Will Artificial Intelligence enable the net zero transition or derail it?

AI makes decisions based on parameters. More parameters create more powerful AI models, which require higher computing power and consume more energy. 

Some predictions see AI parameter growth as exponential. In theory, this means very rapid increases in the capability of AI, accompanied by a similar surge in energy demand.

The International Energy Agency predicts that globally AI electricity consumption could be 10 times higher in 2026 compared to 2022. As a result, data centre electricity consumption could exceed 1000 TWh in 2026, roughly equivalent to that of Japan, or 3.6% of current global electricity demand.

Providing this much additional electricity without using fossil fuels will become increasingly challenging the longer AI growth at this rate continues.

Under an exponential growth scenario, constrained energy grid capacity would be allocated to AI, starving manufacturing and the built environment of the additional electricity they need to decarbonise.

Even if exponential growth in AI computing power is theoretically possible, it will be limited by supply of hardware and energy. There simply aren’t enough microchips or electricity to support exponential growth for long.

However, prioritising AI growth at the fastest rate possible may be an attractive option for many nations, who see it as a shortcut to faster economic growth or to give them an edge against geopolitical competitors.

Under a scenario where fast AI growth is prioritised, its demand for energy could derail the transition to net zero. 

Technological limitations, moral concerns, or regulations could all prevent AI computing power and energy demand from continuing to grow at its current pace.

Under this scenario, AI is not prioritised over all other energy uses. Energy is distributed to AI in a managed way that allows other uses critical to the net zero transition to increase their electricity consumption as well. Meanwhile, AI increases efficiency, leading to lower energy consumption across the economy.

Under this scenario, the benefits of AI are enough to outweigh the increased pressure it puts on the energy grid and accelerate the transition to achieving net zero emissions.

AI is already being utilised to increase renewable energy generation and improve operational efficiencies. The extent to which this could increase the supply of low carbon energy is unknown, but the potential of developments critical to decarbonisation, such as smart grids, can be greatly enhanced by AI.

Arguably, without AI, complete grid decarbonisation is unlikely, especially because of the intermittency of renewables. To keep energy supply flowing when it’s not windy or sunny, we will need to distribute stored energy (for example, from batteries and EVs) and integrate, as of yet, undeployed energy sources such as small modular nuclear reactors, like those being developed by Rolls Royce. AI is likely to accelerate these innovations.

In the long-term, AI computing power could even become decoupled from energy consumption with developments in quantum computing. This could allow us to benefit from the efficiencies AI brings without such additional pressure on energy supply. However, quantum computing remains in its infancy and reduced energy consumption is not proven. If this is a fix, it’s not a short term one.