Monthly GPU servers: every GPU we rent

How monthly GPU server rental works

A monthly GPU server is a whole GPU, or eight, reserved for you for a month at one flat price. You pick a configuration on the deployment page, choose a root password or add your SSH key, and pay the first month from a balance you top up in crypto. The server is delivered in under 10 minutes after payment.

1. Choose a GPU and a configuration. Single-GPU servers run in a dedicated VM with full GPU passthrough. 8-GPU servers are bare metal, with NVLink between the GPUs (Infinity Fabric on the AMD MI355X).
2. Pay the first month. Top up in Bitcoin, Ethereum, Monero, USDT on TRON or Solana, with up to 50% bonus credit on larger top-ups.
3. Get root access within 10 minutes of payment. The IP address and SSH details appear in your dashboard, on Ubuntu with NVIDIA drivers, CUDA and Docker preinstalled, or ROCm on AMD.
4. Renew or cancel month by month. The plan renews from your balance, with a reminder before each renewal, and you can cancel before the next month at no fee.

What every monthly price includes

• Whole GPUs Never shared, sliced with vGPU or partitioned with MIG
• CPU, RAM and NVMe As listed on each configuration, local to the server
• Bandwidth 20 TB a month on single-GPU servers and 50 TB on 8-GPU servers, with no egress fees within it
• Network A public IPv4 address, an IPv6 range and always-on DDoS protection
• Software OS images with NVIDIA drivers, CUDA, Docker and the NVIDIA Container Toolkit, or ROCm on AMD
• Support GPU engineers 24/7, through tickets in your dashboard

There is no setup fee, and the 99.9% availability SLA covers every server. The price on each card is the full monthly price; the per-GPU-hour figure next to it only helps you compare with hourly offers.

Monthly vs hourly: when a monthly server pays off

Divide the monthly price by the hours you will really use the GPU. A single H100 at $2,190 a month works out to $3.00 per GPU-hour if it runs all 730 hours of an average month, and twice that if it runs half the time. If an hourly offer costs less than that figure at your real usage, with storage and egress included, hourly is cheaper; if not, monthly is. Inference endpoints, training runs that last weeks and shared team machines usually land on the monthly side.

The GPU cost calculator runs this for any hourly quote, and monthly vs hourly GPU pricing walks through the break-even math.

Single GPU or 8-GPU server?

A single-GPU server suits one model replica, one fine-tuning job or one developer: serving models that fit in one GPU's memory, LoRA and QLoRA, image generation and rendering. An 8-GPU server is one bare-metal machine with eight GPUs linked by NVLink, for training and for serving models too large for one GPU with tensor parallelism. Eight separate single-GPU servers do not share NVLink, so a model split across GPUs needs the 8-GPU server.

Beyond eight GPUs in one NVLink domain, rack-scale systems such as the GB300 NVL72 are reserved on 12 to 36 month terms. To see how much memory your model needs and which configurations fit, use the VRAM calculator or the LLM GPU requirements tables.

Availability and delivery time

Stock is shown live on every card above. Single-GPU and 8-GPU servers are delivered in under 10 minutes after payment, even when stock is limited. The delivery guide explains how delivery works, and our GPU shortage guide explains why capacity is tight across the industry in 2026.