GPU Compute

On-demand NVIDIA A100 and H100 instances for model training, fine-tuning and inference. NVLink interconnect, fast local NVMe scratch and pre-baked ML images.

Highlights

  • A100 & H100 GPUs
  • NVLink interconnect
  • Pre-baked ML images
  • NVMe scratch disk

Choose your gpu compute plan

All plans are billed once for a 12-month term.

A100 Spot

Cost-efficient interruptible A100.

$2,999 / 12 mo

≈ $249.92 / month · billed upfront

Get A100 Spot Download USD invoice (PDF)
  • 1× NVIDIA A100 40GB
  • 12 vCPU / 85 GB RAM
  • 500 GB NVMe scratch
  • Pre-baked ML images
Most popular

A100 Dedicated

Reserved A100 for steady workloads.

$6,999 / 12 mo

≈ $583.25 / month · billed upfront

Get A100 Dedicated Download USD invoice (PDF)
  • 2× NVIDIA A100 80GB
  • 24 vCPU / 170 GB RAM
  • 1 TB NVMe scratch
  • NVLink interconnect
  • Priority support

H100 Cluster

Flagship H100 cluster for large models.

$14,999 / 12 mo

≈ $1,249.92 / month · billed upfront

Get H100 Cluster Download USD invoice (PDF)
  • 8× NVIDIA H100 80GB
  • 96 vCPU / 640 GB RAM
  • 4 TB NVMe scratch
  • 3.2 Tbps InfiniBand
  • Dedicated TAM

Compare variants

Specification A100 Spot $2,999 / yr A100 Dedicated $6,999 / yr H100 Cluster $14,999 / yr
GPU 1× A100 40GB 2× A100 80GB 8× H100 80GB
vCPU 12 cores 24 cores 96 cores
Memory 85 GB 170 GB 640 GB
Scratch 500 GB NVMe 1 TB NVMe 4 TB NVMe