Skip to main content

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.

03Why Nepal

Rivers that run the grid.

From the Himalayas, to energy, to compute, to the world.

  1. HimalayasSnow and glaciers feed the rivers that run south through Nepal.
  2. EnergyThose rivers generate almost all of Nepal’s electricity: the IEA records the grid as almost entirely hydropower.
  3. ComputeCloudFRM.ai plans to put that energy to work as AI compute in Nepal, drawing on the national grid.
  4. 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: Planned
    Planned to draw on Nepal’s national grid. No power contract is in place yet.
  • FacilityStatus: In progress
    A third-party data centre in Nepal, under NDA. Facility details are not published.
  • NetworkStatus: In progress
    Bandwidth is at quotation; no circuit is contracted yet.
  • ComputeStatus: In progress
    Phase 1 NVIDIA HGX B300 hardware is on order; nothing is racked.
  • OrchestrationStatus: Planned
    Cluster scheduling and management software, planned.
  • ModelsStatus: Waitlist
    Served through the platform once capacity is connected.
  • ApplicationsStatus: In progress
    The 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.

Status: Planned

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.

Status: Waitlist

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 progress
    Nepali-language AI. Its first application, a voice assistant that answers banking questions in Nepali, is in speech-to-text testing.
  • VoiceStatus: In progress
    Voices for AI, starting with a studio for recording speech data.
  • FableStatus: Planned
    An 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.

Status: In progress

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.

  • GPU 01Status: Waitlist
    NVIDIA B300 · 288 GB HBM3e
  • GPU 02Status: Waitlist
    NVIDIA B300 · 288 GB HBM3e
  • GPU 03Status: Waitlist
    NVIDIA B300 · 288 GB HBM3e

MIG

One GPU split into up to seven isolated slices.

  • 1/7 sliceStatus: Waitlist
    One seventh of a B300, walled off from the other slices.
  • 3/7 sliceStatus: Waitlist
    Three sevenths of a B300, for bigger models.

vGPU

A share of a GPU, for lighter work. Coming later.

  • Small shareStatus: Planned
    For development and testing.
  • Large shareStatus: Planned
    For steady, lighter inference.

An example line-up: it shows the kinds of GPU on offer, not how many there will be.