NVIDIA B200 Pricing
Nathanael Chiang
An NVIDIA B200 rents for $4.11 to $14.24 per GPU-hour on demand across the nine providers below. Beam's on-demand machine is the low end at $4.11 per hour. Other GPU clouds list $6.00 to $8.60, while AWS and Oracle sell only 8-GPU shapes at $14.24 and $14.00 per GPU. The median across 20 providers tracked by getdeploying is about $6.70.
Key takeaways
- Beam's $4.11 machine is the cheapest on-demand B200 here. RunPod Community Cloud is next at $5.98.
- CoreWeave, AWS, Oracle and Hyperstack rent the B200 only as 8-GPU nodes.
- Preemptible capacity is far cheaper: Nebius starts at $0.99 per GPU-hour.
B200 cloud prices per hour, by provider
| Provider | SKU | On-demand $/GPU-hr | Notes |
|---|---|---|---|
| Beam | B200 SXM6 | $4.11 | On-demand machine; vCPU, RAM, and NVMe included |
| Hyperstack | HGX B200 | $6.00 | 8-GPU flavor, sold out (getdeploying); per-minute billing; reserved from $5.10 |
| RunPod | B200 SXM6 (Secure Cloud) | $6.79 | Per-second billing; Community Cloud $5.98 |
| Lambda | 1x B200 (SXM6) | $6.99 | 8x node $6.69/GPU; out of stock (getdeploying) |
| Nebius | HGX B200 (SXM6) | $7.15 | Scheduled to rise to $8.50; preemptible from $0.99 |
| Spheron | B200 SXM6 | $8.43 | Live marketplace rate; spot $4.42 |
| CoreWeave | HGX B200 (8-GPU instance) | $8.60 | 8-GPU node $68.80/hr; spot $34.11/hr |
| Oracle Cloud (OCI) | BM.GPU.B200.8 (8x B200 bare metal) | $14.00 | 8-GPU bare-metal shape ($112.00/hr) |
| AWS | p6-b200.48xlarge (8x B200) | $14.24 | Full 8-GPU instance ($113.93/hr), US East |
| Market median | — | $6.70 | Across 20 providers with an on-demand price (getdeploying) |
How these prices were collected
Each figure is an on-demand list price per GPU-hour from the provider's own pricing page, with spot, preemptible and reserved rates excluded. We used single-GPU shapes where offered and divided 8-GPU node prices by eight.
- Beam pricing page
- Hyperstack GPU pricing page, including reserved B200 rates
- RunPod GPU pricing page
- Lambda instance pricing page
- Nebius price list, including preemptible GPU rates
- Spheron B200 rental page
- CoreWeave pricing page
- Oracle Cloud price list for the BM.GPU.B200.8 shape
- AWS EC2 on-demand pricing page
- getdeploying B200 price tracker for the market median
Why B200 cloud prices vary so widely
The biggest split is between hyperscalers and GPU clouds. AWS and Oracle charge about twice the $6.00 to $8.60 most GPU clouds list, and both rent 8-GPU shapes only, at $113.93 and $112.00 per hour. CoreWeave also sells whole nodes, at $68.80 per hour. Shape matters within a provider too: Lambda charges $6.99 per GPU on one card and $6.69 on eight. Marketplace rates move with the host. Spheron's cheapest on-demand B200 is a host in Finland at $8.43, and its US Central listing is $9.36.
Where to rent a B200 for less
Beam's $4.11 is the lowest on-demand B200 price in this data, with vCPU, RAM, and NVMe included. RunPod Community Cloud follows at $5.98, billed per second. Hyperstack ($6.00 on demand, $5.10 reserved) and Lambda both show as out of stock on getdeploying. The one rate below Beam's is preemptible: Nebius starts at $0.99 per GPU-hour, and those instances can be stopped at any time. CoreWeave spot ($4.26 per GPU) and Spheron spot ($4.42) both cost more than Beam on demand.
What this means for your workload
A B200 pays off on compute-heavy work: training runs where time to result matters, and large-model inference in FP8 or FP4. Lambda, CoreWeave and Oracle list 180 GB of memory per GPU; Hyperstack lists 192 GB. For training across a full node, see our guide to multi-GPU distributed training. If your model fits in 141 GB and isn't compute-bound, an H200 costs about half as much on Beam, $2.09 against $4.11.
FAQ
How much does a B200 cost per hour?
On demand, it runs from $4.11 per GPU-hour on Beam to $14.24 on AWS, where you rent a full 8-GPU instance. Most GPU clouds charge $6.00 to $8.60.
What's the cheapest way to rent a B200?
Beam's $4.11 on-demand machine is the lowest price we found, followed by RunPod Community Cloud at $5.98. If your job can handle interruptions, Nebius preemptible B200 starts at $0.99 per GPU-hour.
B200 vs H200: which should I rent?
Rent the H200 if your model fits in 141 GB and hourly cost matters most; on Beam it costs $2.09 against $4.11. Rent the B200 for compute-heavy training or FP4 inference, where the newer Blackwell chip finishes sooner.
