GPU availability snapshot · April 30, 2026
GPU Cloud Availability Tracker
Where you can actually get an H100, A100, or RTX 4090 right now. Daily snapshot across 20 GPU clouds and major regions.
Status legend
| Provider | GPU | Region | Status | ETA | Notes |
|---|---|---|---|---|---|
| RunPod | H100 80GB SXM | US-East | ✓ Available | — | Secure Cloud, on-demand |
| RunPod | H100 80GB SXM | US-West | ✓ Available | — | Secure Cloud |
| RunPod | H100 80GB SXM | EU-Central | ⚠ Limited | 1-3h | Wait for capacity |
| RunPod | A100 80GB | US-East | ✓ Available | — | Both Community and Secure |
| RunPod | A100 80GB | EU-Central | ✓ Available | — | Secure Cloud |
| RunPod | RTX 4090 | Global | ✓ Available | — | Always available via Community |
| Lambda Labs | H100 80GB SXM | US-West | ✓ Available | — | On-demand from $2.49/h |
| Lambda Labs | H100 80GB SXM | US-East | ⚠ Limited | Same day | Capacity tightens during business hours |
| Lambda Labs | H100 80GB SXM | EU | ⌛ Waitlist | 1-3 days | Limited EU capacity |
| Lambda Labs | A100 80GB | US-West | ✓ Available | — | — |
| Lambda Labs | A100 80GB | US-East | ✓ Available | — | — |
| CoreWeave | H100 80GB SXM | US-Midwest | ✓ Available | Sales call | Reservations only — sales-led process |
| CoreWeave | H100 80GB SXM | EU-Amsterdam | ⚠ Limited | 1-2 weeks | Larger clusters available with notice |
| CoreWeave | A100 80GB SXM | US | ✓ Available | — | — |
| Vast.ai | H100 80GB | Global | ✓ Available | — | Marketplace — varies by host |
| Vast.ai | A100 80GB | Global | ✓ Available | — | Community marketplace |
| Vast.ai | A100 40GB | Global | ✓ Available | — | — |
| Vast.ai | RTX 4090 | Global | ✓ Available | — | Always abundant |
| Vast.ai | RTX 3090 | Global | ✓ Available | — | — |
| Paperspace | A100 80GB | US | ⚠ Limited | Same day | Frequent stockouts during US business hours |
| Paperspace | A100 40GB | US | ✓ Available | — | — |
| Paperspace | A100 40GB | EU | ⚠ Limited | Same day | — |
| Hetzner GPU | A100 PCIe 40GB | Germany (Falkenstein) | ✓ Available | — | Hourly + monthly billing |
| Hetzner GPU | A100 PCIe 40GB | Finland (Helsinki) | ✓ Available | — | — |
| OVH GPU | A100 80GB | France (Gravelines) | ✓ Available | — | EU sovereign cloud |
| OVH GPU | T4 | France (Gravelines) | ✓ Available | — | — |
| OVH GPU | V100 | France (Gravelines) | ✓ Available | — | — |
| Google Cloud | H100 80GB (A3) | us-central1 | ✗ Stockout | Unknown | On-demand frequently unavailable; reservations required |
| Google Cloud | H100 80GB (A3) | us-east5 | ⚠ Limited | Hours | Capacity returns intermittently |
| Google Cloud | H100 80GB (A3) | europe-west4 | ✗ Stockout | Unknown | — |
| Google Cloud | A100 80GB (a2) | us-central1 | ⚠ Limited | Hours | — |
| Google Cloud | A100 40GB (a2) | us-central1 | ✓ Available | — | — |
| Google Cloud | TPU v5e | us-central1 | ✓ Available | — | Strong TPU availability |
| AWS | H100 (p5.48xlarge) | us-east-1 | ✗ Stockout | Unknown | Reservations required for guaranteed access |
| AWS | H100 (p5.48xlarge) | us-west-2 | ⚠ Limited | Hours-Days | Best on-demand region for H100 |
| AWS | H100 (p5.48xlarge) | eu-west-1 | ✗ Stockout | Unknown | — |
| AWS | A100 (p4d/p4de) | us-east-1 | ✓ Available | — | — |
| AWS | A100 (p4d/p4de) | us-west-2 | ✓ Available | — | — |
| Azure | H100 (NDA100 v5) | East US | ⚠ Limited | 1-3 days | Quota request needed |
| Azure | H100 (NDA100 v5) | West Europe | ✗ Stockout | Unknown | — |
| Azure | A100 80GB (NDv4) | East US | ✓ Available | — | — |
| Azure | A100 80GB (NDv4) | West Europe | ✓ Available | — | — |
Snapshot from April 30, 2026. Status is determined from public provisioning attempts and provider status pages — actual availability changes minute-to-minute. Marketplace clouds (Vast.ai) reflect host-pool size, not guaranteed individual hosts.
Why H100 is so hard to get in 2026
Frontier model demand
OpenAI, Anthropic, Meta, Google and xAI absorb most NVIDIA Hopper allocation. CoreWeave and Lambda Labs receive the largest specialist allocations; AWS, GCP and Azure prioritize internal teams + reservations.
Reservation-first economics
Hyperscalers bias H100 capacity toward reserved (1-3 year commit) customers. On-demand is the leftover. This is why AWS p5 has more stockouts than RunPod despite a 100× larger fleet.
EU is harder
EU H100 capacity is concentrated in a handful of datacenters (Amsterdam, Frankfurt, Dublin). For European workloads, CoreWeave EU and RunPod EU are the most reliable; Lambda Labs has limited EU presence.
Workarounds
Use A100 80GB if your model fits ≤70B parameters (much more available). Use spot/preemptible for fault-tolerant training (50-70% off + better availability). Or rent from Vast.ai marketplace — never officially "out of stock."
Get notified when availability changes
Stop checking manually. We'll email you when H100 or A100 on-demand capacity opens up at your preferred provider.
No spam. Unsubscribe any time.