Infrastructure
AI infrastructure in Nepal, built on NVIDIA HGX B300
A planned enterprise AI cloud stack: NVIDIA HGX B300 on Supermicro hardware, for deployment in a third-party data centre in Nepal, under NDA. Nothing is in service yet.
Build status
Seven layers, silicon to software, each at today's honest phase. Nothing here is operational; the platform itself is on a waitlist (see below).
| Power | Planned Nepal’s national grid (hydropower). No power contract is in place yet. |
|---|---|
| Facility | Selected A third-party data centre in Nepal, planned, under NDA. Facility details are not published. |
| Network | Planned Tier-1 fiber transit via Indian telecom gateways. Bandwidth is at quotation stage; no circuit is contracted. |
| Compute | Selected NVIDIA HGX B300 on Supermicro: the chosen hardware target. Not yet installed or operational. |
| Orchestration | Target architecture Kubernetes + Slurm, NVIDIA AI Enterprise, Run:ai. The intended stack; nothing is installed yet. |
| Models | Target architecture The intended served-model lineup (brief D2). None is served today; the platform is on a waitlist. |
| Applications | In deployment The gateway, API, dashboard and billing layers exist and are under active development in this repository, ahead of the capacity they will serve. |
The technology stack
The planned stack from silicon to scheduler, with the status of each layer today.
NVIDIA HGX B300
Status: In progressBlackwell-architecture GPU compute platform with high-speed NVLink interconnect and HBM3e memory, designed for multi-billion parameter training and inference.
Supermicro
Status: In progressServer hardware and chassis platform. SuperCloud Composer for infrastructure management and Server Manager for cluster operations are planned.
Third-party data centre
Status: In progressPhase 1 is planned for a third-party data centre in Nepal, under NDA. Facility details are not published.
Engineering
Status: PlannedDeployment is planned with a project-based specialist engineering team covering infrastructure operations, model fine-tuning and deployment.
NVIDIA AI Enterprise
Status: PlannedEnterprise AI software stack with NVIDIA Run:ai GPU orchestration.
Kubernetes + Slurm
Status: PlannedContainer platform and HPC scheduling with PyTorch, TensorFlow and JAX frameworks, monitored via NVIDIA DCGM, Prometheus and Grafana.
The facility
CloudFRM.ai's Phase 1 capacity is planned for a third-party data centre in Nepal, under NDA. Facility details are not published.
| Facility | A third-party data centre in Nepal (planned, under NDA) Status: In progress |
|---|---|
| Power source | Nepal's national grid, which is almost entirely hydropower (IEA). No power contract is in place yet. Status: Planned |
| Grid resilience | Benefits from national transmission investment, including the U.S. Millennium Challenge Corporation (MCC) Nepal Compact's 315 km of 400kV lines. CloudFRM.ai is not a party to that programme. |
| Connectivity | Tier-1 fiber transit via Indian telecom gateways to global networks — bandwidth is at quotation stage; no circuit is contracted yet Status: In progress |
Security & compliance
Physical and logical controls are being designed to meet U.S. export regulations and international data protection standards.
Physical security
Status: PlannedPhysical security is planned to be provided by the third-party data centre operator. Facility details are under NDA and not published.
KYC screening
Status: PlannedTenants will be screened against the U.S. Consolidated Screening List before provisioning.
Network isolation
Status: PlannedStrict VPC isolation and firewall rules are planned, with continuous monitoring and full audit trails.
Have infrastructure questions?
Our engineering team is happy to walk through architecture, compliance, and connectivity in detail.