GPU hosting made understandable

Find the right GPU cloud for what you actually want to run.

Compare GPU providers for AI video, image generation, LLMs and training — without decoding enterprise cloud jargon.

Compare providersEstimate your cost
✓ Retail-friendly✓ Indie developer focused✓ Clear hosting types
Example workload

AI video generation

Example VRAM24 GB+
Hosting optionPod or Serverless
What mattersQueue time + $/job
See how to choose →

AI image generation

Example VRAM16 GB+
Hosting optionPod or Serverless
What mattersSpeed + cost per image
See how to choose →

LLM inference

Example VRAM24 GB+
Hosting optionPod or Managed API
What mattersModel size + response time
See how to choose →

Model fine-tuning

Example VRAM48 GB+
Hosting optionPod or GPU VM
What mattersMemory + training time
See how to choose →

Bursty production API

Example VRAM24 GB+
Hosting optionServerless
What mattersCold starts + concurrency
See how to choose →

Experimenting and testing

Example VRAM16 GB+
Hosting optionPod or GPU VM
What mattersEasy setup + idle cost
See how to choose →
1 / 6Illustrative setups. VRAM needs vary by model.
Your workload. Your GPU. Your next move.

Start with what you want to run

AI image generationAI video generationLLM inferencetraining and fine-tuningServerless GPU hostingComfyUI

Read practical hosting guides →

Compare

GPU hosting providers

Filter by what you want to run, rather than by infrastructure terminology.

Nebius

AI cloud infrastructure for model training and inference, including hosted model services.

Best forLLM inference, Training / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

CoreWeave

GPU infrastructure with managed Kubernetes for training and inference.

Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Radiant / Ori

Radiant's GPU infrastructure includes virtual machines and bare-metal compute.

Pods / VMs
Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Fluidstack

Large-scale AI compute and data-center infrastructure for enterprise workloads.

Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Jarvislabs

GPU virtual machines, development templates and serverless deployments for AI workloads.

Pods / VMsServerless
Best forLLM inference, Training / fine-tuning, Experimenting and testing
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

fal

Image, video and audio model APIs, serverless GPUs and dedicated compute for model development.

Pods / VMsServerlessAPI
Best forAI video generation, AI image generation, Bursty production API, Training / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Hyperstack

GPU virtual machines with API access for AI training, inference and rendering.

Pods / VMs
Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Oracle Cloud

GPU virtual machines and bare-metal instances for AI training and accelerated computing.

Pods / VMs
Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

CUDO Compute

Enterprise GPU infrastructure and cluster operations for AI workloads.

Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Together AI

Hosted model inference, fine-tuning and GPU compute for AI development.

API
Best forLLM inference, Bursty production API, Training / fine-tuning, Experimenting and testing
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Replicate

Hosted APIs for image, video and language models, with fine-tuning and custom deployments.

API
Best forAI video generation, AI image generation, LLM inference, Bursty production API, Training / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

NodeShift

Private AI platform with a model gateway and deployment across cloud or private infrastructure.

API
Best forLLM inference, Bursty production API
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Lambda

On-demand GPU instances and clusters for model training and inference, with API and CLI access.

Pods / VMsUp to 180.00 GB VRAM
Best forLLM inference, Training / fine-tuning
GPU rangeNVIDIA A10, NVIDIA A100 PCIe, NVIDIA A100 SXM, NVIDIA B200 SXM6, NVIDIA GH200, NVIDIA H100 PCIe, NVIDIA H100 SXM5, NVIDIA Quadro RTX 6000, NVIDIA RTX A6000
From$0.69/hr
DetailsVisit provider

Google Cloud

Compute Engine provides GPU-backed virtual machines with configurable machine types.

Pods / VMs
Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

RunPod

GPU Pods for container workloads alongside serverless inference and deployment APIs.

Pods / VMsServerlessUp to 180.00 GB VRAM
Best forTraining / fine-tuning, Experimenting and testing
GPU rangeNVIDIA A100 PCIe, NVIDIA B200 SXM6, NVIDIA H100 PCIe, NVIDIA H100 SXM5, NVIDIA H200 SXM, NVIDIA L40S, NVIDIA RTX 4090, NVIDIA RTX A5000
Check provider pricing
DetailsVisit provider

SaladCloud

Distributed GPU containers, dedicated GPU infrastructure and a managed AI gateway.

Pods / VMsAPI
Best forLLM inference, Bursty production API
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Scaleway

GPU instances for AI development, model training and inference.

Pods / VMs
Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Gcore

GPU virtual machines and bare metal for AI training, inference and accelerated computing.

Pods / VMs
Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

io.net

Distributed GPU infrastructure accessed through IO Cloud.

Best forTraining / fine-tuning, Experimenting and testing
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Verda

Verda provides GPU instances and serverless containers for training and inference.

Pods / VMsServerless
Best forLLM inference, Training / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Voltage Park

Bare-metal GPU clusters for large-scale AI training.

Pods / VMs
Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

TensorDock

GPU cloud virtual machines for deep learning, AI and rendering, with deployment APIs.

Pods / VMs
Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Microsoft Azure

Azure provides GPU virtual machines for AI and high-performance computing.

Pods / VMs
Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Paperspace / DigitalOcean

DigitalOcean Paperspace provides GPU virtual machines with published machine configurations.

Pods / VMsUp to 80.00 GB VRAM
Best forTraining / fine-tuning
GPU rangeNVIDIA A100 40 GB, NVIDIA A100 80 GB, NVIDIA H100, NVIDIA RTX A4000, NVIDIA RTX A5000, NVIDIA RTX A6000
From$0.76/hr
DetailsVisit provider

Akash Network

Decentralized marketplace for container workloads, with AkashML offering hosted model inference.

Pods / VMsAPI
Best forLLM inference, Bursty production API
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Thunder Compute

GPU instances and sandboxes for AI development, evaluations and post-training.

Pods / VMs
Best forTraining / fine-tuning, Experimenting and testing
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Modal

Serverless GPU compute for model inference, fine-tuning, training and evaluation.

ServerlessUp to 180.00 GB VRAM
Best forAI video generation, AI image generation, LLM inference, Bursty production API, Training / fine-tuning, Experimenting and testing
GPU rangeNVIDIA A100 40 GB, NVIDIA A100 80 GB, NVIDIA B200 SXM6, NVIDIA H100 SXM5, NVIDIA H200 SXM, NVIDIA L40S
From$1.95/hr
DetailsVisit provider

Latitude.sh

Dedicated bare-metal GPU infrastructure for machine-learning workloads.

Pods / VMs
Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

CloudRift

GPU rentals using virtual machines, containers and bare metal, plus hosted inference endpoints.

Pods / VMsAPI
Best forLLM inference, Bursty production API, Training / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Vultr

GPU virtual machines for AI, machine learning and accelerated computing.

Pods / VMs
Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Vast.ai

Marketplace GPU containers for training, inference and development, plus serverless deployments.

Pods / VMsServerless
Best forTraining / fine-tuning, Experimenting and testing
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

OVHcloud

Public-cloud GPU instances and AI services for training and inference.

Pods / VMsAPI
Best forBursty production API, Training / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider

Amazon Web Services

Amazon EC2 offers GPU virtual machines for machine learning and accelerated computing.

Pods / VMs
Best forTraining / fine-tuning
GPU rangeSee provider
Check provider pricing
DetailsVisit provider
Simple decision

Pod or serverless?

For most people, this is the first decision that matters.

▣

GPU Pod / VM

A GPU machine stays allocated to you while it is running.

  • Best for steady or continuous work
  • Predictable environment
  • You pay while the machine is running
  • Concurrent jobs usually need your own queue/workers
Choose this when your GPU is busy often.
⚡

Serverless GPU

Workers spin up for jobs and scale with demand.

  • Best for bursts and unpredictable traffic
  • Multiple requests can scale across workers
  • You mainly pay for execution time
  • Cold starts can matter
Choose this when requests arrive in bursts.
Quick calculator

What does one generation actually cost?

Hourly rates are hard to compare. Convert them into an approximate cost per image, video or inference job.

Cost per job = hourly GPU price × job seconds ÷ 3,600
Approx. cost / job$0.049
Approx. monthly compute$48.63
Compute only. Storage, idle time, cold starts, data transfer and provider-specific charges may change the total.
GPU provider?

Get your service in front of people actively comparing GPU hosting.

Claim your listing, verify your data, add referral tracking or upgrade placement.

Partner with GPUPad
Provider details