As the profile of artificial intelligence (AI) on the world stage builds, it is becoming increasingly clear that AI is no longer the domain of science fiction – this technology is having a growing impact on our everyday lives. But, while AI has the potential to have profound and positive impacts on global productivity and economic growth, there are a number of potential risks that need to be considered, one of which is the impact on energy consumption.
Over the last two decades, the number of data centres has grown to support increased cloud storage and digitisation. To date, significant improvements in energy efficiency have helped limit the energy demand from the growth in data centres. However, as AI develops, it brings with it potentially larger demands for compute. For example, a typical ChatGPT query can consume as much as 10 times more energy than a typical Google search query.1
Some estimates already put the contribution of cloud computing and data centres at 2-3% of total global greenhouse gas emissions, which is on par with airlines and shipping2. There are few studies that examine the specific influence of AI on carbon emissions.3 However, given the forecasted trajectory of data centre electricity demand resulting from AI, without improvements in the carbon intensity of energy sources and the energy efficiency of data centres, the result could be a significant increase in carbon emissions.
Aside from environmental targets, there is also a strong incentive to reduce power intensity to help lower costs, with power accounting for approximately 15% of a typical data centre’s cost of goods sold4.
This paper explores three key areas that could play a role in the decarbonisation of data centres.
Three areas to consider:
1. Carbon-free energy
The energy source used to power a data centre – typically from the local electricity grid – is the single biggest influence on its carbon emissions. While converting public electricity grids to renewable sources has the greatest decarbonisation potential, the additional generation and transmission capacity required for this to be a solution at scale could take years to build out. What are the near-term and longer-term options?
2. Advanced compute
Computing is the core function of a data centre and represents approximately 40% of total electricity usage5.
The semiconductor industry has a strong track record of innovation and is increasingly focused on chip efficiency. These efficiency improvements have been largely responsible for keeping data centre power consumption relatively flat over the past ten years (1-2% global electricity demand) despite nearly tripling total data workloads. And there are no signs that the pace of innovation is slowing. What energy and cost savings could the next generation of chips deliver?
3. Cooling
As processors become more powerful, they get hotter. Since heat can decrease the efficiency of processors, eventually leading to failure, cooling within data centres is essential. As cooling comprises roughly 40% of a typical data centre’s energy consumption6, it represents a significant area of development to improve energy efficiency. Currently, most data centres use traditional air cooling, but a promising development that could improve efficiency is liquid cooling technology.
Whether efficiency gains from potential solutions will be enough to offset the soaring growth in data centre capacity will depend on the pace of technological development and policy/regulation. Regardless, the carbon impact of data centres could represent material investment considerations for companies providing the solutions as well as the major data centre operators/users.
1. Source: Goldman Sachs, 14 May 2024, AI poised to drive 160% increase in power demand.
2. Source: Climatiq, 21 April 2022, Measuring greenhouse gas emissions in data centres.
3. Source: nature.com, 12 August 2020, The carbon impact of artificial intelligence.
4. Source: Capital Group estimates, as at June 2024.
5. Source: Open Data Institute, Data centres cloud infrastructures and the tangibility of internet power, as at 16 January 2023.
6. Source: McKinsey, 17 January 2023, Investing in the rising data center economy.