Cheapest H100 SXM — April 2026
RunPod (spot)$1.69/hr
Vast.ai$1.85/hr
RunPod (on-demand)$2.49/hr
Lambda Labs$2.49/hr
CoreWeave$2.89/hr
AWS p5.48xlarge$4.08/hr
Source: GPU Prices live tracker · 5,269 instances · 54 providers
We track 5,269 GPU instances across 54 providers in real time. Prices change daily. This guide summarizes the cheapest options for every GPU tier in April 2026, with notes on what you give up at each price point.
Cheapest by GPU Tier
Budget Tier (<$1/hr) — For Development & Testing
| GPU | Cheapest Provider | Price | VRAM |
|---|---|---|---|
| T4 | GCP (spot) | $0.11/hr | 16 GB |
| RTX 4000 Ada | Vast.ai | $0.17/hr | 20 GB |
| RTX 4090 | Vast.ai (spot) | $0.38/hr | 24 GB |
| A10G | AWS (spot) | $0.44/hr | 24 GB |
| L4 | GCP (spot) | $0.46/hr | 24 GB |
Mid Tier ($1–$4/hr) — For Training & Production
| GPU | Cheapest Provider | Price | VRAM |
|---|---|---|---|
| RTX 4090 | RunPod (OD) | $0.74/hr | 24 GB |
| A100 40GB PCIe | RunPod | $1.19/hr | 40 GB |
| A100 80GB SXM | Lambda Labs | $1.89/hr | 80 GB |
| H100 NVL | RunPod | $2.19/hr | 94 GB |
| H100 SXM | RunPod / Lambda | $2.49/hr | 80 GB |
Enterprise Tier ($4+/hr) — For Scale
| GPU | Provider | Price | Notes |
|---|---|---|---|
| H200 SXM | CoreWeave | $5.25/hr | Available now |
| H100 x8 node | Lambda | $19.92/hr | 8-GPU full node |
| B200 | CoreWeave / Oracle | $7.50/hr | Limited supply |
| GB200 NVL | CoreWeave | $12.00/hr | New in 2026 |
The 5 Cheapest Providers Ranked
Ranked by average price per GPU-hour across all models in their catalog:
1
Vast.ai $0.34/hr avg
Community marketplace, spot-only, variable reliability
2
RunPod Community $0.44/hr avg
Spot-like pricing, wide GPU selection
3
Shadeform $0.61/hr avg
Aggregator — finds cheapest across 10+ clouds
4
Lambda Labs $0.89/hr avg
Reliable, good H100/A100 pricing
5
RunPod Secure $1.12/hr avg
On-demand, persistent storage, easy setup
What You Give Up Going Cheap
The cheapest providers (Vast.ai, RunPod Community) are marketplace models — your instance runs on hardware owned by third parties. You get lower prices but accept: occasional hardware issues, less predictable network performance, and no enterprise SLA. For training jobs with checkpoints, this is fine. For production inference APIs, use RunPod Secure, Lambda, or CoreWeave.
