Our cloud.Your models.
Dedicated GPU clusters and cloud for AI, enterprise, and government.
One provider for AI, enterprise, and government workloads.
GPU cloud
Dedicated NVIDIA HGX B200 and B300 clusters, single-tenant and tested before handover.
Inference and model hosting
Serve open models or your own. Bring the weights, or bring the data and we fine-tune.
Enterprise cloud
Virtual machines, storage, and private networking on the same network as the GPUs.
Government cloud
Infrastructure designed around federal requirements, with a published compliance status.
Hardware that is yours alone.
Dedicated systems and a dedicated fabric partition. No GPUs shared between customers.
Tested before handover.
Every system runs burn-in, fabric, and storage acceptance tests before you get access. You receive the full results.
Engineers you can reach by name.
Each customer has an engineering contact who knows how their cluster is built and has the access to change it.
Contracts procurement can approve.
Reserved terms, written service levels, and pricing you can forecast.
1 of 4
NVIDIA Blackwell, on two platforms.
Eight GPUs per system. Choose HGX B200 for most training and inference, or HGX B300 for long-context and memory-bound work.
| HGX B300 | HGX B200 | |
|---|---|---|
| Architecture | NVIDIA Blackwell Ultra | NVIDIA Blackwell |
| GPUs per system | 8 | 8 |
| GPU memory per system | 2,304 GB HBM3e | 1,440 GB HBM3e |
| GPU interconnect | NVLink, 1.8 TB/s per GPU | NVLink, 1.8 TB/s per GPU |
| Scale-out network | 800 Gb/s per GPU | 400 Gb/s per GPU |
| Suited to | Long-context and reasoning models; Memory-bound inference; Large mixture-of-experts training | Pre-training and continued training; Fine-tuning at scale; Batch and online inference |
Figures are NVIDIA's published HGX platform specifications. CPU, system memory, local storage, and NIC models are confirmed per deployment.
Size a cluster.
Choose a platform and a system count to see what you would be reserving. Your configuration goes with your request.
8 HGX B300 systemsDrag across the racks or use the slider.
64 GPUs
- HBM3e memory
- 18.4 TB
- Scale-out bandwidth
- 51 Tb/s
- NVLink domains
- 8
- Racks
- 2
Figures use NVIDIA's published per-GPU specifications and our design target of 4 systems per rack. Final configuration is confirmed per agreement.
Where we stand today.
We are building our first deployment. Each registration and audit below is marked complete only when it is issued or final.
| Framework | Relevant to | Status |
|---|---|---|
| SAM.gov registrationFederal | Federal | Planned |
| SOC 2 Type IICommercial and public sector | Commercial and public sector | Planned |
| NIST SP 800-171Defense industrial base, federal | Defense industrial base, federal | Planned |
| CMMC Level 2Department of Defense | Department of Defense | Planned |
| FedRAMP ModerateFederal civilian and defense | Federal civilian and defense | Planned |
| Small business certificationFederal, state | Federal, state | Planned |
Questions
When will capacity be available?
We are in the first phase of the build: facility, power, and design partners. We share deployment timelines with prospective partners under NDA, because they depend on agreements we will not announce until they are signed.
Which GPUs do you offer?
NVIDIA HGX B200 and HGX B300 systems, eight GPUs per system, connected by NVLink inside each system and a non-blocking network between systems.
Is there a minimum commitment?
Reserved clusters start at a single eight-GPU system. Term length and size are set per agreement.
Can you support government and regulated workloads?
We are designing our facilities and operations around NIST SP 800-171, with SOC 2, CMMC Level 2, and FedRAMP Moderate on our roadmap. None of these are complete today. The trust center lists the current status of each.
Do you offer anything besides GPUs?
Yes. Enterprise cloud services run in the same facility: virtual machines, block and object storage, private networking, and backup. We also plan managed inference, model hosting, and fine-tuning on the same GPUs.