The Himalayan Compute Company
Hydropower, cold air, and high ground, turned into intelligence. Built in Nepal. Built for the world. Starting at 50 MW. Built out to 1 GW.
Silicon Peaks Compute
01The demand
WorldEvery serious power is building AI factories the way earlier centuries built ports and railways. The argument about whether is over. What remains is where, and who owns what comes out.
Global data-center power demand is on track for roughly 2,000 GW by 2050 against about 600 GW of committed supply. The 1,400 GW between those numbers is the opening.
Countries that host compute keep their data, their language, and their models at home. Countries that only import intelligence pay rent on their own future.
In their words
Tokens are the new commodity. AI factories are the infrastructure that produces them.
Jensen Huang, NVIDIA, 2026
There will be a significant power shortfall next year, so not like the distant future.
Elon Musk, 2026
We see a future where intelligence is a utility like electricity or water and people buy it from us on a meter.
Sam Altman, OpenAI, 2026
FIG. 01The divergence. Global data-center power, GW
2020 to 2050IEA, Energy and AI 2025. Wood Mackenzie. Uptime Institute. Indicative trajectories. REV 2026.09
0GW
demand by 2050
0GW
committed supply
0GW
the opening between them
02The ledger
FIG. 02Roughly 30,000 MW of new hydropower comes online over the next decade, in a country whose peak demand today is about 3,200 MW. Last year Nepal exported 3.9 TWh at about five cents a kilowatt hour.
The same electricity, sold as compute, earns roughly twenty times more.
Nepal Electricity Authority public reporting, FY 2025/26. Under five percent of theoretical potential is built. REV 2026.09
0
rivers and rivulets
0TWh
exported, FY 2025/26, at about $0.05 per kWh
0×
compute ceiling relative to GDP
$0to 7.1T
compute ceiling, Compute Net Worth Index
FIG. 03The spread. Same water, same turbines.
InteractiveIndicative. Anchored to Nepal Electricity Authority public reporting and Silicon Peaks' published compute-ceiling analysis, a twentyfold spread. Assumes full utilization at plant rating. Slide from 50 MW, the first node, to 1 GW, the build-out. Not an offer of securities.
03The reach
FIG. 04A kilowatt hour that starts as falling water at altitude costs less than any other kind, arrives cold, and lands one fiber hop from a quarter of humanity.
IEA, Electricity 2028. BloombergNEF LCOE Survey H2 2025. Operator filings. Axis runs to $0.30. REV 2026.09
0to 1.3
design PUE, cold climate
≤0ms
to tier-1 APAC hubs
0B
people within reach of Kathmandu
Reach
Great-circle from KathmanduFrom Kathmandu. Half the planet within reach.
Proximity to 2.5 billion people.
Arcs: Delhi · Kolkata · Dhaka · Mumbai · Bengaluru · Colombo · Bangkok · Singapore · Dubai · Hong Kong · Tokyo · Seoul · Frankfurt · London · San Francisco. Nearby corridors pulse faster.
Gravity is the cheapest form of energy ever discovered.
04On water, since people ask
Closed loopAI factories carry a reputation for consuming a lot of water. That reputation belongs to an older generation of machines. Modern halls run closed loops: fill once, circulate for years, like the radiator of a car.
The drinking taps stay untouched.
The myth
Data centers drain municipal water and dry village wells.
The fact
Closed loops fill once and recirculate. Daily draw from drinking supplies is near zero. A year of operation uses less water than a small village pond holds.
The myth
AI factories steal electricity from homes and bring back load shedding.
The fact
They fund their own substations and transmission, and they buy the monsoon surplus that today spills past the turbines unsold.
The myth
Server halls are empty warehouses that occupy land and employ nobody.
The fact
Construction employs thousands. Operations sustain high-skill engineering careers for decades, and the tax bill supports a district.
01 THE DEMAND
Nepal: local-language AI, universities, startups and public-service applications.
South Asia and APAC: regional inference, enterprise workloads and research.
Global customers and off-takers: AI platforms, multinationals and dedicated compute capacity.
Kathmandu · South Asia · APAC · Global markets
02 THE ECOSYSTEM
Delivering compute takes an ecosystem: technology, capital, the expertise to build and operate, and customers who need the capacity.
Silicon Peaks is bringing those pieces together around a facility in Nepal.
Technology · Delivery · Operations
03 THE FACILITY
Compute, cooling, power distribution and connectivity, designed as one coordinated environment for running AI workloads.
Each system has a clear role. Their integration turns individual components into useful capacity.
Silicon Peaks Compute
04 THE COMPUTE
Accelerators do the computation. Memory keeps the working data close. High-speed connections let machines operate as one system.
A representative server tray opens to reveal the relationship between the chip and its coldplate.
Accelerators · Memory · Internal network
05 THE COOLING
Cooling starts at the chip. Liquid carries heat through the cooling circuit to distribution and heat-exchange equipment.
Compute, liquid cooling and the power system stay connected as the facility opens.
Chip · Coldplate · Coolant loop · Heat rejection
06 THE FOUNDATIONS
Below the city sit the physical foundations: land, water infrastructure, the power interface and optical fiber.
Those foundations connect the facility to the resources it needs and the customers it serves.
Land · Water · Energy · Connectivity
07 THE OPPORTUNITY
Foreign direct investment. Multinationals. Local enterprise. An ecosystem for builders, skilled work and new businesses.
The ambition is lasting economic progress: stronger productivity, better jobs, and a greater contribution to Nepal's GDP.
FDI · Multinationals · Jobs · Economic growth
08 THE COMPLETE SYSTEM
Customers, partners, a purpose-built facility, Kathmandu's enterprise and the foundations beneath it.
Each part makes the others possible. Together, they connect Nepal's resources and talent to the world's demand for intelligence.
One connected system.
Routes are illustrative audience connections, not contracted capacity or measured latency. FDI, jobs and GDP are ambitions, not forecasts. NVIDIA is the only legible partner mark, shown at our direction; visibility is not a verified commercial claim. Other partners remain anonymized.
06Phase 1
N1P1The first node. What can be said is below. The rest stays under NDA.
Already in motion. No names and no logos, by design. Just the shape of the room.
Some things in the infrastructure of the coming century will be built once.
07The people
Names on requestFour founders, born in Nepal. Educated at the world's top universities, trained in Silicon Valley and on Wall Street, now building at home. Names on request.
A·T
Kathmandu
L·M·B
Kathmandu
P·C·H
San Francisco
R·A
Miami
Educated & trained at
Harvard
Stanford
Columbia
Brigham Young
Heidelberg
NUS Singapore
Tribhuvan, PulchowkWe work under mutual NDA from first contact.
$0B+
enterprise value built at prior companies
0+
years of operating experience
0
prior exits
08Inquire
48 hours, mutual NDAOne paragraph on who you are and what you hold or need. We reply within forty-eight hours, under mutual NDA.
Every reply under mutual NDA. Kathmandu. San Francisco.
09Common questions
FAQSilicon Peaks is the idea that the Himalaya can be for the intelligence age what Silicon Valley was for the software age: a technology hub where clean power, cold air, high ground, and a young technical workforce concentrate. Anchored in Nepal, it serves South Asia and the Asia-Pacific with compute, jobs, and an ecosystem the region currently imports.
A Nepali company, incorporated in Delaware, building AI compute infrastructure where Himalayan water power is abundant, clean, and cold. We work with partners and governments under NDA. In short: the Himalayan Compute Company.
Upwards of 100,000 MW (100 GW) of theoretical hydropower with under 5 percent built, cold air at altitude, and 1.4 billion neighbors one fiber hop away. The inputs of the intelligence age sit in one geography.
Less fresh water in a year than a neighborhood tea shop uses in three months. The water circulates in sealed pipes like the radiator of a car, and the drinking taps stay untouched.
Write to founders@siliconpeaks.ai with one paragraph on who you are and what you hold or need. Replies within 48 hours, under mutual NDA.
Silicon Peaks Compute is a Delaware corporation with a dual base: Kathmandu for land, power, and build; San Francisco for capital and governance.
Silicon Peaks Compute