Nepal's data center bet
Nepal wants to become a regional data center hub, powered by hydroelectricity. The strategic logic is sound. The dry-season power shortage and Kathmandu's water crisis are also real, and neither has a quick fix.
The government's bet
Nepal has more than ten data centers today, most below 5MW, built to serve banks, hospitals, and government agencies. That's the baseline. The government wants to change it.
Finance Minister Dr. Swarnim Wagle's FY 2083/84 budget commits Nepal to building its first "Sovereign AI Compute Center" at Sucharter in Kathmandu. The plan: procure thousands of AI processing units, offer subsidized compute to domestic startups and researchers, and back it all with clean hydroelectricity. Fifteen Nepali AI researchers working abroad will be brought back on government fellowships. The pitch is simple — convert hydroelectric surplus into high-value digital exports instead of selling raw power to India at wholesale rates.
The private sector isn't waiting. Bichuten Data World, backed by NPR 5 billion ($37M) in domestic equity and a Laxmi Sunrise Bank credit line, is building two facilities at Chobhar (Kathmandu) and Birgunj, using Google's platform and India's VVDN Technologies for liquid cooling. Managing Director Naveen Agrawal calls it an "AI-driven facility," not just storage. Chobhar is in final construction, targeting August 2026.
Communications Minister Dr. Vikram Timsina is also studying whether multinationals should be allowed to invest in large-scale facilities. Industry analyst Rohit Chandra Shah of Huawei Nepal — the country's first Accredited Tier Specialist — puts it plainly: what once served only telecoms now serves every sector. Nepal isn't ahead of this trend. It's catching up.
Where the argument works
There are real reasons Nepal makes sense as a data center destination.
When the power isn't there
"Nepal has abundant clean power" is true for about five months. The other seven tell a different story.
Almost all of Nepal's hydropower is run-of-river — output tracks river flow directly. Rivers run high June through October, low November through May. Major stations show it: Chameliya fell from 30MW to 17MW during the 2019 dry season, a 43% drop (Kathmandu Post, Nov 2019). Other stations show similar patterns. The only facilities with real baseload buffer are the Kulekhani complex — Kulekhani I (60MW), II (32MW), and III (14MW, commissioned 2019) — 106MW total.
Nepal still imports power from India during dry months. In FY 2024–25, it imported NPR 12.92 billion worth while exporting NPR 17.46 billion, becoming a net electricity exporter annually for the first time (NEA, August 2025). That's real progress. But the seasonal mismatch hasn't gone away.
Data centers can't turn off in winter. Tier III facilities need 99.9% uptime — under 9 hours of downtime per year. Stacking hyperscale compute load on a grid that still relies on Indian imports every dry season makes that remaining deficit considerably harder to manage.
Climate change makes this worse. The Nepal Economic Forum projects droughts could cut river flows 38 to 57 percent over the long term. In June 2023, unseasonal floods damaged 30 hydropower projects totaling 463MW. Hydropower is clean. It isn't guaranteed.
The government's answer is 100MW of battery storage in Wagle's budget. Right direction, wrong scale. A hundred megawatts of storage covers minutes of demand, not months. Fixing the dry-season gap ultimately requires either reservoir-scale hydropower — a decades-long project — or accepting that dry-season data centers will run on imported power.
What data centers actually drink
Electricity gets the headlines. Water is the quieter problem, and for Kathmandu it may be more urgent.
Servers generate heat. Cooling that heat takes water. A typical enterprise data center uses 1.1 to 1.9 million liters per day. A large hyperscale facility — the kind that serves global cloud clients — can use up to 19 million liters. Google reported consuming approximately 6 billion gallons across all its data centers and offices in 2024, an 8% increase over 2023 and roughly three times its 2016 level, driven largely by AI (Google 2024 Environmental Report).
Kathmandu Valley's total daily water demand is about 280 million liters — and the system is already short. Groundwater extraction exceeds natural recharge. The Bagmati and Bishnumati rivers have shrunk dramatically over three decades of urbanization. Water demand grows about 10% per year. Melamchi added supply. Demand outpaced it anyway.
A single hyperscale data center at Chobhar or Sucharter would consume 7% of the entire Valley's daily water supply, drawn from an aquifer already in deficit. This isn't hypothetical. Northern Virginia's data center cluster grew to consume close to 2 billion gallons of water in 2023 — about 3% of the Potomac River Basin's total. Kathmandu has far less slack.
Bichuten's choice of VVDN liquid cooling helps. Liquid-cooled facilities can hit a Water Usage Effectiveness (WUE) of 0.5 liters per kilowatt-hour or less, versus 1.8 liters or more for conventional air cooling. That's a big difference at scale. It doesn't make the problem disappear. It shrinks it.
There's also an upstream question rarely raised in Nepal. Reservoir-based hydropower evaporates water — published estimates suggest up to 18.27 gallons per kilowatt-hour, far more than thermal plants. Nepal's grid is mostly run-of-river, so direct evaporation is minimal. But the water flowing past turbines is water diverted from the same river systems that feed Kathmandu's drinking water intakes downstream.
What doesn't fit the "clean" narrative
Water and electricity aren't the only issues. The environmental ledger has other entries.
The international comparisons
| Country | What happened | Outcome | Lesson for Nepal |
|---|---|---|---|
| Ireland | Allowed unrestricted data center growth from mid-2010s, primarily near Dublin | Data centers now consume 22% of national electricity, forcing EirGrid to warn of grid instability. Planning permissions effectively frozen in Dublin. | Nepal's grid is far more fragile than Ireland's was. There is no equivalent safety margin here. |
| Singapore | Was Asia's top data center destination. Imposed a moratorium in 2019 when data center growth threatened national energy sustainability and land constraints | Moratorium lifted in phases from 2022 with strict sustainability standards: PUE below 1.3, WUE of 2.0 m³/MWh or lower (per Singapore's Green Data Centre Roadmap, May 2024), 100% renewable energy pathway required | Nepal should set the standards now, before capacity is installed and locked in. Retrofitting is far more expensive. |
| Iceland | Positioned as renewable-energy data center hub using geothermal plus hydro | Sustained success because of truly baseload renewable power (geothermal), cold climate eliminating cooling load, and strict water accounting | The model Nepal should study. The difference: geothermal is baseload. Nepal's hydro is not. Nepal's "Iceland thesis" only holds in monsoon season. |
| Netherlands | Amsterdam became Europe's largest data center hub by the late 2010s | Provincial ban on new data centers in North Holland (2019-2020) due to water stress, land scarcity, and electricity grid strain. Partial reversal with strict criteria. | Amsterdam and Kathmandu share the same structural problem: water-stressed valley locations. The Netherlands resolved it by moving facilities outside the capital region. |
| India | Aggressive push for data center investment from 2021, targeting Mumbai, Chennai, Hyderabad, Pune | Significant investment attracted, roughly $4B committed 2021-2024. But most located near coastal grids, not landlocked interior. | Nepal's Birgunj (Terai, near Indian grid) is better positioned for private investment than Kathmandu. Coastal access to undersea cables matters for hyperscale. |
This chart raises a question that almost nobody in the Nepali data center debate asks: who is this infrastructure actually for?
Nepal's citizens use less electricity per person than every South Asian country except Afghanistan. Rural areas still face dry-season load shedding. Hospitals run on diesel backup. Farms in grid-unreliable areas depend on diesel irrigation pumps.
Data centers selling compute to Indian, European, or American clients will divert grid capacity from domestic users at the margin. The government's answer — monsoon surplus would otherwise be wasted or sold cheap to India — is fair for the wet months. It doesn't hold in the dry season, when the grid is already short.
A framework Nepal doesn't have yet
Singapore only lifted its data center moratorium when operators agreed to hard numerical targets. Nepal is doing the opposite: committing public money and policy support before setting any sustainability standards. Rules need to come before the concrete, not after.
| Policy area | What it should require | Nepal's current status |
|---|---|---|
| Power Usage Effectiveness (PUE) | Mandatory PUE below 1.4 for any DC above 2MW. Hyperscale must achieve below 1.3. | No standard exists. Existing facilities in Kathmandu are mostly air-cooled, estimated PUE 1.8-2.0. |
| Water Usage Effectiveness (WUE) | Maximum WUE 0.5 L/kWh for new builds in the Kathmandu Valley. Liquid or immersion cooling mandatory above 5MW. | No standard exists. Bichuten's VVDN liquid cooling is a voluntary market choice, not a legal requirement. For reference, Singapore's Green Data Centre Roadmap (2024) targets 2.0 m³/MWh; Nepal should aim to set a more stringent standard from the outset. |
| Dry-season energy sourcing | Data centers above 10MW must show a renewable power purchase agreement that covers dry-season demand, or use licensed battery storage. | No requirement. The government's 100MW battery program is too small and too slow for commercial compliance use. |
| Location policy | Hyperscale facilities should be steered to Terai locations (Birgunj, Bhairahawa, Parsa) with better grid interconnects, lower water stress, and industrial land availability. | No explicit policy. Both public (Sucharter) and private (Chobhar) investments are going into the already water-stressed Kathmandu Valley. |
| E-waste | Mandatory end-of-life hardware take-back, licensed processing, or export to certified facilities required before import clearance for server hardware. | No framework. Nepal's e-waste rules cover consumer electronics but not data center hardware at scale. |
| Domestic value capture | Data centers selling to foreign clients above a size threshold should hire locally above a minimum ratio and provide a share of compute capacity at subsidized rates to domestic institutions. | Bichuten has stated domestic intent voluntarily. No legal obligation for future investors. |
What the numbers say
SagarmathaIQ Assessment
The strategic case is real. Nepal has genuine advantages: monsoon surplus, altitude cooling, a domestic digital economy that needs sovereign infrastructure, and a diaspora with capital and skills. Converting hydropower into compute exports instead of selling raw power to India at wholesale rates is economically sound.
The execution has structural problems. Piling new data centers into the Kathmandu Valley without water efficiency standards, dry-season power requirements, or a location policy steering hyperscale to the Terai is the fastest route to an Ireland-style crisis within five years.
The environmental case depends entirely on scale and location. A few well-designed, liquid-cooled facilities below 20MW, running primarily on monsoon surplus near Birgunj or Bhairahawa, are defensible. A Kathmandu Valley hyperscale build-out without regulation is not — not for a city already running a water deficit.
The question nobody is asking: Before Nepal pitches itself to foreign hyperscale clients, it should explain why its own citizens use 479 kWh per person per year — roughly one-thirtieth of Bhutan's figure. Infrastructure that serves the domestic economy and uses clean seasonal surplus wisely is a different proposition from one that primarily serves foreign cloud clients while drawing from Kathmandu's groundwater.