One contract,
sized per cluster.
Every cluster follows the same contract shape. What changes is the size, the region and how fast it ramps.
NVIDIA HGX B300 on a 60-month term. The rate is quoted per deal: size, region and how much you prepay lower it.
- Cluster size and wave schedule
- Region and Tier III facility
- Network design
- Operating and support terms
- Your exit and rollover terms
- HGX B200
- FromUS$3.00/ GPU-hour
- 8 GPUs per node180 GB HBM3e per GPU
- GB200 NVL72
- FromUS$4.00/ GPU-hour
- 72 GPUs per rack186 GB HBM3e per GPU
- GB300 NVL72
- FromUS$5.00/ GPU-hour
- 72 GPUs per rack288 GB HBM3e per GPU
- VR200 NVL72Waitlist
- FromUS$6.50/ GPU-hour
- 72 GPUs per rack288 GB HBM4 per GPU
All rates are from-rates per GPU-hour on 60 months, prepaid. VR200 NVL72 is a waitlist for first allocation, not available capacity.
NVIDIA hardware available to reserve.
Clusters run on NVIDIA Blackwell today. Rollover is how a contract gets to what comes after.
NVIDIA HGX B300
Eight Blackwell Ultra GPUs per node with 288 GB of HBM3e each, reserved by the node. Root on every node in bare metal mode.
NVIDIA GB300 NVL72
72 Blackwell Ultra GPUs and 36 Grace CPUs in one rack, as a single NVLink domain with 20.7 TB of memory.
The same terms for every cluster.
Size, region and ramp are set per deal. The contract structure is the same.
- Term
- 60 monthsThe only term. There are no shorter ones.
- Minimum size
- 8 HGX nodes (64 GPUs) or one NVL72 rack (72 GPUs)Most clusters are 32 to 128 nodes; larger ones are scoped case by case. NVL72 is reserved as whole racks.
- Delivery
- Large clusters in monthly waves of 32 nodesHGX B300 about 10 to 14 weeks from signing. Dates are always counted from signing.
- Exit
- Option in the contractYour proposal sets the terms.
- Rollover
- Into a newer GPU generationPart of the commitment can move into a newer GPU generation when it ships.
- Prepayment
- Required, held in escrowIt is paid into escrow with an independent escrow agent, which releases it in stages as supply, installation and acceptance milestones are met. From the first full month, it is credited in equal parts against each monthly invoice over the term. There is no security deposit.
- Transfer
- Possible with Lightyard's consent
- Network
- RoCE v2 Ethernet, spine-leafNon-blocking on request.
- Facilities
- Tier III, in the region you chooseUS West, US Central, US East, Europe, East Asia, Southeast Asia and South America
- Operation
- Bare metal or managedSame single-tenant hardware either way. We monitor it, replace failed parts and run it to an SLA.
Hourly capacity
We work best on committed, long-term capacity. If you need GPUs by the hour or month to month, this isn't the right contract. Tell us what you need and we'll point you in the right direction.
Before you ask for a proposal.
Anything else on pricing, ask. We usually reply within a few hours.
On a 60-month term, per GPU-hour: HGX B300 from US$3.50 / GPU-hr; HGX B200 from US$3.00 / GPU-hr; GB200 NVL72 from US$4.00 / GPU-hr; GB300 NVL72 from US$5.00 / GPU-hr. VR200 NVL72 is on the waitlist for first allocation, from US$6.50. The rate is quoted per deal.
Capacity that is committed and prepaid is priced lower than capacity billed month to month. How much it moves the rate is set in your proposal.
It is paid into escrow with an independent escrow agent, which releases it in stages as supply, installation and acceptance milestones are met. From the first full month, it is credited in equal parts against each monthly invoice over the term. There is no security deposit.
The contract includes an exit option, on terms set out in your proposal. Otherwise the 60-month term runs as signed.
No. Every cluster is reserved for 60 months, with an exit option. There are no shorter or pilot terms.
Per deal, on the GPU, the size, the region and how much of the contract you prepay. Tell us what you need and the proposal comes back with a rate and delivery dates counted from signing.
The minimum is 8 HGX nodes (64 GPUs) or one NVL72 rack (72 GPUs). Most clusters are 32 to 128 nodes; larger ones are scoped case by case. If your number is different, ask anyway.