Preise mit den Preisseiten der Anbieter abgeglichen; letzte Aktualisierung 2026-08-02
GPU-Mietpreise
Live-Tracker für GPU-Cloud-Preise. Die Spannen spiegeln veröffentlichte On-Demand-Tarife von Hyperscalern und Neoclouds wider.
| GPU | VRAM | On-Demand, $/GPU·h (min) | median | (max) | 1-Jahres-Vertrag, $/GPU·h |
|---|---|---|---|---|---|
| NVIDIA GB200 (per GPU, NVL72) | 186 GB | $10.50 | $16.00 | $27.04 | — |
| NVIDIA B200 (180GB) | 180 GB | $3.75 | $6.19 | $12.36 | — |
| AMD MI355X (288GB) | 288 GB | $2.29 | $4.90 | $8.60 | $4.25 |
| NVIDIA H200 (141GB) | 141 GB | $2.30 | $4.29 | $6.87 | $3.72 |
| NVIDIA H100 (80GB) | 80 GB | $1.49 | $3.35 | $6.98 | $2.35 |
| AMD MI300X (192GB) | 192 GB | $1.85 | $3.00 | $7.86 | $1.82 |
| NVIDIA A100 (80GB) | 80 GB | $1.19 | $1.88 | $5.07 | — |
| NVIDIA L40S (48GB) | 48 GB | $0.79 | $1.54 | $2.25 | — |
| NVIDIA RTX 5090 (32GB) | 32 GB | $0.30 | $0.54 | $0.99 | — |
| NVIDIA RTX 4090 (24GB) | 24 GB | $0.20 | $0.44 | $0.69 | — |
NVIDIA GB200 (per GPU, NVL72)
Rack-scale product with very limited on-demand access: CoreWeave lists the cheapest published rate (~$10.50/GPU-hr), while hyperscaler bare-metal listings reach ~$27/GPU-hr. Most capacity (Together AI, Oracle, Azure) is sold as reserved contracts with quote-only pricing.
Quelle: GetDeploying — GB200 price comparison, CoreWeave pricing, Together AI — GB200 NVL72
NVIDIA B200 (180GB)
Min is a dynamic-priced neocloud listing (Packet.ai via GetDeploying); typical neocloud on-demand sits at $5.89-8.60 (RunPod, Lambda, CoreWeave), with AWS P6-B200 at ~$12.36/GPU-hr the published ceiling after AWS's July 2026 price hike. No clean published 1-year rate: long-term contract listings average ~$3.50-3.80, while Lambda's short-term cluster rates run far higher.
Quelle: GetDeploying — B200 price comparison, RunPod pricing, Lambda pricing, AWS capacity-block price hike (2026)
AMD MI355X (288GB)
Newest AMD flagship with a thin rental market: Vultr's 8x pods start at ~$2.29/GPU-hr, while Oracle bare metal lists $8.60/GPU-hr; the median on-demand price has risen sharply (~232% since Oct 2025) as supply stays tight. CoreWeave publishes a 1-year reservation at $4.25/GPU-hr vs $7.20 on-demand.
Quelle: GetDeploying — MI355X price comparison, Thunder Compute — CoreWeave pricing guide, Spheron — MI300X/MI355X pricing 2026
NVIDIA H200 (141GB)
Min from FluidStack ($2.30), typical neocloud range $3.59-4.50 (RunPod, Nebius, Lambda), max is AWS p5en at ~$6.87/GPU-hr. Median on-demand is up ~23% year-over-year as H200 became the workhorse training GPU; reserved listings average ~$3.72/GPU-hr.
Quelle: GetDeploying — H200 price comparison, RunPod pricing, Nebius AI Cloud prices
NVIDIA H100 (80GB)
Widest rental market (48+ providers): marketplace/neocloud low end $1.49-2.01 (Vast.ai, UpCloud), hyperscaler high end up to $6.98 (Azure). Per SemiAnalysis, 1-year contract rates rebounded ~40% from their Oct 2025 low of $1.70 to ~$2.35/GPU-hr by spring 2026.
Quelle: GetDeploying — H100 price comparison, IntuitionLabs — H100 rental prices across 15+ clouds, Lambda pricing, Vast.ai live GPU pricing
AMD MI300X (192GB)
Aggressively priced for its 192GB of HBM3: Vultr ($1.85) and DigitalOcean ($1.99) anchor the low end while Azure's 8x on-demand tops out at $7.86/GPU-hr. Cyfuture AI publishes a 12-month reserved rate of $1.82/GPU-hr; prices trending down as MI355X arrives.
Quelle: GetDeploying — MI300X price comparison, Thunder Compute — MI300X pricing, Spheron — MI300X/MI355X pricing 2026
NVIDIA A100 (80GB)
Legacy workhorse in steady decline: RunPod community cloud from $1.19, neoclouds at $2.70-2.79 (CoreWeave, Lambda), with GCP a2-ultragpu still listing ~$5.07/GPU-hr at the top. No widely published 1-year rate; most contract demand has moved to H100/H200.
Quelle: GetDeploying — cloud GPU reference, RunPod pricing, GPUCost — GCP GPU pricing
NVIDIA L40S (48GB)
Inference/graphics mid-tier: RunPod community cloud from $0.79, Nebius at $1.55-1.82, CoreWeave at the published top (~$2.25/GPU-hr). Stable pricing; no standard published 1-year rates.
Quelle: GetDeploying — cloud GPU reference, RunPod pricing, Nebius AI Cloud prices
NVIDIA RTX 5090 (32GB)
Consumer flagship on marketplaces only (no hyperscaler offering): Vast.ai/Salad listings from ~$0.30, RunPod Secure Cloud at $0.99 the published top. Supply growing through 2026, pushing prices toward RTX 4090 levels.
Quelle: GetDeploying — cloud GPU reference, Vast.ai — RTX 5090 pricing, RunPod pricing
NVIDIA RTX 4090 (24GB)
Cheapest way to rent CUDA compute: Salad from $0.20 and Vast.ai interruptible listings as low as ~$0.16, with RunPod Secure Cloud at $0.69 the published ceiling. Median down ~27% year-over-year as RTX 5090 supply displaces it.
Quelle: GetDeploying — RTX 4090 price comparison, RunPod pricing, Vast.ai live GPU pricing