Feeding energy-hungry AI tech advances

The growing popularity of artificial intelligence (AI) technology as a powerful new tool is raising questions of how the world can supply sufficient energy to resource-hungry AI while countries are shifting to clean energy.
AI encompasses advanced computer systems able to do tasks in place of human intelligence. It can accelerate work speed, but at the same time requires huge amounts of electricity.
Data centres and cloud computing, which offer computing services including data storage, analysis and management, support AI.
“AI and Google search can do similar work. While we cannot clearly differentiate some of their functions, the former demands much more power,” said Chaiwat Kovavisarach, group chief executive officer and president of energy conglomerate Bangchak Corporation Plc.
He was speaking at the 14th Greenovative Forum, with the theme “Crafting Tomorrow’s Future with Sustainable Energy and AI.” The event, held recently by Bangchak, was joined by 600 participants, including energy gurus, AI buffs and environmental experts.
“Bangchak has a high expectation of AI. We are eager to know how AI can help us work better,” said Mr Chaiwat, adding he is also hopeful about developing AI into an energy efficient technology.
Energy demand and supply are a big issue as the world is depending more on AI. “We have to strike a balance between its benefits and impact, including fewer jobs for workers as they are replaced by AI,” said Prasong Poontaneat, acting chairman of Bangchak’s board of directors.
One major challenge for energy producers is how to add more electricity to the world to power AI, used increasingly by businesses and households, amid global campaigns for a transition from traditional energy sources to renewables.
The question arises of whether renewable energy sources will be sufficient for AI-driven demand for electricity. “The world has never had an energy transition. What we have seen so far is energy addition, which means we keep adding new energy sources,” said Mr Chaiwat.
Humans long used biomass like wood which is burned to produce energy before they discovered coal. Oil, natural gas, nuclear energy and renewable energy were discovered later, only to diversify energy sources. “Today many people still burn wood as a fuel,” said Mr Chaiwat.
Based on various sources of energy, the world can supply 0.59 zettajoules (ZJ) at present. The world needs to add 0.31 ZJ to reach a total of 1 ZJ to meet demand for more AI usage in the future, he said. ZJ is a large unit of energy amounting to billion trillion joules.
Mr Chaiwat said a data centre needs 5 megawatts of electricity. The facility can be upgraded to what he called “Data Centre 2.0,” which utilises the cloud computing technology, to better manage more data. This increases power demand to 30-100MWs.
When AI is applied, a Data Centre 3.0 is needed, demanding 300-1,000MWs. The power demand of a large data centre could reach 5 gigawatts in the future.
“That amount is equal to 100 Bangchak refineries and the data centre compound will span 2 million square metres,” said Mr Chaiwat.
If developers cannot come up with new technology to build a data centre that demands less electricity, or energy conservation campaigns do not work, nuclear energy technology may be needed to supply more power, said Mr Chaiwat.
A data centre, or network of computers that stores, processes and distributes data, is still based on binary coding technology, or the use of computer code with two digits (0 and 1) as a basis for computing and doing a variety of tasks.
Experts are wondering whether they can develop “DNA data storage” technology which imitates the way human DNA works, said Mr Chaiwat. By translating binary code of 0s and 1s into a sequence of DNA nucleotides (A,T,C,G), DNA data storage can store and mange digital data. This technology, currently in the early stage of development, promises an energy efficient feature, but it is costly and prone to technical errors.
“If we can change to DNA data storage, we have to see whether it can help us save the world. We want the world to keep developing and, at the same time, need to take a good care of the planet,” said Mr Chaiwat.
A new better machine that can run AI and requires less electricity is needed. “The current version of the machine is like an oil-powered car we used a long time ago that could run for just 4 kilometres per a litre of oil,” said Mr Chaiwat.
If DNA data storage seems to be a futuristic technology, another option is the use of a small modular reactor (SMR) to supply electricity to the world during the AI era, he suggested.
SMR is a form of nuclear power technology with a capacity of up to 300 megawatts per unit, about one-third of the generating capacity of larger, traditional nuclear power reactors, according to the International Atomic Energy Agency.
The world cannot depend only on solar and wind power because the sun and wind tend to offer an intermittent supply of power, with volume determined by weather patterns. Under Thailand’s 2024 power development plan, to come into effect from 2024 to 2037, two SMRs, each with a capacity of 300MW, will be developed and start operations towards the end of the plan. The SMRs could be developed and run by the Electricity Generating Authority of Thailand.
Though SMRs are designed to be safer than traditional nuclear power plants, safety issues are important, such as the need to prevent radiation leakage that could affect humans and the environment, said Mr Chaiwat. “Bangchak is open-minded about energy development, including the use of nuclear energy technology,” he said, adding the company is interested in conducting a study on nuclear energy.
Mr Chaiwat also suggested a need for people to change behaviour to reduce electricity consumption. Demand for electricity is not only driven by AI but also some lifestyle choices like overusing air conditioners to keep cool. “It doesn’t make sense to turn on air con when people in a room wear jackets to keep themselves warm,” he said.