Nepal as an AI power bank

High tech firms are power hungry, and the climate crisis is driving them to seek renewable sources of energy. Nepal’s surplus electricity generation capacity coincides with this surge in global demand to power generative AI data centres.

Data centres have become major drivers of electricity demand globally in the last two years, but the lack of clean energy remains a major hurdle to expansion.

Along with coverage area and number of racks, a key parameter to measure use of information technology in each country is the amount of power they use1 for data centres. For example, the US currently uses 50MW of data centres per million people compared to just 1MW in India. In Nepal, investment in data centres have just begun.

About 1.5% of global electricity is currently going to power data centres and data transmission. Improvements in IT hardware, efficient cooling methods, and gains from the shift to efficient cloud and hyperscale data centres have allowed growth in energy demand to remain small even as data centre services have grown significantly since 2010.

After 2022 this picture has begun to break down due to the exponential growth in power requirements for AI which far exceeds accompanying gains in energy efficiency.

The growth of AI applications was made possible by breakthroughs in Deep Neural Network models and hardware to run them, enabling image recognition and generation, natural language processing and many other previously ‘high-compute applications’ which required a lot of computational power.

These advances laid the groundwork for large language models such as GPT-4o (OpenAI), Claude (Anthropic), Gemini (Google), Llama-3 (Meta/Facebook), Grok (xAI) among others. The growth of large language models and generative AI has changed the role of data centres from managing content, data storage and data processing to training large AI models and running AI inference to serve user applications from those models.

But training AI models is compute-intensive, and requires high-powered hardware. Similarly, AI inference, which involves using trained models to make predictions based on new data, also demands significant energy due to the large and increasing number of user applications served by AI models.

According to Bloomberg Intelligence, growth in investment into generative AI is expected to push data-centre electricity use by 4-10 times from current levels by 2030. Companies such as Google and Microsoft are already consuming as much electricity as The Netherlands.

Many of the largest companies have also made ‘net-zero’ commitments, to generate no net greenhouse gas emissions by 2050, which they are struggling to uphold due to this unexpected increase in energy demand to power data centres.Big tech is therefore sourcing power from all the clean energy sources they can find: wind, solar and hydropower. They have even invested in nuclear power plants to power AI data centres. As demand rises, companies will be looking for new geographies with abundant clean energy.

Could this new reality present an opportunity for Nepal to market its surplus clean energy from hydropower and solar? The country could promote itself as a destination for data centres, as Bhutan is doing with its Gelephu Mindfulness City.

Nepal’s data infrastructure, even including NEA and Ncell Axiata data centres, is not well developed. The recently announced partnership between the Chaudhary Group’s BLC Holdings, and India’s Yotta Data Services to build Nepal’s first supercloud 4MW data centre in Ramkot could pave the way for private sector investment into data centres in Nepal.

Yotta already runs some of the largest hyperscale data centres in Mumbai and Delhi, and this investment could allow the company to be the provider of digital services in the Subcontinent’s high growth IT sector.

To catch up with per capita IT services at the level of India, Nepal would need seven more Ramkot-sized data centres using about 30MW of power. With its readily available clean energy and proximity to large population centres in Northern India and Bangladesh, Nepal could in theory provide the compute infrastructure necessary to power the future AI training and inference services for local applications and markets.

To provide IT services at the level that the US population currently consumes, Nepal would need 1,500MW of power just to power data centres for its own needs. While this is unlikely to happen anytime soon, Nepal could opportunistically provide data centre services for the upcoming AI needs of some of the 500 million people living in the Ganges basin.

India is already a regional powerhouse in data centres and has about 1,600MW of capacity, and generative AI is estimated to push computational requirements with power needs expected to grow between 25-30% each year through to 2030.

With the right investment climate, and the necessary investment into fibre optic infrastructure and IT parks, Nepal could leverage its plentiful power supply advantage to offer locations for a portion of this investment.

This would provide a much needed boost to Nepal’s ongoing efforts to establish itself as a technology hub and provide jobs to the 15,000 IT professionals that are being trained in the country each year.

Data centres are one part of a larger technology ecosystem. AI model training and maintenance require skilled expertise in every industry from finance, to train fraud detection models, to healthcare, to label X-ray and MRI images and test results and diagnostic data to build intelligent medical tools, and to the legal sector, to supervise generated contract documents.

NTC’s recently installed XGPON network standard that can deliver up to 10 Gbps downstream speed and 2.5 Gbps upstream does provide the basic level of connectivity needed for AI applications, but it is only available in Kathmandu and needs to be expanded to other urban centres.

Nepal’s investment climate for private companies remains far from ideal, and government regulations need to be relaxed to allow global companies and non-resident Nepalis to bring in investment to take full advantage of this new digital infrastructure.

Bikash Pandey is Director of Clean Energy and Circular Economy at Winrock International. People Power is his regular column in Nepali Times on global energy issues relevant to Nepal.