GPU Cloud

NVIDIA H100, H200, B200 and B300 capacity inside the Kingdom - by the hour, by the month, or dedicated to you.

Reserve capacity All services

Available GPU Models

From Hopper to Blackwell Ultra. Pick the memory footprint and bandwidth your workload needs - provisioned on demand, no procurement cycle.

NVIDIA Hopper

H100 Hopper

• Memory80 GB HBM3
• Bandwidth3.35 TB/s

H200 Hopper

• Memory141 GB HBM3e
• Bandwidth4.8 TB/s

NVIDIA Blackwell

B200 Blackwell

• Memory192 GB HBM3e
• Bandwidth8 TB/s

B300 Blackwell Ultra

• Memory288 GB HBM3e
• Bandwidthcontact for availability

Non-Blocking GPU Fabric

Every accelerator sits one hop from every other across a rail-optimized spine. No oversubscription, no contention - just line-rate collective bandwidth for training at pod scale.

SPINE

256-GPU POD - NON-BLOCKING FABRIC

0

Per-GPU Fabric

0

Top-tier SLA Target

<0

Provisioning

Consumption Models

Match your commitment to your workload - burst on-demand, reserve committed capacity, or run a private cluster behind your own perimeter.

On-Demand 01

• Per-hour billing
• Self-serve via Launchpad
• Scale to zero

Reserved 02

• 1-36 month terms
• Priority capacity
• Committed-use pricing

Dedicated 03

• Private bare-metal clusters
• Custom SLA
• Your security perimeter

Product - Launchpad

Spin up your first instance with Launchpad

Open Launchpad →

Reserve Capacity

Reserve capacity