Hardware

NVIDIA platforms
you can reserve.

HGX platforms are reserved by the 8-GPU node. NVL72 platforms are reserved as whole racks. Every cluster is dedicated to one customer.

Platforms

Specs by platform.

Each platform is reserved on a 60-month term. Rates are on the pricing page.

  • Available

    NVIDIA HGX B300

    Generation
    Blackwell Ultra
    GPUs
    8 GPUs per node
    Memory
    288 GB HBM3e per GPU
    Cooling
    Air
    Availability
    about 10 to 14 weeks from signing
  • Available

    NVIDIA HGX B200

    Generation
    Blackwell
    GPUs
    8 GPUs per node
    Memory
    180 GB HBM3e per GPU
    Cooling
    Air
    Availability
    Set in the proposal
  • Available

    NVIDIA GB200 NVL72

    Generation
    Blackwell
    GPUs
    72 GPUs per rack
    Memory
    186 GB HBM3e per GPU
    Rack memory
    13.4 TB
    CPUs
    36 Grace per rack
    Cooling
    Liquid
    Availability
    Set in the proposal
  • Available

    NVIDIA GB300 NVL72

    Generation
    Blackwell Ultra
    GPUs
    72 GPUs per rack
    Memory
    288 GB HBM3e per GPU
    Rack memory
    20.7 TB
    CPUs
    36 Grace per rack
    Cooling
    Liquid
    Availability
    Set in the proposal
  • Waitlist

    NVIDIA VR200 NVL72

    Generation
    Rubin
    GPUs
    72 GPUs per rack
    Memory
    288 GB HBM4 per GPU
    Rack memory
    20.7 TB
    CPUs
    36 Vera per rack
    Cooling
    Liquid
    Availability
    Waitlist for first allocation

Rates and cluster sizesHow reserved clusters workContract terms

Get started

Start with one rack.

Tell us the size and timing you have in mind. We usually reply within a few hours and come back with a written proposal.

sales@lightyard.ai