Pricing
Priced per contract, sized to your workload.
Reserved NVIDIA B300 capacity in 256 GPU blocks and slices of them. Tell us the target and we reply with a term sheet: rate, term and a delivery window.
01
What you reserve
Three contracts, one ledger, the same hardware.
Reserved cluster
8 to 128 GPUs
- →Reserved from 12 months
- →Bare metal, managed Kubernetes
- →Named support
Capacity partner
128 GPUs and up
- →Wholesale or white label
- →Take or pay, 24 to 36 months
- →Non exclusive by default
Dedicated AI factory
Full 256 GPU block
- →One customer, one fabric
- →Take or pay, 24 to 36 months
- →First right on the next block
02
What sets the price
Six inputs. Send them and we size the cluster.
Size
GPUs and nodes reserved, from one node to multiple blocks.
Term
From 12 to 36 months. Longer terms get lower rates.
Contract type
Enterprise reserved cluster, capacity partner block or a dedicated factory.
Network
Ports, bandwidth and connectivity to US hubs.
Storage and support
Parallel file and object tiers, and the support level you need.
Start window
When you need the capacity live. The first block goes live Q2 2027.
03
Always included
In every contract
- ✓Dedicated, single tenant hardware
- ✓Managed Kubernetes on bare metal
- ✓Local NVMe on every node
- ✓Spectrum X fabric and low latency to US hubs
- ✓Monitoring, hardware replacement and named support
Not how we price
- ✕Per token or per call
- ✕Spot, auction or preemptible rates
- ✕Quota tiers or waitlists
- ✕Best effort marketplace hosts