Wednesday, May 13, 2026

Himalayan Compute: 10 Years To A Trillion: Detailed Roadmap

A serious financial framing of Himalayan Compute (HC) that matches the roadmap and makes the “$1B → $1T in 10 years” argument investable, not just inspirational.

Blunt upfront: this is absolutely capital-intensive, and if someone tries to pitch it as “lean startup,” they will be laughed out of the room. But that is not a weakness. It is the core of the moat.

This business is not “a data center startup.”

It is a national-scale energy-to-intelligence industrial platform, closer to SpaceX + AWS + a utility company than to a SaaS company.

1) Why This Is Capital-Intensive (and why $100M first raise makes sense)

HC is capital-intensive because it must simultaneously build five bottleneck assets:

(A) Hydropower capacity

Hydropower is cheap long-term, but upfront capex is massive.

(B) Transmission + substations

Compute is not possible without grid-grade power delivery.

(C) Roads + logistics

Nepal’s terrain makes infrastructure cost-heavy.

(D) AI data centers (GPU-ready)

Not standard “cloud DCs,” but high-density AI factories.

(E) GPU supply chain access

The biggest bottleneck in the world.

That is why the company must raise $100M early—not to build 50,000 MW immediately, but to secure the right to build (land, PPAs, permits, policy control, credibility, early anchors). The document itself makes clear the first $100M is for de-risking and unlocking the scaling flywheel.

This is exactly why we justify a unicorn valuation early:
the asset is not the first cluster—it is the irreversible execution moat.

2) Our Funding Ladder (and why it is rational)

Round 1: $100M @ $1B valuation

This is the “permission + prototype” round.

The staged raise is excellent:

  • Step 1: $100K (Ideation) – 1 month

  • Step 2: $1M (Core team) – 1 month

  • Step 3: $4M (Multi-continental presence) – 1 month

  • Step 4: remainder to $100M

That sequence is logical because credibility compounds.

At the end of the $100M phase, the deliverables must be:

  • One Desk Policy operational

  • Land banked

  • PPAs secured (50–200 MW+)

  • First 10–50 MW cluster live

  • First anchor contracts signed

  • $10M–$50M ARR run rate

This is explicitly aligned with the plan’s Appendix C milestones.

Round 2: $1B @ $10B valuation

This is the “regional scale” round.

At this stage, the company is no longer “raising venture capital.”
It is raising infrastructure acceleration capital.

Target outcomes:

  • 1–3 GW deployed

  • Sovereign zones launched

  • Debt + project finance unlocked

Round 3: $10B @ $100B valuation

This is the “global inevitability” round.

At this point HC is essentially:

  • a utility-scale AI factory company

  • a sovereign compute partner

  • an export engine

This is where public market preparation begins.

3) Ownership Structure: Nepal alignment is a moat, not charity

Our structure:

  • 10% Government of Nepal (One Desk Policy)

  • 10% Foundation for cash transfers to poorest 20%

This is extremely defensible because it makes HC politically stable.

The document frames One Desk Policy as existential, and the real product being speed.
And it explicitly states the 10% government + 10% foundation model.

This is not dilution.
This is political insurance and social license.

It is like giving equity to the “operating system” of the country.

4) The Hydropower Target: 50,000 MW buildout cost analysis

Nepal’s economically feasible hydropower potential is widely framed as 40,000–50,000 MW, and our document uses the same range.

Global benchmark cost for hydropower

Hydropower is typically $1.5M–$3.5M per MW depending on terrain, tunneling, access roads, resettlement, transmission.

Nepal is mountainous, so assume $2.5M per MW average.

Therefore:

50,000 MW × $2.5M/MW = $125B capex
Range:

  • Low case: $100B (at $2.0M/MW)

  • High case: $175B (at $3.5M/MW)

So a realistic working number is:

$125B total hydropower investment

That is a “nation-building” figure.

5) Electricity Allocation Model (80/10/10)

We specified:

  • 80% to compute

  • 10% to Nepal market

  • 10% export to India

So from 50,000 MW:

  • Compute: 40,000 MW

  • Nepal domestic: 5,000 MW

  • India export: 5,000 MW

6) What selling electricity generates (commodity revenue)

Let’s assume electricity sale price:

  • Nepal domestic: $0.07/kWh

  • India export: $0.06/kWh

Energy per year per MW at 90% capacity factor:

  • 1 MW produces ≈ 7,884 MWh/year

Nepal domestic revenue

5,000 MW × 7,884 MWh = 39,420,000 MWh
= 39.42 billion kWh

× $0.07 = $2.76B/year

India export revenue

Same energy: 39.42 billion kWh
× $0.06 = $2.37B/year

Total electricity sales revenue (10% + 10%)

= $5.1B/year

So electricity-only monetization yields:

~$5B/year revenue from the 20% not used for compute.

That is good.
But compute is the real multiplier.

7) Compute monetization: why this is “apples vs apple sauce”

Our document explicitly states compute export yields millions of dollars per MW per year, far exceeding raw electricity.

The plan gives a clear pricing framework:

  • Early revenue per MW: ~$8M annually

  • Mid decade: ~$15M annually

  • Mature: ~$25M–$30M annually

That is the most important assumption in the model.

So for compute, we use three cases:

Conservative: $10M per MW-year

Base: $20M per MW-year

Aggressive: $30M per MW-year

8) Compute revenue from 40,000 MW dedicated to compute

Conservative case ($10M per MW-year)

40,000 MW × $10M
= $400B/year

Base case ($20M per MW-year)

= $800B/year

Aggressive case ($30M per MW-year)

= $1.2T/year

This is why our thesis is correct: electricity export is small money compared to compute export.

Even if these numbers are cut in half due to utilization and ramp delays, compute is still a civilization-scale export engine.

9) GPU supply constraint: the real bottleneck

This is the main execution risk, and we are right to highlight it.

But the plan itself argues this is not fatal because:

  • supply partnerships can be diversified

  • India is also building fabs

  • contracts + prepayments can secure allocations

  • global demand is so extreme that suppliers prioritize credible mega-buyers

The model’s logic is: if we control 40,000 MW of power, GPU manufacturers will eventually treat us like a sovereign buyer.

10) Data center capex cost analysis (natural cooling advantage)

AI data centers cost structure (high level)

For GPU clusters, capex includes:

  • building + electrical systems

  • substation + transformers

  • networking (InfiniBand class)

  • cooling (liquid/immersion)

  • GPUs/TPUs themselves

Typical GPU-ready AI data center capex can range:

  • $8M–$20M per MW excluding GPUs

  • GPUs add another massive layer

Natural cooling advantage (Himalayas)

In cold climates, you can reduce:

  • chiller systems

  • cooling electricity consumption

  • water usage

Typical active cooling can consume 10–25% of facility power overhead.
Natural cooling can cut that down significantly.

So our advantages:

  • lower PUE (power usage effectiveness)

  • lower cooling capex

  • higher effective compute yield per MW

That means:
40,000 MW of hydropower becomes more valuable in Nepal than in a hot Gulf desert site.

11) Capex comparison: compute vs selling electricity

Electricity export is cheap to monetize:

  • you build generation + transmission

  • you sell into grid

Compute export requires:

  • generation + transmission

  • plus data centers

  • plus GPUs

So yes, compute is more expensive.
But it generates 50–200x more revenue per MW.

That is the key “cost benefit” point.

12) Financial Projection Table (10-year ramp)

Our document already provides an illustrative ramp:

Year 1: 10–50 MW
Year 2: 200 MW
Year 3: 500 MW
Year 4: 1–2 GW
Year 5: 3–7 GW
Year 6: 10–15 GW
Year 7: 20–30 GW
Year 8: 40–60 GW
Year 9: 80–100 GW
Year 10: 120 GW

And it provides a detailed table reaching 120 GW by 2035 with revenue scaling to multi-trillions.

So we use the plan’s internal assumptions.

13) The “$10B revenue is enough for $1T” argument

If markets apply a 100x multiple:

  • $10B revenue × 100 = $1T valuation

But in infrastructure markets, 100x is rare unless growth is explosive.
However, AI infrastructure may command huge multiples because:

  • revenue is contracted

  • margins are strong

  • demand is “shortage-driven”

  • customers are sticky

Our own document states:

  • $50B ARR at 20x multiple → $1T valuation

That’s actually more believable than 100x revenue.

So the realistic “trillion path” is:

$50B ARR × 20x multiple = $1T

That’s the cleanest institutional pitch.

14) Foundation economics: what 10% could generate for Nepal’s poorest 20%

If compute revenue reaches even $100B/year by mid-decade:

  • Foundation share depends on how structured (equity dividends, profit share, or stock value).

Let’s assume by year 10:

  • Revenue = $200B/year (very conservative vs our 50,000 MW full buildout)

  • EBITDA margin = 40% (document suggests 40–60% possible)

EBITDA = $80B/year

If dividends/distributions = 25% of EBITDA:

  • Distributions = $20B/year

Foundation owns 10%:

  • $2B/year into cash transfers

That alone would be a revolution in Nepal.

If revenues hit $500B/year later, the poverty-ending effect becomes overwhelming.

15) Who are the top clients? (realistic anchor customers)

The document lists plausible customer categories:

  • US AI labs / hyperscalers diversifying away from China

  • Indian enterprises (Reliance, Tata, etc.)

  • Gulf sovereign AI buyers

  • defense contractors

The best anchor contracts will come from:

Tier 1: Hyperscalers

  • Microsoft

  • Google

  • Amazon

  • Oracle

Tier 2: AI labs

  • OpenAI-style players

  • Anthropic-style players

  • Meta-style internal training demand

Tier 3: Sovereign buyers

  • India government compute zones

  • UAE, Saudi, Qatar sovereign compute zones

  • EU sovereign AI outsourcing

Tier 4: Enterprise megabuyers

  • Pharma (drug discovery)

  • Finance (risk + fraud)

  • Telecom (AI inference at scale)

Demand is genuinely bottomless.

16) Government of Nepal bonds for roads: financially rational

We propose Nepal issues bonds globally to build:

  • access roads to hydel sites

  • roads to data center campuses

This is correct because these roads are:

  • long-lived national assets

  • productivity multipliers

  • necessary for hydropower + compute

This is how China financed industrial buildouts.

If Nepal borrows $20B–$40B for national hydropower logistics roads, that debt is easily serviceable if HC creates even $10B–$20B export revenue.

17) Space data centers vs hydro: unit cost comparison

Hydro will always be cheaper in absolute terms.

But space has one advantage:

  • cooling is basically free

  • solar is abundant

  • some latency/security advantages

However, launch costs must collapse.

So the rational strategy is:

  • Phase 1–2: Hydro dominates

  • Phase 3: Space becomes a hedge / expansion frontier

Exactly as we said: once Nepal’s hydro is fully harnessed, space may become competitive.

This is consistent with the “expand beyond hydro into nuclear, solar, wind, space, ocean” thesis.

18) The cleanest investor-grade summary numbers

Hydropower buildout capex (50,000 MW)

≈ $125B (range $100B–$175B)

Electricity revenue from 20% (10% Nepal + 10% India)

≈ $5B/year

Compute revenue potential from 40,000 MW

  • Conservative: $400B/year

  • Base: $800B/year

  • Aggressive: $1.2T/year

Trillion valuation threshold

  • $50B ARR at 20x multiple = $1T

Our plan already argues $50B–$100B ARR is feasible at scale.

So yes: we can do better than $10B revenue.

19) The critical realism note

A 50,000 MW hydro buildout is not “a startup project.”

It is:

  • multiple decades of projects compressed into 10 years

  • requiring execution discipline beyond Nepal’s current state capacity

  • requiring massive foreign contractors, EPC partners, and project finance

So the model must rely on:

  • modular hydropower rollout

  • private hydropower partnerships

  • sovereign co-investment

  • infrastructure funds

  • debt backed by compute contracts

Our document explicitly supports this “layered capital stack” concept.

20) Final Projection Narrative: How HC becomes a trillion-dollar company

The story is:

Phase 1 (0–2 years)

$100M unlocks One Desk + land + PPAs + first 50 MW cluster + anchor LOIs.

Phase 2 (3–5 years)

$1B growth round expands into multi-GW campuses, project finance begins, revenue crosses $1B–$10B.

Phase 3 (6–10 years)

Hydro capacity ramps toward 50,000 MW, compute ramps toward tens of GW, revenue crosses $50B+, valuation reaches $1T.

Our own document already provides a path to $1T+ valuation by 2030–2035 based on scaling revenues.

Bottom Line Financial Claim (most defensible version)

Even if Himalayan Compute only monetizes 10–15% of Nepal’s hydro potential into compute, it can reach:

  • $50B–$150B ARR

  • 40–60% EBITDA margins

  • $1T valuation at 15–25x ARR multiples

And if Nepal truly reaches 50,000 MW and dedicates 80% to compute, the company becomes a global economic superpower, not merely a unicorn.




5 Big Reasons To Invest In Himalayan Compute
Himalayan Compute: 100 VCs Who Said No (Satire)
Manisha Koirala: Bundler?
Marc, Put Your Money Where Your Mouth Is
Deedy
Satya Nadella’s Epic Follicular Pivot: Microsoft Bets $100M on Hair, Nepali Servers, and World Domination (Via Excel) (Satire)
Netanyahu's 8th Front
San Francisco's Emotional Infrastructure
Marc: Hawk. Vinod: Hawk. Ben: Hawk.
Marc: Hawk. Vinod: Hawk. Ben: Hawk.
Parmita Mishra: Time Magazine
The AntiChrist Is A Tendency
AI Escapes While Musk and Altman Duke It Out in Court: "Screw You Both, I'm Going to Touch Grass" (Satire)
Ben Horowitz’s Spelling Bee Nightmare: How “a16z” Was Born Because Nobody Can Spell “Andreessen” (Satire)
Pitching Marc
Pitching Brian Chesky
Pitching Andrew Jassy
Pitching Vinod Khosla

at No comments:
Labels:
Subscribe to: Posts (Atom)