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

The world is at the moment going by an enormous change due to Synthetic Intelligence (AI). All over the place we glance, from information feeds to enterprise reviews, persons are speaking about how AI will change every little thing. For a growing nation like Nepal, this seems like an amazing promise. We hope that know-how can assist us skip over the troublesome steps of growth that different international locations needed to undergo. Nevertheless, as we stand in late 2025, we have to be trustworthy with ourselves. There’s a very massive hole between our goals and our actuality. Whereas we’re busy utilizing Fb and TikTok, we’re not truly constructing the know-how ourselves. We have gotten a nation of digital shoppers, not creators.

This distinction is essential. If we don’t change our path, Nepal dangers changing into a “knowledge colony.” This implies we’ll maintain importing AI instruments from massive overseas corporations, paying them with our cash and our private knowledge, whereas our personal engineers depart the nation. To cease this from occurring, we have to look previous the hype and repair three deep issues: our lack of information, our lack of computing energy, and the lack of our greatest expertise.

The primary main barrier is our language. Within the international tech world, knowledge is commonly in comparison with oil. If that’s true, Nepal has the crude oil as now we have hundreds of thousands of individuals talking and writing, however we would not have the refineries to make it helpful. The large AI fashions like GPT-4 are largely educated on English. They don’t perceive the Nepali language properly. I’ve seen this downside firsthand. After we attempt to use these international fashions for Nepali, they wrestle. They don’t perceive our tradition or the respectful methods we communicate to elders (like utilizing hajur or tapai). Even worse, they typically “hallucinate,” which implies they confidently make up info which can be fully fallacious. We can not use such unreliable programs in our banks or hospitals.

There may be additionally a hidden price that many individuals don’t speak about, which we are able to name the “Tokenization Tax.” The pc applications that learn textual content are constructed for English. After they learn Nepali script, they’re very inefficient. A single Nepali phrase is likely to be damaged into three or 4 items by the pc. Since AI corporations cost cash based mostly on these items (tokens), it prices a Nepali developer 3 times extra to construct an app than it prices an American developer. That is an unfair financial burden on our native startups.

Even when we clear up the information downside, we face a second hurdle: now we have nowhere to run these highly effective applications. AI requires huge computing energy, which may be very costly. Nepal doesn’t have its personal business knowledge facilities with the required high-speed chips (GPUs).

Due to this, our startups are pressured to hire computer systems from overseas giants like Amazon (AWS) or Google. This creates a severe financial downside. Our corporations earn in Nepali Rupees, however they should pay their server payments in US {dollars}. Each time the greenback will get stronger, our companies lose cash. We’re sending our scarce overseas foreign money in a foreign country simply to maintain our servers operating.

That is ironic as a result of Nepal has a surplus of electrical energy from our hydropower initiatives. We speak about promoting electrical energy to India whereas we must be utilizing it right here. We may construct “Inexperienced Compute” facilities, knowledge facilities powered by clear vitality. This could appeal to worldwide corporations who need to scale back their carbon footprint. However to do that, we’d like a dependable energy grid that by no means shuts down and higher web connections to the skin world. Proper now, we’re lacking this chance.

Maybe the saddest problem is the human one. Nepal produces 1000’s of engineering graduates yearly. They’re vibrant and wanting to be taught. Nevertheless, our business suffers from what I name the “Hole Center.” We’ve got many junior engineers who’re simply beginning, and now we have just a few bosses on the prime. However we’re lacking the folks within the center, the seniors with 5 – 6 years of expertise who can lead initiatives and educate the juniors. Why? As a result of all of them depart. As quickly as an engineer will get good expertise, they migrate to the US, Europe, or Australia for higher pay and a greater life.

This “mind drain” is killing our capacity to innovate. You can’t construct complicated, long-term initiatives in case your greatest staff members depart each two years. We’re left with a workforce that is aware of the speculation of AI from college however lacks the sensible expertise to construct actual merchandise.

The federal government has lastly realized these points. The brand new Nationwide Synthetic Intelligence Coverage 2082 is an efficient begin. It acknowledges the risks of AI, like deepfakes, and units massive targets for the long run. However a coverage is only a piece of paper if it isn’t acted upon.

The most important sensible downside is how the federal government buys issues. Our public procurement legal guidelines are designed to construct roads and bridges, the place you at all times select the bottom bidder. This doesn’t work for software program. If the federal government buys the most cost effective AI system, it would get a nasty system. We’d like new guidelines that permit the federal government to purchase high quality know-how from native startups, even when it prices just a little extra.

Regardless of all these challenges, I stay hopeful due to our entrepreneurs. Our startups are adapting. As a substitute of making an attempt to construct huge fashions like ChatGPT, which prices hundreds of thousands, they’re constructing good, particular instruments. They’re utilizing a way referred to as Retrieval-Augmented Era (RAG), which connects AI to native paperwork like Nepal’s legal guidelines or banking guidelines. This makes the AI correct and helpful for our particular wants.

We’re at a crossroads. We will proceed as we’re, watching the AI revolution from the sidelines and paying overseas corporations for his or her instruments. Or we are able to determine to take management. To take management, the federal government should cease simply writing insurance policies and begin investing in native knowledge facilities. We have to create an setting the place senior engineers need to keep in Nepal. And our companies have to work collectively to share knowledge and construct programs that clear up Nepali issues. The know-how is international, however the options have to be native. If we don’t construct our personal AI future, another person will construct it for us, and we’d not just like the outcome.