Vultr is a global cloud provider with 30+ data-center regions. Its Cloud GPU platform offers fractional and full NVIDIA GPUs (A16, A40, L40S, A100, H100, GH200) and AMD Instinct GPUs, billed by the hour, plus bare-metal GPU servers and managed Kubernetes.
Table of Contents
Vultr GPU at a Glance
- 32 global data-center regions
- Fractional GPUs (slices of A16, A40, A100)
- NVIDIA L40S, A100, H100, GH200 and AMD MI300X/MI325X
- Fractional A16/A40 GPUs from a few cents/hr · A100, L40S, H100 and GH200 on demand
Pros
- Low-cost fractional GPUs for small AI models
- Serve AI close to users worldwide
- Kubernetes, object storage and CDN in one place
Cons
- Big GPUs not available in every region
- Pricing is higher than GPU marketplaces
Vultr GPU Pricing (2026)
Vultr prices change by region and commitment. Third-party trackers list the HGX H100 at roughly $2.30 per GPU-hour, with fractional A16 and A40 GPUs starting at a few cents per hour. Reserved and committed-use plans cut prices by 20–35%.
| GPU | VRAM | Best For |
|---|---|---|
| A16 / A40 (fractional) | 2–48 GB slices | Small AI models, VDI, light inference |
| L40S | 48 GB | LLM inference, image generation |
| A100 | 40 / 80 GB | Training and large-model inference |
| H100 / GH200 | 80 / 96 GB | Large LLMs, fine-tuning |
| AMD MI300X / MI325X | 192 / 256 GB | Very large models with lots of VRAM |
Key Features
- Fractional GPUs: pay for a slice of a GPU, ideal for small models like Gemma 4 E4B or embedding services.
- Global reach: deploy AI inference close to users on every continent.
- Full cloud platform: managed Kubernetes, load balancers, object storage and databases.
Who Should Use Vultr?
Vultr is best for developers who want to serve AI apps globally and like a simple, all-in-one cloud. For the cheapest raw GPU hours, RunPod and Vast.ai are usually cheaper; for 24/7 monthly servers, see GPU Mart.
FAQ
A fractional GPU is a slice of a physical GPU (for example part of an A16 or A40) with its own VRAM. It is much cheaper than a full GPU and enough for small AI models or remote desktops.
Yes, Vultr offers HGX H100 and GH200 instances, as well as AMD Instinct MI300X and MI325X.
Verdict
Vultr is a flexible, developer-friendly GPU cloud with great global coverage and cheap fractional GPUs. It is a strong choice for globally distributed AI inference.
Prices were checked in September 2026 and change often. Always confirm current pricing on the provider’s website before you order.