Infrastructure
The place.
A datacentre in the Kathmandu Valley, running on Nepal's hydropower. Where your data is stored, processed and kept.
Being commissionedKathmandu Valley
The five-layer cake
Every layer, named. From energy to your app.
NVIDIA's Jensen Huang describes AI as a five-layer cake: energy, chips, infrastructure, models and applications. Here is each layer at CoreValley.
- 01ApplicationsFraud scoring · voice in Nepali · assistants · forecastingWhat you build: the products and services your customers use.
- 02ModelsOpen models · your fine-tunes · your ownTrain, fine-tune and serve models with the tools you already know: JupyterLab, Kubeflow, Ray, MLflow.
- 03InfrastructureDatacentre · cooling · networking · storage · schedulingWhat turns chips into one machine, run by our engineers in Kathmandu.
- 04ChipsNVIDIA H200 NVL · 141 GB HBM3e per GPUWhole GPUs, or up to seven isolated slices of one with NVIDIA MIG.
- 05EnergyNepal's hydropower · 99.99 % facility uptimeWhere every AI stack starts. Ours runs on Nepal's hydro-dominated grid.
The five layers are NVIDIA's framing1.
Capacity
Starting with H200 NVL. Growing from there.
Our first NVIDIA H200 NVL GPUs are being commissioned now. We'll scale to sixteen GPUs, then add other NVIDIA GPUs as demand grows.
- NowNVIDIA H200 NVLBeing commissioned; enterprise early access is open.
- NextSixteen GPUsMore H200 NVL as early-access customers come on board.
- LaterMore GPUsOther NVIDIA GPUs and new generations, as demand grows.
Power
Runs on hydropower. Nepal's own.
Our datacentre runs on Nepal's grid, which is built on hydropower. We say it plainly and footnote it, rather than dressing it up as a certificate we don't hold.
Network
Under 5 ms inside the valley. At least twenty round trips in a quick blink.
From inside the Kathmandu Valley, a round trip to our datacentre typically takes under five milliseconds.
Typical figure; yours depends on your network and location5
The facility
What we can say about the building.
Enough to judge it, nothing that helps anyone break in.
In the Kathmandu Valley
In Nepal, close to the teams who use it.
Being commissioned
Our first GPUs are coming soon; enterprise early access is open.
99.99 % facility uptime
The facility's commitment for power, cooling and the building.
Details on request
We don't publish the address or security arrangements. Regulated customers get a written description of their data flows during onboarding.
Residency
Your data is safe with us. Here is exactly what that means.
Stays in Nepal
- Datasets and files you upload
- Model weights, checkpoints and adapters
- Prompts and answers of your endpoints
- Notebooks and their outputs
- Logs of your jobs
Nothing you upload, train or generate is copied outside Nepal. Regulated customers get a written description of their data flows during onboarding.
Questions
About the place.
Where exactly is the data?
In our datacentre in the Kathmandu Valley. We don't publish the address; regulated customers get a written description of their data flows during onboarding.
What runs there?
NVIDIA H200 NVL GPUs. We don't publish the detailed configuration; customers get what they need during onboarding.
How fast is the network?
From inside the Kathmandu Valley a round trip typically takes under 5 ms; yours depends on your network and location.
What powers the datacentre?
Nepal's grid, which is built on hydropower.
What does 99.99 % cover?
It is the datacentre facility's availability commitment for power, cooling and building infrastructure. Service-level terms for CoreValley services are set out in your agreement.