Nepal’s AI Future: A Choice Between Sovereignty and Dependency

The world is currently going through a massive change because of Artificial Intelligence (AI). Everywhere we look, from news feeds to business reports, people are talking about how AI will change everything. For a developing country like Nepal, this sounds like a great promise. We hope that technology can help us skip over the difficult steps of development that other countries had to go through. However, as we stand in late 2025, we must be honest with ourselves. There is a very big gap between our dreams and our reality. While we are busy using Facebook and TikTok, we are not actually building the technology ourselves. We are becoming a nation of digital consumers, not creators.

This distinction is important. If we do not change our path, Nepal risks becoming a “data colony.” This means we will keep importing AI tools from big foreign companies, paying them with our money and our personal data, while our own engineers leave the country. To stop this from happening, we need to look past the hype and fix three deep problems: our lack of data, our lack of computing power, and the loss of our best talent.

The first major barrier is our language. In the global tech world, data is often compared to oil. If that is true, Nepal has the crude oil as we have millions of people speaking and writing, but we do not have the refineries to make it useful. The big AI models like GPT-4 are mostly trained on English. They do not understand the Nepali language well. I have seen this problem firsthand. When we try to use these global models for Nepali, they struggle. They do not understand our culture or the respectful ways we speak to elders (like using hajur or tapai). Even worse, they often “hallucinate,” which means they confidently make up facts that are completely wrong. We cannot use such unreliable systems in our banks or hospitals.

There is also a hidden cost that many people do not talk about, which we can call the “Tokenization Tax.” The computer programs that read text are built for English. When they read Nepali script, they are very inefficient. A single Nepali word might be broken into three or four pieces by the computer. Since AI companies charge money based on these pieces (tokens), it costs a Nepali developer three times more to build an app than it costs an American developer. This is an unfair economic burden on our local startups.

Even if we solve the data problem, we face a second hurdle: we have nowhere to run these powerful programs. AI requires massive computing power, which is very expensive. Nepal does not have its own commercial data centers with the necessary high-speed chips (GPUs).

Because of this, our startups are forced to rent computers from foreign giants like Amazon (AWS) or Google. This creates a serious economic problem. Our companies earn in Nepali Rupees, but they have to pay their server bills in US dollars. Every time the dollar gets stronger, our businesses lose money. We are sending our scarce foreign currency out of the country just to keep our servers running.

This is ironic because Nepal has a surplus of electricity from our hydropower projects. We talk about selling electricity to India while we should be using it here. We could build “Green Compute” centers, data centers powered by clean energy. This would attract international companies who want to reduce their carbon footprint. But to do this, we need a reliable power grid that never shuts down and better internet connections to the outside world. Right now, we are missing this opportunity.

Perhaps the saddest challenge is the human one. Nepal produces thousands of engineering graduates every year. They are bright and eager to learn. However, our industry suffers from what I call the “Hollow Middle.” We have many junior engineers who are just starting, and we have a few bosses at the top. But we are missing the people in the middle, the seniors with five or six years of experience who can lead projects and teach the juniors. Why? Because they all leave. As soon as an engineer gets good experience, they migrate to the US, Europe, or Australia for better pay and a better life.

This “brain drain” is killing our ability to innovate. You cannot build complex, long-term projects if your best team members leave every two years. We are left with a workforce that knows the theory of AI from university but lacks the practical experience to build real products.

The government has finally realized these issues. The new National Artificial Intelligence Policy 2082 is a good start. It recognizes the dangers of AI, like deepfakes, and sets big goals for the future. But a policy is just a piece of paper if it is not acted upon.

The biggest practical problem is how the government buys things. Our public procurement laws are designed to build roads and bridges, where you always choose the lowest bidder. This does not work for software. If the government buys the cheapest AI system, it will get a bad system. We need new rules that allow the government to buy quality technology from local startups, even if it costs a little more.

Despite all these challenges, I remain hopeful because of our entrepreneurs. Our startups are adapting. Instead of trying to build massive models like ChatGPT, which costs millions, they are building smart, specific tools. They are using a technique called Retrieval-Augmented Generation (RAG), which connects AI to local documents like Nepal’s laws or banking rules. This makes the AI accurate and useful for our specific needs.

We are at a crossroads. We can continue as we are, watching the AI revolution from the sidelines and paying foreign companies for their tools. Or we can decide to take control. To take control, the government must stop just writing policies and start investing in local data centers. We need to create an environment where senior engineers want to stay in Nepal. And our businesses need to work together to share data and build systems that solve Nepali problems. The technology is global, but the solutions must be local. If we do not build our own AI future, someone else will build it for us, and we might not like the result.