What is the cheapest B200 rental?
The B200 is NVIDIA's Blackwell datacenter GPU and sits at the top of the market, so it rents well above Hopper cards like the H100. Early Blackwell supply is tight, which keeps rates high and availability uneven across providers. The table below shows the cheapest verified B200 listings we currently track.
| GPU | VRAM | $/hr | Where | |
|---|---|---|---|---|
| B200 | 180 GB | $3.50 | Vultr on-demand | Rent → |
| B200 | 180 GB | $5.89 | RunPod secure cloud | Rent → |
| B200 | 180 GB | $5.98 | RunPod community cloud | Rent → |
| B200 | 180 GB | $6.00 | Hyperstack on-demand | Rent → |
| B200 | 180 GB | $6.11 | DataCrunch on-demand | Rent → |
| B200 | 180 GB | $6.25 | Modal serverless | Rent → |
| B200 | 180 GB | $6.25 | fal serverless | Rent → |
| B200 | 180 GB | $6.69 | Lambda on-demand | Rent → |
The B200 is a large step up from the H100 in both compute and memory, aimed at frontier-scale training and the largest inference workloads. For those jobs it can finish faster and, despite the higher hourly rate, cost less per unit of work than an older card. For anything that already runs comfortably on an H100 or A100, the B200 is overkill and the premium is hard to justify.
Because Blackwell is new, prices move more than for established cards as capacity comes online and providers compete. Expect the cheapest rate to shift week to week, and expect the gap between a hyperscaler and a specialist provider to be wide.
As with every card here, we only list B200 rates we can verify, and we flag the reliability tier so an early community listing is not mistaken for a firm production rate.
Related questions
- What is the price difference between the H100 and H200?
- How much does an H100 cost per hour?
- How much does it cost to train an LLM?
Numbers on this page come from today's verified snapshot. Full table on the homepage; method in the methodology.