Cutting-Edge NVIDIA Compute
Current NVIDIA parts — including B300, GB200, and B200 — for parallel work across training and inference.
GPUs in edge facilities
Close to bare metal.
Scaled like a cloud.
H100, A100, and RTX 4090 can start now. Train on a cluster or serve from one node, billed by the second.
Pricing
Compute Instances
Pick the machine for the job, not a package name.
Available
GB300 NVL72
GPU4
vRAM279GB
vCPUs144
RAM960GB
Local Storage61.44TB
$28
Spot /hr
Available
HGX B300
GPU8
vRAM270GB
vCPUs192
RAM4096GB
Local Storage61.44TB
$35.84
Spot /hr
Available
RTX PRO 6000 Blackwell
GPU8
vRAM768GB
vCPUs128
RAM1024GB
Local Storage7.68TB
$11.09
Spot /hr
Platform
What the compute actually includes
Elastic Configuration
Bare metal, elastic GPU, or a CPU VM. Set vCPU, memory, and disk for the job. Provisioning is measured in minutes.
Cost-Effective Scaling
Prices stay on the page, from $0.60 per GPU-hour. On-demand, weekly, monthly, or reserved — match the way you spend.
Global Low-Latency Network
A fast backbone ties sites together. When you need isolation, open a VPC of your own.
High-Performance Storage
High-IOPS storage can sit under a millisecond, for jobs that hammer the disk.
Enterprise-Grade Security
Firewalls and security groups follow the workload. You write the rules for traffic.
Workflow
From a chosen machine to a running job, in minutes
01
Select
Browse our AI Cloud and choose the instance type that perfectly matches your performance needs.
02
Configure
Customize your environment with automated storage and network provisioning for a seamless setup.
03
Pay
Choose a billing model that fits your budget—flexible prepaid options or postpaid terms for established workflows.
04
Deploy
Launch your fully provisioned instance with a single click. Your resources will be live and ready to use instantly.
05
Monitor
Track real-time usage and costs with clear monthly statements and 7-day forecasts.
06
Release
Spin down instances anytime you're done. Automatically release resources and stop billing immediately—pay only for what you use.
Infrastructure
Facilities built for dense compute
Network, power, and storage are sized for GPUs that stay busy, so a training run is not cut by the floor underneath it.
Infiniband Networking
3.2 Tbps non-blocking between nodes, so scale stays close to linear.
NVMe Storage Tiers
Local NVMe can read at 200GB/s, so a dataset does not sit and wait.
Cirruslink AI Node / Region US-EAST