How much does an NVIDIA H100 GPU cost?
The NVIDIA H100 is a high-performance GPU built on the Hopper architecture. It is designed for demanding AI workloads such as large language model training, high-throughput inference, and data-intensive processing. With up to 4.9 TB per second of memory bandwidth and support for FP8 precision, the H100 delivers significant improvements over the previous generation A100.
If you are trying to determine the cost of using an H100, the answer depends on several factors. Costs vary depending on whether you purchase the hardware outright, rent it in the cloud, or pay for just the GPU or an entire system that includes CPU, RAM, and storage.
This guide outlines H100 pricing across major providers and explains how Northflank offers one of the most developer-friendly, fully integrated setups.
💭 What is Northflank?
Northflank is a full-stack AI cloud platform that helps teams build, train, and deploy models without infrastructure friction. GPU workloads, APIs, frontends, backends, and databases run together in one place so your stack stays fast, flexible, and production-ready.
Sign up to get started or book a demo to see how it fits your stack.
NVIDIA does not publish direct retail pricing, but you can find H100 GPUs through OEMs and authorized resellers. As of mid-2025:
- H100 80GB PCIe typically ranges from $25,000 to $30,000
- H100 80GB SXM ranges from $35,000 to $40,000
PCIe vs SXM
PCIe is easier to deploy and shows up in more off-the-shelf systems. SXM offers better performance with higher bandwidth and power, often used in tightly coupled multi-GPU servers.
Pricing depends on configuration, cooling, and whether the purchase includes a full server. For example, an eight-GPU DGX H100 system can exceed $300,000. These setups require specialized infrastructure, so many teams opt to rent H100s in the cloud instead.
If you're curious how the H100 stacks up against the A100, check out this article.
To make sense of H100 pricing, you need to account for more than just the GPU itself. Many platforms list low hourly rates but charge separately for the CPU, RAM, and storage needed to run workloads. Others offer bundled pricing but with tradeoffs in performance or stability.
Here’s how H100 hourly on-demand pricing break down across several popular platforms:
| Provider | H100 SXM (USD/hr) | Notes |
|---|---|---|
| Northflank | $2.74 | Fully bundled (GPU, CPU, RAM, storage). Fast startup, no quota required, full-stack AI platform. |
| Modal | $3.95 | GPU-only pricing. CPU and RAM billed separately. Serverless model execution. |
| RunPod | $4.18 | GPU only. Setup takes time, and automation is limited |
| Fireworks AI | $5.80 | GPU-only pricing for hosted model serving. No quotas. Fast auto-scaling. |
| Baseten | $6.50 | Fully managed model hosting. Includes CPU, RAM, and NVMe storage. |
| AWS | $7.57 | May Require quota approval. Bundled node (CPU, RAM, disk). Startup takes minutes. |
| GCP | $11.06 | GPU bundled with VM (CPU, RAM, disk). Requires regional GPU quota. |
| Azure | $6.98 | Pricing includes CPU, RAM, and storage. Quotas apply. |
| OCI | $10 | Bare-metal with full machine access (CPU, RAM, NVMe). Quota may be required. |
| Lambda Labs | $3.29 | Bundled pricing. Full-node access (CPU, RAM, storage) |
Many platforms make it hard to see what you're getting. You often have to manage CPUs, memory, storage, or deal with quota approvals before you can even start. That adds friction and slows you down.
Northflank keeps things simple. You get access to H100s with everything included. GPU, CPU, memory, and storage are already set up, so you can focus on running your code, not configuring infrastructure.
Northflank is also more than just GPU hosting. It’s a full-stack platform for AI teams. You can train models, serve APIs, run frontends or backends, and manage databases all in one place. Built-in CI, logs, metrics, and autoscaling help you move faster from idea to production without switching tools or writing extra config.
If you're looking for speed, simplicity, and a complete setup that works, Northflank gives you a better way to build and deploy AI.
The H100 is a powerful GPU, but cost and usability depend entirely on where you run it. Some platforms bury you in hidden fees or make setup painful. Others seem affordable but fail on reliability.
Northflank gives you fast, consistent access to H100s with clear pricing and no extra complexity. You get everything in one place: GPU, CPU, RAM, and storage, already configured and production-ready.
If you're ready to try it yourself, sign up and deploy your first H100. If you want to see how it fits your workflow, book a quick demo.
