The two opposing impacts of AI on energy transition
By Jeff Lin, Thematic Equity Investment Portfolio Manager at M&G
The increased adoption of Artificial Intelligence (AI) will require additional physical datacentre infrastructure, which includes buildings and associated power and cooling, and hence increased electricity consumption. This is likely to create further demand for renewable energy as we move ahead.
It is likely that we are only at the initial stage of such growth. The massive increase in data centre products from enablers such as Nvidia is just starting to be commercialised and is growing at tremendous speed. Nvidia’s datacentre quarterly revenue run rate has quadrupled in the last 12 months, as the company sees strong demand across hyperscale public cloud companies (such as Amazon, Microsoft and Google), consumer internet and corporate enterprise[1].
Microsoft believes that in its fiscal year (FY) 2022 ending June, it consumed 18,153,454 MWh of electricity, up 33% versus FY 2021[2]. While the company has not yet disclosed its electricity consumption for FY 2023, we do believe the growth in Generative AI computing will continue to accelerate the rise in electricity consumption.
Efficiency improvements from AI offset AI energy consumption
While we expect AI to be a significant driver of electricity consumption over the next decades, we believe it will also play a key role in managing the supply and demand of electricity. As electricity generation moves to renewable sources from carbon-based ones, the supply of electricity will become more fragmented, distributed and less predictable. Homes are likely to become local sources of solar energy, power may come from long distances where there is wind, sun, or hydroelectric power, and storage of excess power will require stationary or mobile (including EVs) batteries.
However, wind and solar energy is not always available, either because of the time of day or the weather. AI will be needed to balance electricity consumption with production. For example, based on weather predictions, and hence available power, non-critical electrical loads can be “shifted” to more favourable times. AI can recommend reducing power consumption and storing excess power in batteries ahead of adverse weather that could reduce available power. Companies such as Schneider Electric and Microsoft are working together to transform grid management with the goal of maintaining grid reliability and accelerating customer adoption of distributed energy storage resources, including EVs and rooftop solar[3].
AI can also be used to reduce energy consumption. Today, data from trucks is analysed and used to optimise drivers’ habits. Data such as idle time, throttle position, and speed is being used to train drivers to minimise fuel consumption. In the future, autonomous transportation with fully electric vehicles will help minimise energy consumption because of optimised routing, reduced traffic, and better asset utilisation.
While AI will increase datacentre energy consumption, the capabilities of AI will be needed to manage the transition to fully renewable electricity sources.
1 Source: Nvidia, company earnings reports, 2023
2 Microsoft, 2022 Environmental Sustainability Report, 2022 Factsheet
3 Source: Schneider Electric capital markets day