Company
Nepal's AI infrastructure, built for Nepal.
We exist so that researchers, startups and enterprises can build and run AI without foreign clouds, foreign currency or foreign support hours.
H200 available now · enterprise early access
Our mission
CoreValley is building GPU infrastructure in Nepal, so Nepali teams can train models, fine-tune LLMs, run inference and ship AI products without sending data abroad or paying in dollars.
Until now, serious AI work in Nepal has meant three compromises: data shipped overseas, invoices in USD, and support that answers while you sleep. That does not work for universities, for regulated industries, or for the next generation of Nepali AI companies.
So we built the alternative: the NVIDIA H200, the same class of GPU the world’s AI labs train on, in a hydro-powered datacenter in Kathmandu. Billed in rupees, and run by engineers who understand both the technology and the local context.
What we run
Four products, one platform.
Notebooks for teams, GPU environments for training, endpoints for serving and dedicated servers for steady load. All on NVIDIA H200s in Kathmandu today, with Blackwell coming soon — one account, one rupee invoice.
GPU pods
formerly GPU Workspace
Rent a whole H200 or a slice of one, with root access and your own images. Pay by the second.
See GPU podsJupyterHub
formerly AI Lab
Managed GPU notebooks for teams, labs and courses. Nobody has to run a shared server.
See JupyterHubModel endpoints
formerly Inference & Endpoints
Open-weight models behind an OpenAI-compatible API, served from inside Nepal and billed per token.
See model endpointsDedicated & bare metal
formerly part of GPU Workspace
Whole H200 servers that only you use, reserved by the month, in your own private cluster.
See dedicated servers- Hardware
- NVIDIA H200, available now. RTX PRO 6000 Blackwell coming soon; L40S after it.
- Software
- Ready from first boot: CUDA, PyTorch, TensorFlow, Jupyter, vLLM and DeepSpeed.
- Billing
- In rupees, metered per second. eSewa, Khalti, bank transfer or corporate invoice.
- Power and latency
- A hydro-powered datacenter in Kathmandu, with under 5 ms latency inside the city.
Principles
What guides us.
Sovereign by default
Your infrastructure, your data and your support stay in Nepal. Local compliance is a design requirement, not an add-on for enterprise deals.
Practical over theoretical
We optimise for real work: fast setup, firm quotes in rupees, and engineers in Kathmandu who answer in Nepal time.
Accessible to researchers
University labs and student projects should reach serious GPUs without an enterprise procurement cycle. Enterprise early access comes first; research access is next.
Honest about constraints
We publish what we actually run. Today that is the NVIDIA H200. Where a GPU has not landed or a control is still in progress, we say so.
Who we serve
Who we build for.
Early access is open to enterprises on NVIDIA H200. More GPUs, and access for more teams, are coming soon.
Enterprises and public sector
Open now · H200Isolated environments, data that stays in the country and engineers in Kathmandu — for banks, hospitals, government and any regulated workload that cannot cross a border.
Universities and researchers
Coming soonManaged notebooks so students and faculty can work with modern models without running a cluster or opening a foreign cloud account.
Startups and ML teams
Coming soonGPU pods with root access, the freedom to install what you need, and a path to endpoints when you ship.
Questions
Frequently asked.
Where is CoreValley hosted?
In a hydro-powered datacenter in the Kathmandu Valley, Nepal — the region we call np-ktm-1. Compute and storage stay in the country. Nothing is replicated abroad.
Which GPUs do you run?
The NVIDIA H200, with 141 GB of memory, is available now through enterprise early access — as a whole card or as a slice. The RTX PRO 6000 Blackwell is coming soon, with the L40S after it.
How do you bill?
In Nepali rupees. Usage is metered per second and invoiced monthly, and dedicated servers can be reserved by the month. Pay by eSewa, Khalti, bank transfer or corporate invoice on net terms. Rates are not published yet, so every request gets a firm quote within one working day.
Are you open to new customers?
Yes. Early access is open to enterprises on NVIDIA H200 now. Tell us the model, the dataset size and the GPU hours you expect, and we size capacity with you before you commit. Universities, startups and other teams can join the list today — access for more teams is coming soon.
What does early access mean?
The H200 platform is live, and we onboard enterprise teams one at a time. You tell us the workload, we send a capacity plan and a firm rupee quote within one working day, and our team sets up your project with you. More GPUs, and access for more teams, are coming soon.
Why does it matter that the GPUs are in Kathmandu?
Two reasons. Your data never has to leave the country, and latency inside Kathmandu is under 5 ms — so an app serving users in Nepal does not wait on a round trip overseas.
What powers the datacenter?
Hydropower. Our Kathmandu datacenter runs on Nepal's hydroelectricity, so your compute is low-carbon and its power price is not tied to a gas market on another continent.
How do I reach the team?
Email info@corevalley.ai or use the contact form. We reply within one working day, in Nepal time.
Contact
Where to find us.
- Location
- Kathmandu Valley, Nepal
- np-ktm-1 · hydro-powered
Tell us what you want to run. We reply within one working day with a capacity plan and a firm rupee quote.