GPUPriceBook

Best spot & community GPU prices

Spot, community and interruptible tiers cut GPU rental prices by roughly 30-65% in exchange for capacity that can be reclaimed. The lowest discounted rate we track is $0.20/hr for a NVIDIA RTX 4090 on Vast.ai. Use these tiers for checkpointed training, batch inference and sweeps — not for latency-critical serving. Every published discounted rate is ranked below, with the saving vs on-demand.

Source: Provider pricing pages. Data as of June 2026.

Discounted GPU rates ranked cheapest first

RankGPUProviderSpot /hrOn-demand /hrSavingTier name
#1RTX 4090Vast.ai$0.20/hr$0.52/hr62%Interruptible
#2L4Google Cloud$0.33/hr$0.71/hr54%Spot
#3RTX 4090RunPod$0.34/hr$0.69/hr51%Community Cloud
#4A100 80GBVast.ai$0.35/hr$0.74/hr53%Interruptible
#5A6000Hyperstack$0.40/hr$0.50/hr20%Spot
#6L4AWS EC2$0.43/hr$0.81/hr47%Spot
#7V100Microsoft Azure$0.57/hr$3.06/hr81%Spot
#8L40SNebius$0.74/hr$1.55/hr52%Reserved (preemptible)
#9A100 40GBMicrosoft Azure$0.75/hr$3.40/hr78%Spot
#10A100 80GBMicrosoft Azure$0.82/hr$3.67/hr78%Spot
#11RTX 6000 AdaNebius$0.95/hr$1.80/hr47%Reserved (preemptible)
#12H100 SXMVast.ai$0.95/hr$1.93/hr51%Interruptible
#13A100 80GBHyperstack$1.08/hr$1.35/hr20%Spot
#14L40SAWS EC2$1.27/hr$1.86/hr32%Spot
#15H100 PCIeVast.ai$1.40/hr$2.78/hr50%Interruptible
#16A100 80GBAWS EC2$1.41/hr$3.43/hr59%Spot
#17A100 40GBAWS EC2$1.42/hr$2.75/hr48%Spot
#18RTX 6000 AdaHyperstack$1.44/hr$1.80/hr20%Spot
#19H100 PCIeHyperstack$1.52/hr$1.90/hr20%Spot
#20H100 PCIeMicrosoft Azure$1.75/hr$6.98/hr75%Spot
#21A100 40GBGoogle Cloud$1.84/hr$3.67/hr50%Spot
#22H100 SXMRunPod$1.99/hr$3.29/hr40%Community Cloud
#23H100 PCIeRunPod$1.99/hr$2.39/hr17%Community Cloud
#24H100 SXMNebius$2.15/hr$3.85/hr44%Reserved (preemptible)
#25H100 SXMMicrosoft Azure$2.27/hr$12.29/hr82%Spot
#26H200AWS EC2$2.38/hr$7.91/hr70%Spot
#27H200Nebius$2.45/hr$4.50/hr46%Reserved (preemptible)
#28A100 80GBGoogle Cloud$2.53/hr$5.07/hr50%Spot
#29B200AWS EC2$2.69/hr$14.24/hr81%Spot
#30H100 SXMAWS EC2$3.37/hr$6.88/hr51%Spot
#31B200Nebius$3.95/hr$7.15/hr45%Reserved (preemptible)
#32H100 SXMGoogle Cloud$4.98/hr$10.98/hr55%Spot
#33H200Google Cloud$5.35/hr$10.85/hr51%Spot

Source: Provider pricing pages. Data as of June 2026.

Snapshot June 2026 — cloud GPU prices change weekly; verify on the provider's pricing page before relying on a figure.

Frequently asked questions

What are the best spot / interruptible GPU prices right now?

The lowest discounted rate we track is $0.20/hr for a NVIDIA RTX 4090 on Vast.ai (Interruptible). Spot/community/interruptible tiers typically cut on-demand prices by 30-65%. Snapshot June 2026 — cloud GPU prices change weekly; verify on the provider's pricing page before relying on a figure.

What's the difference between spot, community and interruptible?

They are all discounted, reclaimable capacity. AWS/GCP/Azure call it 'Spot'; Vast.ai calls it 'Interruptible'; RunPod's cheaper tier is 'Community Cloud'; Nebius/Verda publish 'reserved/preemptible' rates. The common trade-off: lower price in exchange for the instance possibly being taken back.

When should I use spot GPUs?

Use spot for fault-tolerant, checkpointed or batch work: training that saves checkpoints, offline/batch inference, hyperparameter sweeps and data processing. Avoid it for latency-critical online serving or long unattended runs where a mid-run reclaim is expensive.

How much can spot save me?

Usually 30-65% off the on-demand rate, varying by provider and supply. The savings column above shows the discount for each row where both prices are published. Marketplace interruptible (Vast.ai) can be the deepest discount but also the most variable.

Keep exploring

Last updated: 2026-06-21