Renting a single H100 80GB costs anywhere from $2.69 to $6.98 per hour depending on the provider — a 2.6× spread for the same silicon. Go spot/interruptible and the floor drops to $1.60/hr. On a week-long fine-tuning job, picking the right provider is the difference between a $450 bill and a $1,170 one.
CloudMart tracks GPU pricing across 10 providers and refreshes it twice a week. Here's where H100 (and A100) rental prices actually stand in 2026, who's cheapest at each tier, and which provider fits which workload.
H100 80GB: single-GPU on-demand prices
| Provider | On-demand /hr | Spot /hr | Notes |
|---|---|---|---|
| RunPod | $2.69 | $1.99 | Per-second billing, Secure + Community tiers |
| Vast.ai | $2.80 | $1.80 | P2P marketplace — cheapest, variable reliability |
| Lambda Labs | $2.99 | — | Flat rate, no interruptions, ML-focused tooling |
| Crusoe | $3.90 | $1.60 | Cheapest spot H100 we track |
| OVHcloud | $4.99 | — | EU-based, predictable pricing |
| CoreWeave | $6.16 | — | InfiniBand interconnect, built for clusters |
| Google Cloud | $6.98 | $2.09 | A3 instances, deep GCP integration |
Prices are per GPU-hour as tracked on June 12, 2026. GPU pricing moves fast — check the live GPU marketplace for today's numbers.
8× H100 clusters (distributed training)
For multi-GPU training, the per-node spread is even wider:
| Provider | 8× H100 /hr | Per GPU |
|---|---|---|
| RunPod | $21.52 | $2.69 |
| Lambda Labs | $23.92 | $2.99 |
| Crusoe | $31.20 | $3.90 |
| CoreWeave | $49.24 | $6.16 |
| Google Cloud | $55.84 | $6.98 |
A caveat on the cheap end: at cluster scale, interconnect matters as much as the hourly rate. CoreWeave's premium buys InfiniBand networking with near-linear scaling across nodes — for serious distributed pre-training, a cheaper cluster that scales at 70% efficiency can cost more per useful FLOP. For single-node jobs (which covers almost all fine-tuning), the cheap providers are simply cheaper.
Don't need an H100? A100 80GB is the value play
For most fine-tuning and inference workloads, an A100 80GB delivers ~60–70% of H100 training throughput at roughly half the price:
| Provider | A100 80GB /hr |
|---|---|
| Paperspace | $1.15 |
| RunPod (SXM) | $1.64 |
| Vast.ai | $1.80 |
| Lambda Labs | $1.99 |
| Crusoe | $2.42 |
| CoreWeave | $2.70 |
And if you're doing inference on a quantized 7B–13B model, you may not need datacenter GPUs at all: an RTX 4090 24GB runs $0.55/hr on Vast.ai ($0.35 spot) and $0.74/hr on RunPod.
What a real job costs
- LoRA fine-tune, 7B model, ~6 hours on 1× A100 80GB: $7–$12 on Paperspace/RunPod. This is the workload most people actually have.
- Full fine-tune, 13B model, ~3 days on 1× H100: ~$194 on RunPod on-demand, ~$115 on Crusoe spot (with checkpointing).
- Week of 8× H100 distributed training: ~$3,600 on RunPod, ~$9,400 on Google Cloud. At this scale, negotiate — listed prices are a starting point.
Which provider should you pick?
- Cheapest possible experiment: Vast.ai spot. Accept that hardware quality varies and your job can be reclaimed.
- Iterating on fine-tunes: RunPod — per-second billing and pre-built PyTorch templates make start/stop cycles cheap.
- A long run you can't afford to lose: Lambda Labs — flat rate, no interruptions, no surprise bills.
- Spot with checkpointing discipline: Crusoe at $1.60/hr is the lowest H100 number on the board.
- 8+ GPU distributed training: CoreWeave — the InfiniBand premium pays for itself at scale.
- Already on AWS/GCP/Azure: The integration tax is real, but so is the convenience. GCP spot at $2.09/hr is genuinely competitive.
Not sure what GPU your job needs?
Tell the Planner your model size and workload — it picks the GPU, provider, and estimates the job cost.
Plan my training run →Related: LLM API pricing compared — if you just want to call a model rather than host one, per-token APIs are almost always cheaper than renting GPUs.