Cheapest H100
Cloud Providers
On-demand, spot, and reserved H100 80GB SXM5 pricing across 9 cloud providers — ranked cheapest first. All prices in USD per GPU per hour.
$1.65
Cheapest H100
RunPod Community
$2.19
Best Reliable Rate
Lambda Labs on-demand
$12.29
Most Expensive
AWS p5 per GPU
6×
Price Spread
Specialist vs AWS
RunPod Community
Community CloudCHEAPEST SPOTH100 SXM5 80GB
On-Demand
$1.65
per GPU/hr
Spot/Preemptible
$0.75–1.29
1yr Reserved
N/A
Spot-like — can be interrupted. Best for batch workloads. Prices have fallen as supply increased.
Lambda Labs
Specialist CloudBEST VALUEH100 SXM5 80GB
On-Demand
$2.19
per GPU/hr
Spot/Preemptible
N/A
1yr Reserved
~$1.75
Most reliable specialist cloud. No commitment required. SLA included.
CoreWeave
Specialist CloudH100 SXM5 80GB
On-Demand
$2.50
Spot/Preemptible
N/A
1yr Reserved
~$2.00
Enterprise-grade infra with InfiniBand clustering. Best for 16+ GPU reserved deployments.
RunPod Secure Cloud
Secure CloudH100 SXM5 80GB
On-Demand
$2.35
Spot/Preemptible
N/A
1yr Reserved
~$1.99
Dedicated pod, SLA-backed. Reliable on-demand option.
Crusoe
Green CloudH100 SXM5 80GB
On-Demand
$2.65
Spot/Preemptible
N/A
1yr Reserved
~$2.15
Clean energy (flare gas capture). Good for ESG-conscious teams. Competitive pricing.
Together AI
Serverless InferenceSERVERLESSShared H100 pool
On-Demand
Pay per token
Spot/Preemptible
N/A
1yr Reserved
N/A
No GPU provisioning — serverless inference API. ~$0.54/M tokens for Llama 70B class models.
Google Cloud (GCP)
HyperscalerH100 80GB (a3-highgpu-8g ÷ 8)
On-Demand
$3.93 per GPU*
Spot/Preemptible
$1.18–1.90
1yr Reserved
$3.12
*Must buy 8-GPU node at $31.45/hr. Best spot pricing among hyperscalers. Good VPC/compliance integration.
Azure
HyperscalerH100 80GB (ND96isr_H100_v5 ÷ 8)
On-Demand
$4.09 per GPU*
Spot/Preemptible
$1.23–2.05
1yr Reserved
$3.28
*Must buy 8-GPU node at $32.77/hr. Enterprise security, Active Directory integration, compliance.
AWS
HyperscalerMOST EXPENSIVEH100 SXM5 80GB (p5.48xlarge ÷ 8)
On-Demand
$12.29 per GPU*
Spot/Preemptible
$3.69–6.15
1yr Reserved
$7.77
*p5.48xlarge at $98.32/hr for 8 GPUs. Premium pays for AWS ecosystem (VPC, IAM, CloudWatch, compliance).
Which Provider to Choose
If you need:
Cheapest possible, interruptions OK
→ RunPod Community
Batch inference, training jobs with checkpointing, dev/test workloads. Save 30–60% vs on-demand.
If you need:
Production inference, no commitment
→ Lambda Labs
$2.49/hr on-demand, no minimum. Best for steady-state inference APIs without long-term lock-in.
If you need:
Scaling to 16+ GPUs, enterprise
→ CoreWeave
Reserved pricing at scale, InfiniBand networking for distributed training, strong SLAs.
If you need:
Need AWS/GCP ecosystem (IAM, VPC, compliance)
→ GCP or Azure
GCP has the best H100 spot pricing among hyperscalers. Azure for Teams/Office integrations. Expect 2–4× premium.
If you need:
Serverless / no GPU management
→ Together AI
Pay per token, no provisioning. Works for standard models. ~$0.80/M tokens for Llama 70B class.
If you need:
Clean energy requirement
→ Crusoe
Powered by flare gas capture. Competitive rates at $2.79/hr. Good for ESG reporting.
Frequently Asked Questions
What is the cheapest way to get an H100 GPU in the cloud?
RunPod Community Cloud offers H100 SXM5 at approximately $1.65/hr on-demand and $0.75–1.29/hr for spot/community instances as of May 2026. Lambda Labs offers the best balance of price ($2.19/hr) and reliability for on-demand H100 access. For the absolute cheapest option with interruption tolerance, RunPod community spot is the winner.
Why is AWS H100 so much more expensive than Lambda Labs?
AWS p5.48xlarge charges $98.32/hr for an 8-GPU H100 node — equivalent to $12.29/hr per GPU, versus $2.19/hr on Lambda Labs. The premium pays for AWS's ecosystem (VPC, IAM, CloudWatch, compliance certifications, enterprise support), not for the GPU hardware itself. For pure GPU workloads without AWS-specific integrations, specialist clouds are now 5–6× cheaper.
Is Lambda Labs or CoreWeave better for H100?
Lambda Labs is better for 1–8 GPU on-demand usage with no commitment. CoreWeave is better for teams needing 16+ GPUs consistently, reserved pricing, or enterprise-level networking (InfiniBand clusters). As of May 2026, Lambda's on-demand rate ($2.19/hr) slightly undercuts CoreWeave's rate ($2.50/hr), but CoreWeave's reserved pricing ($2.00/hr) is competitive at scale.
What H100 spot pricing is available?
Spot/preemptible H100 pricing as of May 2026: RunPod Community $0.75–1.29/hr (best spot rate), GCP spot $1.18–1.90/hr, Azure spot $1.23–2.05/hr, AWS spot $3.69–6.15/hr. Lambda Labs does not offer spot instances. Spot instances can be interrupted with little notice — use them for fault-tolerant batch workloads with checkpointing.