NVIDIA H200 Pricing
Eli Mernit
An NVIDIA H200 rents for $2.09 to $10.60 per GPU-hour on demand. Beam's on-demand machine is the low end at $2.09 per hour, other specialist clouds charge $3.50 to $4.40, and AWS and Azure 8-GPU instances work out to $7.91 and $10.60 per GPU. The median across 36 providers tracked by getdeploying is about $4.45.
Key takeaways
- Beam's $2.09 machine is the only on-demand H200 here under $3.50. Atlas Cloud ($3.50) and RunPod ($3.59) come next.
- Runcrate, Hyperstack, AWS and Azure list the H200 only in 8-GPU configurations.
- H200 rates are climbing: the getdeploying median is up about 3% over 90 days and about 26% over 12 months.
H200 cloud prices per hour, by provider
| Provider | SKU | On-demand $/GPU-hr | Notes |
|---|---|---|---|
| Beam | H200 SXM5 | $2.09 | On-demand machine; vCPU, RAM, and NVMe included |
| Atlas Cloud | H200 | $3.50 | 1 to 8 GPUs per instance |
| RunPod | H200 SXM (Community Cloud) | $3.59 | Per-second billing; Secure Cloud $4.59 |
| Hyperstack | H200 SXM | $3.99 | 8-GPU configuration; per-minute billing; reserved from $2.79; out of stock (getdeploying) |
| Runcrate | H200 SXM5 | $4.40 | One listing, an 8x node in Tokyo; per-second billing |
| AWS | p5en.48xlarge (8x H200) | $7.91 | Full 8-GPU instance ($63.30/hr), US East |
| Azure | ND96isr H200 v5 (8x H200) | $10.60 | Full 8-GPU VM ($84.80/hr), US East 2; $13.78/GPU in West US |
| Market median | — | $4.45 | Across 36 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 and reserved rates excluded. We used single-GPU shapes where offered and divided 8-GPU instance prices by eight.
- Beam pricing page
- Atlas Cloud GPU pricing page
- RunPod GPU pricing page
- Hyperstack GPU pricing page, including reserved H200 rates
- Runcrate pricing page
- AWS EC2 on-demand pricing page
- Azure virtual machine pricing page (ND96isr H200 v5)
- getdeploying H200 price tracker for the market median
H200 vs H100: when the extra memory pays off
The H200 has the same Hopper compute as the H100 SXM, so the upgrade is memory: 141 GB of faster HBM3e against 80 GB, about 76% more. LLM decoding reads the weights for every token. Faster memory means more tokens per second, and the extra room fits longer contexts and bigger batches. On Beam the step up is small. The H200 machine costs $0.26 more per hour than the H100 PCIe machine, and it is cheaper per GB of memory, about 1.5 cents per GB-hour against 2.3 cents.
Where to rent an H200 for less
Nobody in this data undercuts Beam's $2.09 on-demand rate, which includes vCPU, RAM, and NVMe. Atlas Cloud ($3.50, one to eight GPUs per instance) and RunPod Community Cloud ($3.59, billed per second) are the next cheapest. Hyperstack's reserved H200 starts at $2.79 but takes a commitment, and its on-demand configuration is out of stock on getdeploying. Skip the 8-GPU-only offers unless you need all eight cards: AWS and Azure bill $63.30 and $84.80 per hour for a full instance.
What this means for your workload
Rent an H200 when memory is the limit: a model too large for one 80 GB card, long-context serving with a growing KV cache, or a busy endpoint that wants big batches. Behind an OpenAI-compatible LLM API, one H200 can replace a pair of H100s whenever the weights fit in 141 GB, with no tensor-parallel overhead.
FAQ
How much does an H200 cost per hour?
It costs $2.09 per GPU-hour on Beam and $3.50 to $4.40 at other specialist clouds, on demand. AWS and Azure 8-GPU instances work out to $7.91 and $10.60 per GPU.
What's the cheapest way to rent an H200?
Beam's $2.09 on-demand machine is the lowest price in our table, with vCPU, RAM, and NVMe included. Among other providers, Atlas Cloud ($3.50) and RunPod Community Cloud ($3.59) are the cheapest, and both rent single GPUs.
H200 vs H100: which should I rent?
Pick the H100 if your model fits in 80 GB with room to spare, since it costs less per hour. Pick the H200 when you need 141 GB or faster memory for large models, long contexts, or big batches.
