Nepal: The Next Frontier for AI Infrastructure?
Every data center on Earth runs on three inputs: power, cooling, and connectivity. The industry has spent two decades optimizing the third while taking the first two for granted. AI changed that equation overnight.
Training a single frontier AI model now consumes as much electricity as a small city. Rack densities have surged from 8–15 kW to 50–130 kW. Cooling bills are the fastest-growing line item on every operator’s P&L. And the grids feeding the world’s largest data center markets (Northern Virginia, Singapore, Johor) are buckling under the load, with some now threatening moratoriums on new builds.
The result is a trilemma that no established market has solved: operators need power that is simultaneously cheap, clean, and secure. Cheap power is usually dirty (Indonesia: 67% coal). Clean power is usually expensive (Singapore: $0.18/kWh). Secure jurisdictions often lack infrastructure. Pick two. The industry has accepted this as a binding constraint.
It doesn’t have to be. There is a country where industrial electricity costs $0.061 per kilowatt-hour, 57% below the world average. Where 95% of the grid runs on hydropower, producing near-zero carbon emissions without offsets or RECs. Where 6,000 Himalayan rivers provide unlimited cold water for liquid cooling. Where the government is building $900 million in new transmission infrastructure. And where the open border with the world’s fastest-growing data center market is 1,770 kilometers long.
That country is Nepal.
The Demand Explosion
A $151 Billion Market: Asia Is the Epicenter
The Asia-Pacific data center market was valued at roughly $78 billion in 2025 and is projected to reach $151 billion by 2030, growing at a 14% CAGR. The region is expected to add over 31 GW of installed power capacity in that timeframe alone. Hyperscale colocation in APAC is projected to nearly triple, reaching $50 billion by 2030.
India, specifically, is one of the fastest-growing markets on the planet. Its data center market stood at roughly $6–7 billion in 2025 and multiple analysts project it reaching $17+ billion by 2030, a CAGR of over 20%. Power demand from Indian data centers alone is forecast to grow nearly five-fold to 57 TWh by 2030. Over $32 billion in new data center investments have been announced in India in the last two years, from AWS, Microsoft, Google, and domestic players alike.
But India’s explosive growth comes with constraints: grid reliability challenges, continued coal dependence (renewables are targeted at only 32% of generation by 2030), water stress in key metros like Mumbai and Chennai, and rising land and power costs. Average PUE in Indian data centers remains 1.45–1.55, well above global best practice.
Directly north of India’s most power-hungry corridor lies a country with the opposite problem: too much clean electricity and not enough demand to absorb it.
The India Proximity Advantage
Next Door to the World’s Fastest-Growing Market
Nepal shares an open border and 1,770 km of contiguous frontier with India. Cross-border fiber optic connectivity is growing. The 400 kV Dhalkebar–Muzaffarpur transmission line already enables electricity trade between the two countries, and Nepal has been authorized to sell surplus power on the Indian Energy Exchange since 2021.
This proximity means Nepal can serve as an overflow infrastructure zone for India’s data center boom. Latency to northern Indian metros (Delhi NCR, Lucknow, Patna) can be kept within acceptable thresholds for most workloads, including inference. For training workloads and batch processing, latency is nearly irrelevant; what matters is power cost and carbon intensity. On both counts, Nepal dominates.
Consider the economics: a hyperscale facility consuming 50 MW at Nepal’s industrial rate of $0.061/kWh would spend roughly $26.7 million per year on electricity. The same facility in Singapore at $0.18/kWh would spend $78.8 million, nearly three times as much. Over a 15-year facility lifecycle, that delta exceeds $780 million in power costs alone.
How Nepal Stacks Up Against Competing Markets
The Full Picture: Strengths and Gaps Side by Side
Nepal dominates on power cost, carbon intensity, renewable share, natural cooling, and water; but trails badly on connectivity and regulatory maturity. This table makes the trade-offs explicit. The question for any operator is which variables matter most for their specific workload mix.
| Nepal | Malaysia | Indonesia | Vietnam | Singapore | |
|---|---|---|---|---|---|
| Industrial electricity | $0.061 | $0.070 | $0.090 | $0.080 | $0.180 |
| 50 MW facility annual cost | $26.7M | $30.7M | $39.4M | $35.0M | $78.8M |
| Grid carbon intensity | ~0.02 | ~0.54 | ~0.78 | ~0.45 | ~0.41 |
| Renewable share | 95% | ~8% | ~15% | ~27% | ~3% |
| Subsea cables | None | Yes (many) | Yes (many) | Yes (several) | Yes (hub) |
| Intl bandwidth | ~1.3 Tbps | >50 Tbps | >30 Tbps | >20 Tbps | >100 Tbps |
| Ambient cooling | Excellent | Poor | Poor | Poor | Poor |
| Water for cooling | Abundant | Adequate | Varies | Adequate | Scarce |
| Proximity to India | Border | Sea | Sea | Sea | Sea |
| DC regulatory framework | None | Developing | Developing | Developing | Mature |
Nepal’s Distributed Power Infrastructure
175 Plants, 3,422 MW, and Growing Fast
As of March 2025, the nation’s installed capacity stands at 3,422 MW, of which 95% comes from hydropower, with small contributions from solar (107 MW) and a legacy thermal plant (53 MW). There are 175 operational hydropower plants commissioned to date. These are spread across dozens of river basins and districts, giving the grid a distributed resilience that centralized systems lack. A landslide or climate event that knocks out one plant or basin does not cripple the system. As extreme weather events become more frequent globally, this kind of built-in geographic redundancy is a structural advantage that few competing markets can match.
But the pipeline is what changes the calculus entirely. An additional 260 hydropower projects are currently under construction, representing 10,696 MW of new capacity. The three largest (Arun-3 at 900 MW, Tila-1 at 440 MW, and Betan Karnali at 439 MW) are each individually larger than many countries’ entire renewable additions. Nepal’s economically feasible hydropower potential is estimated at 43,000 MW, with a theoretical ceiling of 83,000 MW. Less than 8% of the economic potential has been developed so far.
Nepal already generates more power than it consumes during the monsoon months and began exporting surplus to India in 2021. As the construction pipeline comes online, Nepal is projected to become a year-round net exporter. That surplus is cheap, stranded energy: exactly what data centers need.
Representative sample of major plants (51 operational, 7 under construction). Nepal has 175 operational plants totaling 3,422 MW; smaller and micro-hydro (<1 MW) plants are excluded. Locations are district-level coordinates. Border data from official Nepal government geoportal [26]. Power plant data from NEA and Wikipedia [2][4].
Grid Modernization: The Missing Backbone
$900M+ in Transmission Upgrades Underway
Generation capacity means nothing if the grid can’t deliver it. Nepal’s transmission network currently spans roughly 6,507 circuit-kilometers of high-voltage lines (66 kV through 400 kV), but has historically lagged behind generation investment. That is changing fast. Multiple major grid modernization programs are now simultaneously under construction, backed by the ADB, EIB, and the Government of Nepal.
A $582 million transmission project (designated a National Pride project) is building approximately 315 km of double-circuit 400 kV transmission lines connecting river basins to the national grid. Three new 400 kV gas-insulated substations are being constructed at Ratmate (Nuwakot), New Damauli (Tanahun), and New Butwal (Nawalparasi). The New Butwal substation will link directly to the planned 400 kV Butwal–Gorakhpur cross-border line to India.
Separately, the ADB approved a $311 million loan in late 2024 to finance 290 km of additional transmission lines: from Dailekh to Jumla, New Butwal to Lamahi, Nijgadh to Ramauli, and Teenpiple to Okharpauwa. This project also includes smart grid components: SCADA (Supervisory Control and Data Acquisition) network deployment, smart meter rollout expansion, automation of 39 existing substations, and establishment of a data recovery center.
On the cross-border front, Nepal and India signed a landmark agreement in October 2025 to construct two additional 400 kV transmission lines: Inaruwa–New Purnea and Lamki–Bareilly, with a target completion by 2030. The existing 400 kV Dhalkebar–Muzaffarpur line has been upgraded to 1,000 MW capacity. India has committed to importing 10,000 MW from Nepal over the next decade. NEA currently exports roughly 1,000 MW daily to India and Bangladesh.
Per capita electricity consumption has tripled, from 131 kWh in 2015/16 to over 400 kWh in 2023/24, reaching roughly 700 kWh by the end of 2026. Total national consumption hit 13.7 TWh in 2025, up from under 4 TWh a decade earlier. The grid is being built to handle not just today’s demand but the coming decade of industrial electrification, EV adoption, and (potentially) data center loads.
The Demand–Supply Mismatch, and the Opportunity
Seasonal Surplus Is a Data Center’s Best Friend
Nepal’s electricity system exhibits two structural patterns that are directly relevant to data center planning. First, a seasonal mismatch: the monsoon season (June–September) delivers high river flows that push hydropower generation well above domestic demand, creating massive surplus. The dry season (December–March) sees run-of-river output drop by more than 50%, requiring imports from India to cover the gap. In July 2025, peak system demand reached 2,901 MW; installed capacity was already above 3,400 MW and climbing.
Second, an intraday peak pattern: Nepal’s daily load curve spikes during 6:00–9:00 AM and 6:00–9:00 PM (morning cooking/commute and evening lighting/cooking). Off-peak hours (particularly overnight and midday) see dramatically lower demand. NEA has begun implementing Time-of-Use (ToU) tariffs to incentivize load-shifting, and off-peak industrial rates can be as low as $0.023/kWh at high voltage.
For data centers, this is a structural advantage. AI training workloads are flexible: they can be scheduled during off-peak hours and surplus-season months. A 50 MW data center operating predominantly during Nepal’s surplus window would absorb power that otherwise has no buyer. NEA’s Independent Power Producers (IPPs) reported that they would need to curtail electricity worth $192 million in FY 2024/25 during the wet season alone because they couldn’t find export markets fast enough. Data centers could absorb this stranded energy at deeply discounted rates.
The Water Cooling Advantage
6,000 Rivers Meet 3,000x Heat Transfer Efficiency
Water is a constraint most operators overlook. Conventional air-cooled facilities rely on evaporative cooling towers that consume enormous quantities of freshwater, a growing crisis in water-stressed regions like Northern Virginia, Phoenix, and Mumbai. Liquid cooling, now essential for AI-density racks above 30 kW, offers a different thermodynamic profile.
Water’s thermal conductivity is roughly 25 times that of air, and its specific heat capacity is about four times higher. Combined, liquid cooling systems can move heat with up to 3,000 times the efficiency of air cooling. In practice, data centers transitioning from air to 75% liquid cooling have demonstrated a 15.5% improvement in total energy usage effectiveness and over 10% reduction in total facility power. Liquid-cooled facilities routinely achieve PUE values of 1.03–1.06, versus the global air-cooled average of 1.55.
Nepal’s advantage here is structural. The country is threaded with over 6,000 rivers and rivulets originating in the Himalayas, among the world’s largest freshwater systems. Cold, clean, high-altitude river water is ideal for direct liquid cooling loops. Unlike arid data center destinations that must desalinate or recycle water at significant cost, Nepal has a virtually unlimited supply of cooling medium at near-optimal intake temperatures year-round.
At elevations above 1,400 meters (common across Nepal’s central hills and valleys) ambient air temperatures run 5–10°C cooler than lowland Asian data center hubs. This directly reduces the energy required for any supplemental air cooling and extends the hours of free-cooling operation. The combination of abundant cold water and cool ambient air means Nepal can achieve world-class PUE figures without the expensive mechanical cooling infrastructure required in tropical or arid climates.
True Green Compute: Not Greenwashing
A Near-Zero-Carbon Grid, by Default
When hyperscalers like Google, Microsoft, and AWS announce “100% renewable” targets, they typically mean purchasing Renewable Energy Certificates (RECs) to offset fossil-fuel consumption, a paper exercise that does not change the actual electrons flowing into their facilities. Nepal’s grid is different. 95% of electricity generation is hydropower. The grid’s carbon intensity is among the lowest of any country on Earth, approaching the physical floor. A data center in Nepal is genuinely powered by clean energy by default: no RECs, no offset games, no greenwashing.
As ESG mandates tighten and the EU’s Corporate Sustainability Reporting Directive (CSRD) begins requiring verified Scope 2 and Scope 3 emissions disclosures, the ability to point to an inherently clean grid becomes a competitive advantage for data center operators courting sustainability-conscious enterprise clients.
Geopolitical Neutrality & Data Sovereignty
A Non-Aligned Jurisdiction Between Two Superpowers
Nepal is a federal democracy, landlocked between India and China, and has maintained a policy of non-alignment for decades. It has no active military conflicts and no history of state-sponsored cyber operations. For organizations seeking jurisdictional diversification (particularly for sensitive AI training data, financial workloads, or healthcare datasets) Nepal offers a neutral sovereign territory that is neither aligned with the US-led tech ecosystem nor the Chinese one.
This neutrality, combined with proximity to a 1.4-billion-person Indian market, creates a distinct data sovereignty proposition. As India tightens data localization regulations and enterprises seek alternatives to established (and increasingly expensive) hubs like Singapore, Nepal can position itself as a compliant, cost-effective “near-shore” data jurisdiction.
The Case, Summarized
Why Nepal Solves the Trilemma
The data center industry faces three constraints that pull in opposing directions. Cheap power is often dirty power. Clean power is often expensive. Secure jurisdictions often lack infrastructure. Nepal is unusual because its natural geography and hydrology resolve these tensions simultaneously:
Nepal will not replace Singapore or Mumbai overnight. Resources alone don’t build data centers. The case has real gaps, and they deserve honest examination.
The Challenges
Connectivity is the weakest link
Nepal has zero subsea cable landings and only ~1.3 Tbps of total international bandwidth (NTA, 2023), compared to 50+ Tbps for Malaysia and 100+ Tbps for Singapore. All international traffic transits overland through Indian upstream providers (primarily Airtel and Tata) or partially through China Telecom. The national fiber backbone spans approximately 4,932 km, with nearly half the population living more than 10 km from a fiber node. Fixed broadband ranks 88th globally at 66 Mbps (Ookla, 2024). For latency-sensitive workloads, the extra hop through Indian transit adds meaningful milliseconds.
The trend, however, is moving fast. Nepal’s domestic internet traffic has grown from under 100 Gbps in 2019 to over 1 Tbps in 2025. Electricity demand has been on a similar curve. Fiber is being strung on the same transmission towers that major grid projects are building, and these projects include dedicated fiber components. Cross-border fiber capacity to India is expanding alongside the new 400 kV transmission corridors. The government is studying an international bandwidth platform to reduce dependency on Indian transit providers. Nepal’s connectivity today is where its power grid was a decade ago: underdeveloped, but on a steep trajectory. For latency-critical inference workloads, Nepal is not the right location today. For training workloads, where power cost matters more than ping time, the connectivity gap is less of a constraint.
The dry season problem
Nepal’s run-of-river dominated grid loses over 50% of generation capacity during winter (December–March). During these months, Nepal still imports 274–700 MW from India to cover domestic demand. A 24/7 data center cannot afford seasonal power interruptions.
The construction pipeline addresses this directly. Several major storage-type projects are underway (Dudhkoshi Storage 330 MW, Tanahu 140 MW, Tamor Storage 308 MW) that will provide year-round dispatchable power. NEA projects Nepal will cease dry-season imports entirely by 2026–27. Data center operators can also negotiate firm PPA contracts with storage-backed generators, and hybrid solar+battery installations can supplement dry-season hydro. The seasonal gap is a transitional problem, not a structural one.
Political and regulatory immaturity
Nepal has experienced frequent government changes, bureaucratic delays, and weak enforcement of contracts. There is no dedicated data center policy, no Tier certification infrastructure, and the regulatory framework for foreign investment in digital infrastructure is underdeveloped. NEA remains a monopoly utility that controls 100% of transmission and distribution.
This will not change quickly. But the trajectory is positive. The Electricity Regulatory Commission (ERC) has been established with ADB support, NEA has been profitable for multiple consecutive years, and the government has designated key grid projects as National Pride projects. Nepal’s Special Economic Zone Act provides tax holidays and duty exemptions. The institutional framework hasn’t yet been designed around data centers specifically, which means first-movers face bureaucratic friction but also negotiate the most favorable terms.
Natural disaster exposure
Nepal sits in an active seismic zone. The 2015 earthquake killed nearly 9,000 people and caused $7 billion in damage. Monsoon-season landslides and flooding routinely damage infrastructure, including transmission lines and hydropower plants. Climate change is accelerating glacial melt and altering river flow patterns.
Seismic risk is manageable with modern engineering. Data centers in Japan, which has higher seismic risk, operate at Tier IV reliability. The greater concern is transmission infrastructure vulnerability to landslides, which the new 400 kV backbone projects are designing around with improved routing and HTLS (high-temperature low-sag) conductors. Climate change cuts both ways: glacial melt increases short-term river flow (more power) while potentially reducing long-term water availability. This deserves serious hydrological modeling, not dismissal.
Competing markets are not standing still
Malaysia ($0.07/kWh, 2.4 GW DC pipeline in Johor alone) and Indonesia ($0.09/kWh, lowest annual power cost at $42M for a hyperscale facility per Wood Mackenzie) both have subsea cable landings, growing DC ecosystems, established regulatory frameworks, and larger domestic markets. Vietnam has rapidly grown solar to 16+ GW.
The carbon math changes the picture. Malaysia’s grid is only 8.3% renewable with carbon intensity of ~0.54 kg CO₂/kWh, and its DC emissions are projected to rise from 5.9 to 40 MtCO₂e by 2030 (Ember). Indonesia’s grid is 67% coal at ~0.78 kg CO₂/kWh. Nepal’s grid is 95% hydro at ~0.02 kg CO₂/kWh: 27x cleaner than Malaysia and 39x cleaner than Indonesia. As CSRD and Scope 2/3 reporting mandates tighten, this gap becomes a competitive moat, not a footnote.
Nepal also has structural cooling advantages: 5–10°C cooler ambient temperatures and unlimited cold river water versus tropical 30°C+ climates that burn energy on HVAC. Nepal is not competing for all workloads. It is competing for carbon-sensitive, power-intensive, latency-tolerant workloads, and on that narrow framing, it has no rival in Asia.
Talent and supply chain gaps
Nepal lacks a deep pool of data center operations engineers, network architects, and security specialists. The supply chain for servers, cooling equipment, and electrical infrastructure depends entirely on imports through Indian or Chinese border crossings. Construction timelines are notoriously unpredictable.
Hyperscale operators routinely build in greenfield locations (Iowa, Luleå, Johor) and import both talent and equipment. Nepal produces ~10,000 engineering graduates annually, and remote monitoring (which NEA is deploying via SCADA) reduces on-site staffing needs. The real question is whether the power cost delta ($50M+ per year for a 50 MW facility vs. Singapore) is large enough to justify the operational complexity. For many operators, it will be.
The Timeline
Nepal is a 5–10 year play, not a 2-year deployment. The grid upgrades underway will mature by 2028–2030. The construction pipeline will deliver year-round surplus by approximately the same timeframe. The regulatory and connectivity gaps need deliberate policy intervention: specific legislation modeled on what Singapore and Malaysia have done for their own DC industries.
But the resource fundamentals are permanent structural advantages that no competitor can replicate: cheap, clean, abundant power; unlimited cold water; natural cooling; geopolitical neutrality. Malaysia cannot move its grid to 95% renewable. Indonesia cannot cool its data centers with Himalayan snowmelt. Singapore cannot triple its land area. The question is not whether Nepal’s resources are world-class. The question is whether its institutions can keep pace with its rivers.
Comparisons between Nepal and Switzerland have been made for years: mountainous, landlocked, hydropower-rich, neutral, wedged between larger, economically powerful neighbours. The parallel has mostly been aspirational. Nepal will never be Switzerland. But if data is the 21st century’s gold, shouldn’t a mountainous neutral country be building 21st century banks?
Sources & References
About the author: I’m Aabhash Dhakal, a PhD candidate at the University of Würzburg working on Decision-Focused Machine Learning and Optimization. My research interests include energy systems and transport. More about me.
I’m working on a newsletter covering topics at the intersection of ML, optimization, and infrastructure. If this kind of analysis interests you, stay tuned.