Data center development · Bhutan
Standardize the block.
Pre-sell the capacity.
Then build the next one.
bitit develops hydro-powered compute capacity in Bhutan. We take a site from power and permits through to an operating 100 MW block — then build the same block again, funding each phase from the one before it.
How the company works
A developer, not a landlord.
Most compute projects raise money, build a shell, and hope tenants arrive. We run the sequence in the other order — capacity is committed before the block is fully built, and each completed block pays for the ground under the next one.
-
01
Secure the ground
Land, grid interconnection, water, road access and the permits that let a block exist. This is the slow part, and it is the part that is worth owning.
-
02
Standardize the block
One design, 100 MW, assembled from a repeated 1 MW pod. Standardizing the pod is what makes the second block cheaper than the first, and the sixth cheaper than the second.
-
03
Pre-sell before steel
We target half a block committed before it goes to full build. A signed buyer is the cheapest financing available, and it is also the honest test of whether the capacity is wanted.
-
04
Build the template, then copy it
The first block absorbs the learning — the drawings, the vendors, the commissioning sequence. Everything after it is a copy, built by people who have built it before.
-
05
Operate and collect
Hosting fees on delivered capacity, billed against contracted load. Buyers own their machines; we own the building, the power and the uptime.
-
06
Recycle the capital
Cash from an operating block funds the next one rather than a dividend. The end state is an income-producing asset sold whole, not a company run forever.
Step 06 funds step 01 of the next block
The unit we sell
One hundred megawatts, sold whole.
A block is the smallest thing that makes sense to build twice. It is sized so a single buyer can take all of it, and structured so several buyers can share one without arguing about the walls.
| Block capacity | 100 MW IT load |
|---|---|
| Smallest build increment | 1 MW pod |
| Pods per block | 100 |
| Pilot ahead of the first full block | 20 MW |
| Commitment targeted before full build | 50 % |
| Power source | Hydroelectric grid supply |
| Sold as | A whole block, or capacity within one |
| Buyer owns | Machines |
| bitit owns | Building, power, uptime |
Figures describe the development standard. Commercial terms, tariff and delivery schedule are set per block and per buyer.
The site
Built where the power already is.
Bhutan generates its electricity almost entirely from hydropower. Siting compute next to that generation is the whole argument — it shortens the distance between the turbine and the rack, and it puts the load on a grid that was built for industry.
Predominantly hydroelectric, from run-of-river and storage schemes on Himalayan catchments.
Large, steady industrial loads are the kind of customer a hydro grid is easiest to plan around — and the kind that can be curtailed and restarted without breaking anything.
Interconnection, tariff and the licensing route are confirmed with the relevant authorities before a block is committed, not assumed from a brochure.
Development plan
Phase one is one block.
We would rather commission 20 MW that works than announce 600 MW that does not. The pilot exists to prove the pod design and the commissioning sequence at a size where mistakes are affordable.
Proves the 1 MW pod, the power path and the commissioning sequence. Small enough to correct, large enough to be a real reference.
The first complete block, taken to full build once commitment reaches the target. Everything learned here becomes the drawing set.
Copies of the template, each sold whole, each funded substantially by the block before it.
Who we build for
Two kinds of buyer.
Operators who want their own capacity
Buyers who have outgrown rented compute and would rather hold the asset. They take a block, or a defined share of one, and keep their own machines inside it.
- Takes a whole block or a fixed share
- Owns the hardware, not the building
- Contracted load, hosting fee against it
Latency-tolerant AI workloads
Work that cares about cost per delivered token far more than about the round trip to the user — batch inference, fine-tuning, rendering, anything queued rather than typed at.
- Batch inference and offline scoring
- Training and fine-tuning runs
- Render and simulation queues
Contact
Tell us what you need powered.
Capacity enquiries, co-development, and everything else reach the same inbox. Say how much load and when, and you will get a straight answer about whether we can meet it.