rates.
European full-stack AI cloud (Verda; formerly DataCrunch). GPU clusters, VMs, batch jobs, and serverless inference for training and serving-EU-friendly infrastructure for AI teams.
Pricing: Pay-as-you-go and reserved-see Verda.
GPU cloud marketplace to provision NVIDIA and other accelerators across many provider regions from one control plane. Emphasizes price discovery, reservations, and unified deployment for AI training and inference workloads.
Pricing: Marketplace and provider-dependent rates; see platform for current GPU pricing.
Integrated AI stack combining on-demand and reserved GPU capacity with hosted RL training, evaluations, sandboxes, and model deployment. Targets teams building and improving agents and custom models on shared or large multi-node clusters.
Pricing: Usage-based GPU and platform pricing; enterprise and cluster quotes available.
Large-scale vetted GPU cluster operator with CLI and web purchasing for elastic H100-class capacity, short-term bursts, and longer reservations with resale of idle nodes. Positions as traditional-cloud-style support from bare-metal operations upward for AI training workloads.
Pricing: Market-based hourly GPU pricing and contract options; see site for current node rates.
Neo-cloud GPU operator owning 24,000+ NVIDIA H100 and Blackwell-class GPUs across U.S. Tier 3+ datacenters. Offers on-demand HGX nodes from ~$1.99/hr, dedicated reserves, InfiniBand clusters, and Reference Platform NVIDIA Cloud Partner validation for large-scale AI training and inference.
Pricing: On-demand from ~$1.99/hr; dedicated reserve contracts via sales.
EU-sovereign AI GPU cloud with on-demand VMs, serverless runs, dedicated inference endpoints, and InfiniBand clusters scaling to 8,000 GPUs. Features per-second billing, GDPR-compliant European datacenters, Docker-native deployments, and compiler-level workload profiling for right-sized hardware selection.
Pricing: Per-second GPU billing; long-term contracts from 3 months via sales.
AI-native infrastructure platform with on-demand and reserved NVIDIA H100, H200, Blackwell, and Vera Rubin GPU capacity. Includes Inference Engine for serverless model APIs, Cluster Engine for Kubernetes orchestration, bare-metal servers, and managed multi-node training clusters across global regions.
Pricing: Usage-based GPU and inference pricing; enterprise quotes available.
NVIDIA Preferred Partner cloud with owned GPU hardware for AI training, inference, and rendering. Features hourly on-demand instances, single-tenant bare metal, InfiniBand GPU clusters, and REST/MCP APIs for programmatic provisioning from B300 down to A30 SKUs with preinstalled CUDA.
Pricing: On-demand from ~$0.43/hr; cluster quotes via sales.
Developer-first on-demand GPU cloud optimized for low-cost ML prototyping and production jobs. Features NVIDIA A100 and H100 instances from ~$0.78-$1.38/GPU-hr, VS Code extension for one-click connection, CLI and web console, and prebuilt templates for PyTorch, vLLM, and Jupyter workflows.
Pricing: Pay-as-you-go from ~$0.78/GPU-hr; no long-term contracts.
Serverless GPU inference platform for custom ML models. Deploy from Hugging Face, Git, or Docker with autoscaling, dynamic batching, private endpoints, and sub-second cold starts. SOC 2 Type II; customers report up to 90% savings vs fixed GPU clusters.
Pricing: Pay for GPU hours used; deploy on site.
GPU supercomputing cloud for AI training and inference. Offers on-demand instances, 1-Click Clusters, and dedicated superclusters with NVIDIA H100, B200, GB300, and NVL72 racks. SOC 2 Type II; managed clusters and co-engineering for large labs and enterprises.
Pricing: On-demand and reserved GPU pricing; contact for superclusters.
Full-stack AI neo-cloud for bare-metal and VM GPUs, Kubernetes/Slurm clusters, inference endpoints, and fine-tuning. Large sovereign AI-factory footprint in Europe; acquired Anyscale to deepen the AI cloud platform stack.
Pricing: Reserved clusters and platform quotes-see Nscale.
Enterprise GPU rental marketplace aggregating Tier 3/4 data-center capacity-H100, H200, B200/B300 and more-via one dashboard. Spot and dedicated tiers, bare metal or VMs, aimed at teams avoiding hyperscaler lock-in.
Pricing: On-demand from roughly $0.58/GPU-hr; volume and enterprise quotes-see Spheron.
NVIDIA developer cloud to prototype and ship AI on GPU instances across 20+ clouds from one terminal. Launchables are one-click GPU environment templates; Cloud SDK wires your own cloud into Brev. Successor to brev.dev after NVIDIA acquisition.
Pricing: See NVIDIA Brev for instance and credit options; Inception and academic GPU credit programs.
Developer-first automated inference cloud on AMD Instinct GPUs. Sovereign production inference with high-bandwidth Ethernet fabric, self-serve capacity, and infrastructure tuned for serving-not NVIDIA-only fleets.
Pricing: Usage-based and reserved; see hotaisle.xyz.
AWS EC2 P5 instances with NVIDIA H100 GPUs for large-scale AI training and inference. High-throughput networking and Elastic Fabric Adapter for multi-node model training in the Amazon cloud.
Pricing: On-demand, reserved, and Savings Plans-see AWS EC2 P5 pricing.
Google Cloud GPU accelerators for AI training and inference-NVIDIA GPUs across Compute Engine and GKE with A3/A2 machine families, regional capacity, and managed AI stack integration.
Pricing: Pay-as-you-go Compute Engine rates; committed use discounts available.
Oracle Cloud Infrastructure GPU compute for AI training and inference. Bare-metal and VM shapes with NVIDIA GPUs, RDMA cluster networking, and OCI regions for enterprise AI workloads.
Pricing: Pay-as-you-go and committed capacity-see Oracle Cloud GPU.
Microsoft Azure GPU-accelerated VMs for AI training and inference. NC, ND, and NV families with NVIDIA and AMD accelerators, including H100/H200-class options for enterprise deep learning and HPC.
Pricing: Pay-as-you-go and reserved instances-see Azure GPU VMs.
Decentralized GPU cloud marketplace for AI and ML rentals. Rent community and datacenter GPUs on demand for training and inference without long hyperscaler contracts.
Pricing: Marketplace pricing-see Clore.
Cost-focused GPU cloud with large pooled capacity for AI training and inference. On-demand access across thousands of GPUs aimed at cutting hyperscaler bills while keeping reliable infrastructure for ML teams.
Pricing: Pay-as-you-go GPU rentals; see Oblivus pricing.