Ethernet
B200 Ethernet
$6.75M
- 64 × NVIDIA B200, 8 servers
- High-speed Ethernet between servers
- Good fit for inference and fine-tuning
Boxes & pricing
Four boxes. 64 GPUs each.
Pick the GPU, B200 or B300, and the fabric, Ethernet or Cluster. Every box is a 20 ft ISO container with eight servers in two racks. List prices, USD, per box.
Cluster
B200 Cluster
$7.5M
- 64 × NVIDIA B200, 8 servers
- InfiniBand fabric and central NVMe storage
- Built for multi-node training. UPS options available
Ethernet
B300 Ethernet
$8.25M
- 64 × NVIDIA B300 (Blackwell Ultra), 8 servers
- High-speed Ethernet between servers
- More memory per GPU for larger models
Cluster
B300 Cluster
$9.0M
- 64 × NVIDIA B300 (Blackwell Ultra), 8 servers
- InfiniBand fabric and central NVMe storage
- The full training box. UPS options available
List prices per box. Final pricing, delivery timing, and terms are confirmed in a signed purchase agreement and depend on hardware availability.
Ethernet vs Cluster
Which fabric do you need?
If your jobs fit on one 8-GPU server at a time, or you mostly serve models, Ethernet is usually enough. If you train across many servers at once, Cluster’s InfiniBand fabric and shared storage keep the GPUs fed.
| Ethernet | Cluster | |
|---|---|---|
| GPUs | 64 × B200 or B300 | 64 × B200 or B300 |
| Servers | 8 (2 racks × 4) | 8 (2 racks × 4) |
| Server fabric | High-speed Ethernet | InfiniBand |
| Storage | No central NVMe | Central NVMe storage |
| Best for | Inference, fine-tuning | Multi-node training |
| UPS | Ask us | Options available |
| Cooling | In-frame dry cooler. Air-cooled servers by default; liquid options available | |
| B200 list | $6.75M | $7.5M |
| B300 list | $8.25M | $9.0M |
- B200NVIDIA Blackwell, at the lower list price.
- B300NVIDIA Blackwell Ultra, with more memory per GPU for larger models.
Larger builds
More than 64 GPUs.
Boxes can be paired on one site, and larger formats are available for bigger deployments.
Paired 20 ftTwo boxes side by side on one site. Available.
128+ GPUsMulti-box builds for larger training and inference. Available.
40 ft40 ft configurations. Available.
Need more than 64 GPUs?
Paired 20 ft boxes, 128-GPU and larger builds, and 40 ft configurations available.