Talent, founders, and the playbook

The country has perhaps a few thousand people who could plausibly build AI products in Nepal — Pulchowk and KU graduates, returning diaspora, mid-career engineers shifting in from web and mobile, a small but growing cohort of self-taught builders. This is more than people often think and less than the rhetoric suggests. Where they spend their time over the next five years matters more than almost any other variable in this story.

The pipeline as it actually is

The pipeline that produces working AI engineers in Nepal has four legs.

Undergraduate engineering. IOE Pulchowk, Kathmandu University, Kantipur Engineering College, NCIT, and a handful of others produce a few hundred technically capable graduates per year. The curriculum is uneven on modern ML; the strong students teach themselves and find ways to work on real projects before they graduate.

Domestic apprentice work. Fusemachines, Cloudfactory, Genese, Leapfrog Technology, the bigger product startups, and the major banks all train juniors on real problems. A two-year tour at one of these is, in practice, the standard finishing school for a Nepali AI engineer.

Foreign graduate study. A substantial fraction of strong Nepali engineers go abroad for an MS or PhD — to the US, increasingly to India, occasionally to Europe and Australia. A meaningful fraction come back. Many do not. The diaspora pipeline (next section) is part of the picture.

Self-taught builders. A growing cohort of young Nepalis learns ML the way an earlier cohort learned web development — through online courses, open-source projects, and small consulting jobs. By the time they are 25, the strongest of them are technically indistinguishable from the formally-trained graduates.

What founders actually need

If you talk to Nepali AI founders honestly about what holds them back, three answers come up repeatedly.

Patient, local capital. Nepali banks do not lend to early-stage product companies. Domestic VC barely exists at the seed stage. Most early money is friends-and-family, founder savings, or grants. This means founders default to consulting work to fund the company, which means the product never gets the focus it needs. A small, patient seed-fund infrastructure — even modest by global standards — would change the trajectory of dozens of companies.

A buyer they can sell to. The biggest commercial AI buyers in Nepal are the banks and the telcos. Selling to them as a young startup is procurement-hard: long sales cycles, requirements designed to filter out small vendors, occasional opacity around vendor selection. A domestic public-sector buyer of AI products — a procurement framework that explicitly favours small Nepali firms for public-good AI, with clear, fair evaluation — would unlock real revenue for the early-stage companies.

Talent at the senior level. Hiring a junior engineer in Nepal is comparatively easy. Hiring a senior person who has shipped a production ML system at scale is hard — most of them are abroad. Even one or two senior people in a startup can change what is possible; the same startup without them tops out at junior-quality work.

What a credible four-year playbook looks like

For a serious Nepali AI founder starting today, a working four-year playbook tends to look roughly like this.

Year one. Pick a vertical where the founder has domain access nobody else has — a cooperative network, a hospital chain, a province-level government contact, a specific commodity supply chain. Spend the first six months in the vertical understanding the actual workflow rather than building. Ship a small, focused product that solves one specific pain in that workflow. Get two paying customers, even if at break-even pricing.

Year two. Productise. Hire a small team — two engineers, one domain person — and turn the bespoke first product into something that can be deployed by a third customer without the founder personally configuring it. Take the first outside money, ideally from a mix of patient domestic capital and one strategic foreign investor with India or Southeast Asia operating experience.

Year three. Expand the product surface horizontally — adjacent problems in the same vertical, not new verticals. Begin to build network effects: data accumulating in the product makes the model better, which keeps customers in. Hire the first senior person, often a returning diaspora engineer who has done a tour at a big tech firm.

Year four. Either you have a category-defining position in your vertical (twenty-plus paying customers, defensible data moat, real revenue), in which case you raise a serious Series A from a regional or international fund and expand to a second country — usually India first; or you sell to a strategic buyer or a larger Nepali firm and start the next thing.

This playbook does not work for every kind of AI startup, but it has worked, with variations, for several of the small number of credible Nepali AI companies that exist today.

The diaspora question, slightly

The Nepali AI diaspora — engineers at OpenAI, Anthropic, Google DeepMind, Meta AI, Microsoft Research, plus the broader engineering diaspora at the big tech firms and the AI-heavy startups — is, person for person, probably the country’s largest single AI asset. The question is not whether to engage them; it is how.

What works: structured advisory relationships, time-boxed sabbaticals, summer engineering institutes hosted in Kathmandu, equity in Nepali companies, joint research arrangements with universities. What does not work: asking them to “come home” without a credible role to come home to.

The country’s most useful policy lever here is not visa schemes or tax breaks. It is making the work itself attractive — real problems, real scale, real autonomy. A diaspora engineer comes back for a chief-AI-officer role at a Nepali bank with a real budget faster than for a tax break.

Check your understanding

Quick check

—

A founder is starting an AI company in Nepal. Which strategy is best supported by the experience of the small set of credible Nepali AI companies that exist today?

Pick the largest possible global market first, build a generic LLM wrapper, and try to scale internationally before any Nepali traction.
Pick a vertical where the founder has rare domain access (a cooperative network, a hospital chain, a specific supply chain); spend the first six months understanding the actual workflow before building; ship a focused product solving one specific pain; get two paying customers; expand horizontally within the vertical.
Hire only abroad to avoid the local talent constraint.
Build a foundation model from scratch using only Nepali data.

What comes next

This leaves us with one section. What does a realistic, actionable five-year agenda look like — for the country, for the institutions, for the people who will build this? The closing section is the synthesis.