01AI infrastructure in Nepal
Powering Intelligence.
CloudFRM.ai is building AI infrastructure in Nepal, from power to compute to the models and applications that run on it. Phase 1 hardware is on order, and the platform is on a waitlist until capacity is in service.
02From power to intelligence
It starts as snow on the Himalayas.
How the pieces fit, from the river to what people make. Each step is a chapter further down.
03Why Nepal
Rivers that run the grid.
From the Himalayas, to energy, to compute, to the world.
- HimalayasSnow and glaciers feed the rivers that run south through Nepal.
- EnergyThose rivers generate almost all of Nepal’s electricity: the IEA records the grid as almost entirely hydropower.
- ComputeCloudFRM.ai plans to put that energy to work as AI compute in Nepal, drawing on the national grid.
- GlobalCompute in Nepal, for developers and organisations in Nepal and around the world.
Energy use will be published once hardware is in service and metered.
04Infrastructure
Where the build stands.
Layer by layer, and where each one is today.
- PowerStatus: PlannedPlanned to draw on Nepal’s national grid. No power contract is in place yet.
- FacilityStatus: In progressA third-party data centre in Nepal, under NDA. Facility details are not published.
- NetworkStatus: In progressBandwidth is at quotation; no circuit is contracted yet.
- ComputeStatus: In progressPhase 1 NVIDIA HGX B300 hardware is on order; nothing is racked.
- OrchestrationStatus: PlannedCluster scheduling and management software, planned.
- ModelsStatus: WaitlistServed through the platform once capacity is connected.
- ApplicationsStatus: In progressThe Bhasa voice assistant is in speech-to-text testing.
The film is illustrative footage, not a CloudFRM.ai site.
05Enterprise
Private and sovereign AI.
For organisations that need their data, their models and the compute that runs them kept under their control, in Nepal, on dedicated infrastructure.
06Compute
Build on CloudFRM.ai compute.
GPU capacity for training, fine-tuning and inference, planned on NVIDIA HGX B300 for Phase 1. It is on a waitlist: nothing is in service yet.
07Models
Open models, served here.
A catalogue of open models, each shown as usable only once it is served. None is served yet.
08AI platform
One platform for what you build.
What developers and organisations will build with, on one account and the same compute.
- InferenceStatus: WaitlistAn API for running models, being built, with usage metered per request.
- AI AgentsStatus: PlannedAgent Studio: a planned place to build agents that use models and tools.
- RAGStatus: PlannedAnswers drawn from your own documents, not only from what a model was trained on.
- Private AIStatus: PlannedThe platform on infrastructure dedicated to one organisation.
- DataStatus: PlannedA place on the platform to keep and prepare the data your models use.
09Products
What we are making.
None is linked from this site yet: a product is linked here only once there is something to use and its own site has been reviewed.
- BhasaStatus: In progressNepali-language AI. Its first application, a voice assistant that answers banking questions in Nepali, is in speech-to-text testing.
- VoiceStatus: In progressVoices for AI, starting with a studio for recording speech data.
- FableStatus: PlannedAn original anime universe. It stays dark until its first chapter is ready.
10Real creation
AI that speaks Nepali.
The first application of this work is a voice assistant that answers banking questions in Nepali. It is in speech-to-text testing: we are measuring how well it understands spoken Nepali before anyone relies on it.
Other work appears here only once there is something real to show.
11Developers
Start building.
Developer accounts open from the waitlist once capacity is connected. Until then, see how the platform will work and which models are planned.
12Company
Built in Nepal.
Cloud Future Automation Research Matrix Pvt Ltd (CloudFRM.ai) is a company in Kathmandu, Nepal, building AI infrastructure from the power up. Every capability on this site is labelled with where it stands today.
13Join the waitlist
Build with CloudFRM.ai.
The platform is on a waitlist while capacity is brought into service. Start building when accounts open, or tell us what you need now.
GPUs
A whole GPU, or just the slice you need.
Use a whole GPU, an isolated slice of one, or a share of one. Nothing is in service yet.
Full GPU
A whole NVIDIA B300, yours alone.
MIG
One GPU split into up to seven isolated slices.
vGPU
A share of a GPU, for lighter work. Coming later.
- Small shareStatus: PlannedFor development and testing.
- Large shareStatus: PlannedFor steady, lighter inference.
An example line-up: it shows the kinds of GPU on offer, not how many there will be.


