GPU hosting comparison
Rent a GPU by the hour instead of buying one. Below is a cross-provider price board for the most-rented accelerators, followed by every provider we track. Open a provider from the menu for its full catalogue.
On-demand USD prices are indicative snapshots from June 2026 for orientation only — they are not live. Spot/community and committed-use rates are often 30–70% lower. Always confirm on the provider's pricing page before committing.
Cheapest by GPU
Lowest advertised on-demand rate for each popular GPU, and where to get it.
| GPU | Cheapest | At | All providers (on-demand) |
|---|---|---|---|
| H100 80GB | $1.65 | Vast.ai | Vast.ai $1.65 · Hyperstack $1.95 · Nebius $2.00 · Together $2.39 · Crusoe $2.45 · Lambda $2.49 · RunPod $2.69 · Modal $3.95 · CoreWeave $4.76 |
| A100 80GB | $0.80 | Vast.ai | Vast.ai $0.80 · Paperspace $1.15 · RunPod $1.19 · Lambda $1.29 · Together $1.30 · Hyperstack $1.35 · Crusoe $1.45 · CoreWeave $2.21 |
| RTX 4090 24GB | $0.35 | Vast.ai | Vast.ai $0.35 · RunPod $0.69 |
| H200 141GB | $3.15 | Together | Together $3.15 · Lambda $3.49 · Nebius $3.50 · Crusoe $3.90 · RunPod $3.99 · CoreWeave $4.25 |
| B200 192GB | $4.99 | Lambda | Lambda $4.99 · Nebius $5.50 · Crusoe $5.50 · CoreWeave $5.99 |
Providers
RunPod
Per-second billing across Secure Cloud (vetted data centres) and the cheaper Community Cloud. Fast cold starts, serverless endpoints and a large template library make it a favourite for tinkering and bursty inference.
H100 from $2.69 · 6 GPU types
Lambda
An ML-focused cloud from the workstation maker. Clean on-demand instances and reserved 1-Click Clusters with InfiniBand for multi-node training; the Lambda Stack ships CUDA, PyTorch and TensorFlow preinstalled.
H100 from $2.49 · 5 GPU types
Vast.ai
A marketplace that brokers spare capacity from many hosts, so prices are the lowest around — but reliability and location vary by host. A real-time search and bid/interruptible mode let you trade price against guarantees.
H100 from $1.65 · 5 GPU types
CoreWeave
A hyperscale specialist built for large-scale training and inference: InfiniBand-connected clusters, Kubernetes-native scheduling and reserved capacity at fleet scale. Aimed at teams, not hobbyists.
H100 from $4.76 · 5 GPU types
Paperspace (DigitalOcean)
Now part of DigitalOcean, Paperspace pairs simple hourly GPUs with Gradient notebooks and deployments. A gentle on-ramp with free-tier GPUs and a polished notebook experience.
H100 from $2.24/GPU · 4 GPU types
Modal
Serverless GPUs defined entirely in Python — no Dockerfiles or YAML. Per-second billing, scale-to-zero and instant fan-out make it ideal for batch jobs, inference APIs and pipelines.
H100 from $3.95 · 5 GPU types
Nebius
A European AI cloud (spun out of Yandex) with managed GPU clusters, Kubernetes and Slurm. Aggressive on-demand H100/H200 pricing plus reserved B200 capacity at data-centre scale.
H100 from $2.00 · 4 GPU types
Crusoe
A climate-aligned cloud running on stranded and renewable energy. Reserved H100/H200 and B200 clusters with InfiniBand, aimed at sustained large-scale training.
H100 from $2.45 · 4 GPU types
Together AI
GPU clusters and a serverless inference platform from a research-focused cloud. Rents InfiniBand H100/H200 clusters for training alongside per-token inference for open models.
H100 from $2.39 · 4 GPU types
Hyperstack
NexGen Cloud's on-demand and reserved GPUs with transparent hourly pricing and strong European availability from NVIDIA-partner capacity.
H100 from $1.95 · 4 GPU types
Choosing a provider
- Lowest price — Vast.ai brokers spare capacity and is consistently cheapest, at the cost of variable reliability and location.
- Simple & fast — RunPod and Modal bill per second with quick starts and serverless endpoints, great for inference and experiments.
- Serious training — Lambda, CoreWeave and Crusoe offer InfiniBand multi-node clusters and reserved capacity for large runs.
- Cheap clusters — Nebius, Hyperstack and Together AI undercut the big names on on-demand H100/H200 while keeping managed Kubernetes or Slurm.
- Notebooks & onboarding — Paperspace pairs hourly GPUs with managed notebooks and a free tier.