AI in Nepal 2026: Policy, Startups and Outlook

For years, talk about AI in Nepal ran ahead of anything on paper. That changed in August 2025, when the government approved the National AI Policy 2082, the country's first attempt to set rules and ambitions for artificial intelligence in one document. By the middle of 2026, with a new House of Representatives elected in March and Balen Shah sworn in as Prime Minister, the policy is the reference point for almost every conversation about where Nepal's technology sector is headed. The harder question is how much of it will actually get built.

The policy that set the agenda

The National AI Policy 2082 is a framework rather than a law. It is built around human-centred values, data protection, transparency and accountability, and it positions AI as a driver of socio-economic development rather than a threat to be contained. In other words, the document treats artificial intelligence as something Nepal should adopt deliberately, with guardrails, instead of letting it arrive unmanaged through imported apps and services.

What makes the policy unusually candid is that it names its own obstacles. Rather than promising a finished digital future, it lists the conditions Nepal does not yet have. Those admissions matter more than the aspirations, because they describe the real starting line.

The gaps the policy identifies include:

• A shortage of accessible, quality datasets to train and test models
• The absence of an adequate legal and institutional framework for data security
• Insufficient infrastructure and a thin pool of skilled people
• Limited investment mechanisms for AI startups
• Concerns about biased outcomes and privacy violations

Each of these is a structural problem, not a quick fix. A policy can announce a target; it cannot conjure a labelled Nepali-language dataset, a data-protection enforcement body, or a generation of engineers who decide to stay home.

The institutions Nepal plans to build

The policy's implementation plan rests on new institutions and shared infrastructure. The intent is to move AI out of scattered private projects and give it a public backbone the way roads or the power grid support other sectors.

Planned element	Stated purpose
National AI Centre	Coordinate implementation, research and international cooperation
Provincial AI Excellence Centres	Training, innovation and research closer to the provinces
Regulatory sandboxes	Safe, supervised spaces to test AI products before wider release
Core infrastructure	Data centres, cloud systems and high-performance computing

The design is sensible on its own terms. A National AI Centre gives the state a single point of accountability; provincial centres push skills beyond the Kathmandu Valley; sandboxes let regulators learn how a product behaves before it reaches the public. The catch is that all four depend on sustained funding and technical staffing, and Nepal has a long record of announcing centres that open slowly or never quite open at all. The test for 2026 and 2027 is whether any of these moves from the policy text into a building with a budget and a payroll. Setting them up sits inside the government's wider digital-governance plan, which ties AI to broader public-sector digitisation.

Who is actually building AI in Nepal

While the state assembles its framework, a small private sector has been working on AI for years. Companies such as Fusemachines Nepal, an AI and education firm, and Paaila Technology, known for service robots, are among those at the forefront of AI development in the country. They matter not because they rival global labs, but because they have trained Nepali engineers, shipped working products, and shown that AI work can be done from Kathmandu rather than only consumed from abroad.

Around them sits a wider layer of smaller teams and student projects building chatbots, optical character recognition for Devanagari, document tools and analytics. For most ordinary users, the practical entry point is still imported software, which is why a separate guide to the best AI tools for Nepali students and businesses covers the day-to-day side. The state-of-play question here is different: how does a homegrown AI industry grow when capital, compute and talent all pull toward larger markets?

The international link-up

One answer in 2026 is partnership. The second cohort of the India-Nepal Startup Partnership Network (IN-SPAN) began on 1 June 2026, taking 25 Nepali startups into an eight-week, fully funded programme at the IIT Madras Pravartak Technologies Foundation in Chennai. Programmes like this give early founders mentoring, infrastructure and a route into a much larger ecosystem next door. They are useful, but they are also a reminder that some of the deep technical scaffolding Nepali startups need is, for now, easier to reach across the border than at home.

The budget and the digital push

Policy needs money behind it, and the FY2026/27 budget tries to supply some. It backs a National Digital Infrastructure, including a government cloud and a National Data Policy, and makes an explicit push on AI, data analytics and cybersecurity. A government cloud is the kind of plumbing that, if it works, would give the National AI Centre and provincial centres somewhere to actually run workloads, and a National Data Policy could begin to address the data-security gap the AI policy admits.

The risk is the familiar one: a budget line is a promise of intent, not a built system. Government cloud projects in particular are expensive to run and maintain, and they live or die on procurement quality and long-term staffing rather than on the announcement itself.

The sovereign "AI Factory" debate

One live argument cuts to the heart of all this. A Kathmandu Post column has argued that Nepal should build a sovereign "AI Factory" — its own concentrated compute capacity — rather than depend entirely on foreign cloud providers. The case for it is about control: owning the hardware means owning where Nepali data sits, who can see it, and whether critical services can be switched off from outside the country.

The case against is cost and reliability. High-performance computing is expensive to buy, hungry for stable electricity, and quickly outdated. A country still working on consistent power supply and skilled operators has to weigh whether scarce capital is better spent on a national data centre or on renting capacity and investing in people. Both sides agree on the destination; they disagree on the order of the steps. For now, this is a genuine open debate rather than a settled plan, and how it resolves will shape what kind of AI infrastructure Nepal actually owns.

The real gaps, stated plainly

It is worth being honest about Nepal's AI maturity in 2026. The country has a policy, a handful of capable firms and a budget signal. It does not yet have the foundations that turn those into widely used systems. Several constraints are structural:

1. Brain drain. Many of Nepal's strongest engineers leave for jobs abroad, thinning the talent the policy assumes will be available.
2. Cost of compute. Training and running modern models is expensive, and access to high-end hardware is limited.
3. Electricity reliability. Data centres and high-performance computing need stable, continuous power, which is not guaranteed everywhere.
4. Small local-language datasets. Quality Nepali-language data, especially in Devanagari, is scarce, so models often work better in English than in the language most people use.
5. Weak data-protection enforcement. Rules exist mostly on paper, and the enforcement machinery the AI policy calls for is not yet in place.

These gaps are not only technical. They are also about trust. Without enforced data protection, public confidence in government AI projects is fragile, and misuse becomes easier — a worry already visible in cases of AI deepfakes of Nepali politicians circulating during the recent political turbulence. Adoption and safeguards have to advance together, or the policy's human-centred language stays decorative.

What comes next

The next two years are about execution, not vision. The measures of progress are concrete: does the National AI Centre open with a real budget? Does even one provincial Excellence Centre run a training cohort? Does a regulatory sandbox actually admit a product? Does the government cloud host live services? If those answers turn positive, Nepal will have built the base it currently lacks. If they stall, 2026 will be remembered as the year the policy was written rather than the year AI in Nepal took hold. The framework is finally in place; what it produces is still unwritten.

Frequently Asked Questions

When did Nepal approve its National AI Policy?

Nepal approved the National AI Policy 2082 in August 2025. It is a framework for ethical, responsible and inclusive AI, built around human-centred values, data protection, transparency and accountability.

What institutions does the policy plan to create?

It plans a National AI Centre to coordinate implementation and research, provincial AI Excellence Centres for training and innovation, regulatory sandboxes for safe experimentation, and core infrastructure such as data centres, cloud systems and high-performance computing.

Which companies are building AI in Nepal?

Firms such as Fusemachines Nepal, an AI and education company, and Paaila Technology, a robotics and AI company known for service robots, are among those at the forefront of AI development in the country.

What is the IN-SPAN programme?

The India-Nepal Startup Partnership Network's second cohort began on 1 June 2026, placing 25 Nepali startups in an eight-week, fully funded programme at the IIT Madras Pravartak Technologies Foundation in Chennai.

What are the biggest obstacles to AI in Nepal?

The main constraints are brain drain, the cost of compute, unreliable electricity, scarce Nepali-language datasets in Devanagari, and weak enforcement of data-protection rules. The AI policy itself names several of these gaps.