Cloud GPU providers
Compare 32 providers offering GPU compute. Filter by GPU, region, billing model and credits to find the best fit for your workload.
Serverless GPUs, zero idle cost
Sub-200ms cold starts, autoscale from zero, pay only while your code runs.
We may earn a referral fee from some providers when you sign up through our links — marked with a partner badge. This does not affect ranking, ordering, or pricing.
AceCloud provides GPU cloud instances for AI and high-performance workloads.
Other rental types are available.
Decentralized compute marketplace: independent providers around the world bid on your deployment and you pay the winning bid per GPU-hour, in USD or AKT. H100, H200, A100, RTX PRO 6000 and consumer cards, deployable from Akash Console in minutes.
AM ThermaLink provides distributed GPU compute for AI, rendering, and other high-performance workloads. Its infrastructure is designed to reduce environmental impact by recovering and reusing server heat rather than allowing it to go to waste.
Polish cloud provider with enterprise infrastructure and GPU hosting.
US-based GPU cloud offering dedicated NVIDIA RTX GPUs and lean CPU instances, with a browser console, per-VM firewall, BYOK Windows and free backups included on every VM.
Cheap NVIDIA H100/H200 clusters.
Other rental types are available.
AI Journey Starts Here: enterprise AI infrastructure and GPU cloud services for training, inference, and high-performance computing.
On-demand & spot GPU sandboxes, B300 to RTX 4090, billed per second
Simple, reliable cloud computing platform designed to accelerate your AI/ML and high-performance computing workloads.
RTX Pro 6000 GPU cloud for open-source speech & image AI.
Specialized AI model hosting, with on-demand GPU instances billed per second
Dedicated bare-metal NVIDIA GPU servers with full OS and runtime control over SSH/IPMI, hosted in a Tier III datacenter with hardware-level redundancy and direct engineering support.
Hyperstack provides GPU cloud computing with high-performance NVIDIA GPUs for AI, rendering, and compute workloads.
Dedicated NVIDIA bare-metal GPU servers from a German provider, in EU, US and APAC.
Other rental types are available.
Lambda provides cloud GPU instances for AI training, inference, and other compute-intensive workloads with competitive pricing.
Latitude.sh provides bare metal GPU infrastructure with per-region availability and hourly pricing.
Decentralized GPU marketplace built on Bittensor: independent providers list their machines and compete on price. H100, H200, B200 and consumer cards, billed per second, with SSH access in under a minute.
Self-serve GPU rental platform based in Canada with in-house datacenter infrastructure.
AI and agent cloud platform providing on-demand GPU rentals for training and inference workloads with high performance and cost efficiency.
High-Performance Computing is no longer a privilege reserved solely for large corporations. Enterprise-class GPUs, industry-leading prices.
GPU marketplace that aggregates live on-demand capacity from multiple upstream clouds into a single catalog, deployed and billed from prepaid credits.
GPU cloud marketplace aggregating bare metal and VM instances across multiple data centers worldwide.
Runpod provides cloud GPU compute for AI workloads with simple pricing and global availability.
European cloud provider offering GPU instances with competitive pricing and NVIDIA hardware.
European cloud provider offering GPU instances for AI workloads and high-performance computing.
TensorWave® expertly leverages the next generation of AMD accelerators to provide scalable, memory-optimized infrastructure for the most demanding AI workloads.
Bare-metal NVIDIA B300 and GB300 nodes with NVLink 5 and a dedicated RDMA fabric in a US data center, rented as whole nodes with root access and billed per minute from a prepaid balance.
Other rental types are available.
Theta EdgeCloud is a hybrid cloud-edge computing platform offering GPU instances for AI training and inference.
Thunder Compute provides cloud GPU instances for AI and compute-intensive workloads with prototyping and production deployment options.
Rent cloud GPUs from anywhere. Save up to 90% vs AWS, Azure, GCP with a few clicks.
Verda (formerly DataCrunch) provides high-performance GPU cloud instances for AI/ML workloads with competitive pricing and enterprise-grade infrastructure.
High performance SSD cloud servers, compute instances, and dedicated servers. Deploy in seconds with locations worldwide.
Missing a provider?
How to choose a provider
A few practical considerations depending on what you're building.
- Look for providers that offer credits
- Choose simple, notebook-style interfaces
- Pick consumer GPUs (RTX 30/40) for cost
- Auto-shutdown to avoid surprise bills
- Compare API and CLI integration
- Per-second billing for spiky workloads
- Persistent volumes for state
- Spot / preemptible options for inference
- Enterprise SLA and dedicated support
- Compliance — SOC 2, HIPAA, ISO 27001
- Volume discounts and reserved capacity
- EU-sovereign or air-gapped if needed
Frequently asked questions
How many cloud GPU providers can I compare?
Which cloud GPU provider is the cheapest?
Do any providers offer signup credits?
How do I choose a cloud GPU provider?
Run a GPU cloud?
List your platform on GPUs.io and reach teams comparing cloud GPU pricing, specs and regions. Submit your cloud and we'll review it for inclusion.