Cheapest GPU cloud for AI training

A training run holding H100s for days at a stretch, where the hourly card rate is essentially the entire bill.

List prices checked

How this is scored

Scored on 500 H100-hours a month — roughly one card held for three weeks solid. On-demand rates only: spot and reserved pricing is excluded because hosts define preemption and availability so differently that the numbers are not comparable, and a cheap rate you can be evicted from is not the same product.

All 12, ranked for this workload

#ServiceProviderCost for this workloadNotes
1Vast.ai H100Vast.ai$825/mo at 500 H100-hours—
2Together AI H100Together AI$875/mo at 500 H100-hours—
3Hyperstack H100Hyperstack$950/mo at 500 H100-hours—
4RunPod H100RunPod$995/mo at 500 H100-hours—
5Paperspace H100Paperspace$1,120/mo at 500 H100-hours—
6CoreWeave H100CoreWeave$1,155/mo at 500 H100-hours—
7Lambda H100Lambda$1,245/mo at 500 H100-hours—
8Modal H100Modal$1,975/mo at 500 H100-hours—
9Replicate H100Replicate$2,520/mo at 500 H100-hours—
10Google Cloud A3Google$3,490/mo at 500 H100-hours—
11Azure ND H100 v5Microsoft$3,490/mo at 500 H100-hours—
12AWS EC2 P5Amazon Web Services$3,785/mo at 500 H100-hours—

Published list prices only. Rows that do not publish a rate for an axis this workload depends on are excluded rather than shown as cheap.

Questions

Cheapest GPU cloud for AI training in 2026?
Vast.ai H100 from Vast.ai — $825/mo at 500 H100-hours. Together AI H100 is second at $875/mo at 500 H100-hours. This ranking is specific to the workload described above; a different usage shape reorders it.
How is this ranking weighted?
Scored on 500 H100-hours a month — roughly one card held for three weeks solid. On-demand rates only: spot and reserved pricing is excluded because hosts define preemption and availability so differently that the numbers are not comparable, and a cheap rate you can be evicted from is not the same product.
Why does this differ from the general cheapest list?
Because a single price axis never describes a real workload. Ranking by headline rate answers "who is cheapest per unit"; this page answers "who is cheapest for this job", and the two orders are often very different — which is the whole reason it exists as a separate table.
What is not accounted for?
Committed-use discounts, regional price variation, per-request charges, support plans, and minimum retention. These are published list prices applied to one stated workload — a shortlist to price properly with the vendor, not a quotation.

Other workloads