Skip to content

H200 available now · enterprise early access

The platform under the GPU

You rent the GPU.We run the rest.

You bring the model and the data. We run the network, storage, power, security and billing under the card, all in Kathmandu.

Scroll to dive

What you see

The GPU.

An NVIDIA H200 with 141 GB of memory, available now and billed by the second in rupees. It's the part everyone compares, and the smallest part of the job.

h200 · available now — rtx pro 6000 blackwell · coming soon

Below the waterline

Nine layers you never operate.

Everything that turns a card into a service you can count on. Keep scrolling, one layer at a time.

Layer 01 / 09

Networking.

Every customer gets a private network, and nothing reaches your work unless you allow it. Latency inside Kathmandu stays under 5 ms.

vpc · cilium · egress policy · <5 ms

Layer 02 / 09

Storage.

Your files outlive the job on replicated volumes, with fast local disks for scratch. Upload a dataset once, not every run.

block · object · nvme scratch

Layer 03 / 09

Power & cooling.

Our Kathmandu datacenter runs on Nepal's hydropower. Racks stay cool, so a card holds full speed for the whole job, not just the first ten minutes.

hydropower · thermal · redundancy

Layer 04 / 09

Scheduling.

Launch a pod, a notebook or an endpoint, and we put it on the right card. You get your own private Kubernetes cluster without having to run it.

kubernetes · vcluster · orchestration

Layer 05 / 09

Monitoring.

Watch GPU use, job health and spend as they happen. You see the cost before the invoice does.

gpu metrics · traces · alerts

Layer 06 / 09

Security & isolation.

On whole cards and hardware slices (MIG), a neighbour's crash can't reach your work. Every action lands in an audit log you can export.

mig · audit log · least privilege

Layer 07 / 09

Data residency.

Your data stays in Kathmandu, and nothing is copied abroad. The jurisdiction is Nepal.

np-ktm-1 · in-country

Layer 08 / 09

NPR billing.

Pay by the second, in rupees, by eSewa, Khalti, bank transfer or invoice. No dollar card, no exchange-rate surprises.

per second · esewa · khalti · invoice

Layer 09 / 09

Local support.

Engineers in Kathmandu, working Nepal time. You talk to the people who run the racks.

kathmandu · nepal time

The whole iceberg

All of it, in Nepal.

We run all nine layers in Kathmandu, so you only think about your model. H200s are available now through enterprise early access.

Tell us what you want to run. We reply within one working day with a capacity plan and a firm rupee quote.