Image: Victoria Serghievici
Victoria Serghievici from Coherent Solutions discusses the growing energy demands of AI and how to mitigate them.
The rise of artificial intelligence (AI) applications and the quest for artificial general intelligence (AGI) demands increased energy resources on a major scale. According to the International Energy Agency, data centres feeding AI with massive amounts of data are projected to consume about 3-4pc of the world’s electricity by 2026.
Let us put this into perspective. The global electricity supply was around 29,734 TWh in 2023, and if data centres soon consume about 1,189 TWh (4pc) annually, this could exceed four times the total electricity consumed in a year in the UK. A large proportion of the energy used to generate electricity comes from fossil fuels, which has serious environmental impacts.
As AI advancements require extensive energy, “there is no way to get there without a breakthrough,” said OpenAI CEO Sam Altman, referring to the need for more and cleaner energy to meet AI-driven demands. The solution is to focus on renewables – primarily solar and potentially the energy of nuclear fusion, which is currently uncertain – and to adopt practical methods to reduce emissions.
Businesses need effective strategies to decarbonise
The world added 50pc more renewable capacity in 2023 compared to the previous year. Playing in unison, the AI industry broadly calls for balancing technological advancement with sustainable practices. Organisations that develop and leverage AI solutions prioritise green initiatives today for a more sustainable future and to enable AI’s expansion without growing its carbon footprint.
Energy providers today are increasingly developing farms that aim to provide 100pc clean, renewable energy from wind and solar sources. They empower households and businesses to make a positive environmental impact by switching from fossil fuels to sustainable energy, helping to reduce carbon footprints and contribute to a cleaner future.
There are some key practices that businesses of all sizes can adopt to significantly decrease their ecological toll. Together, these practices can help organisations steer their ventures forward in line with global environmental goals.
Use energy-efficient AI hardware
Keeping focused on sustainability and reducing energy consumption requires the right tools to measure and manage environmental impact. Accurate data and insights are indispensable to make informed decisions.
Most businesses that are not running their own data centres can use public cloud providers, such as AWS and Microsoft to run their AI workloads. Also, advanced tools such as AWS Carbon Footprint can help measure a business’s estimated carbon footprint and develop sustainability goals.
Adopt green data centres
Data centres have a significant environmental footprint due to their high energy requirements. To mitigate this impact, large-scale businesses that use on-premises data powerhouses can leverage renewable energy sources such as solar, wind and hydro.
Businesses can further reduce energy consumption by using liquid cooling and free cooling techniques instead of energy-intensive chillers. Machine learning algorithms can help refine cooling systems, workload distribution and minimise energy waste to make data centres more sustainable.
Dynamically adjust computing power
Modern servers and storage devices are designed to consume less power, and businesses can pair them with software that dynamically adjusts computing power based on real-time demand. For example, Kubernetes and Apache Hadoop YARN automatically scale resources for large-scale data processing, while cloud tools such as AWS Auto Scaling and Azure Automation optimise resource allocation.
Using AI itself can contributes to carbon reduction. AI helps to identify areas where emissions can be reduced, streamline logistics and improve resource usage. It is crucial to keep machine learning models accurate and efficient to optimise computational resources and make sure AI operations run smoothly without consuming excessive energy.
The promise of nuclear fusion
Powerful large language models (LLMs), data analysis modes and advanced AI applications will require lots of energy, which is currently challenging to estimate. For example, training a single model such as GPT-4 can produce approximately 300 tons of CO2.
As a theoretical clean energy solution, nuclear fusion technology could generate nearly limitless power without producing long-lived radioactive waste. Nuclear fusion is still very much a research domain, far from practical applications, although breakthroughs have been made. The team from Princeton University’s Plasma Control Group used AI that helped stabilise plasma – a crucial step to maintaining safer and more efficient fusion reactions. Looking ahead, this development could potentially make it feasible to use fusion energy as a practical power source.
Empowering change to reduce emissions
As AI advances, so does its energy consumption. Businesses can make strides in reducing their AI carbon footprint by adopting a multifaceted approach. The key to change is fostering a conscious culture and promoting AI development with sustainability in mind. This strategic decision is an ethical choice for the planet that also helps strengthen brand reputation and customer loyalty.
Victoria Serghievici is a machine learning engineer at Coherent Solutions. She develops scalable models to enhance operational efficiency and support energy providers in making a cleaner world. She is passionate about continuous learning and keeps up with the latest ML and AI advancements.
Find out how emerging tech trends are transforming tomorrow with our new podcast, Future Human: The Series. Listen now on Spotify, on Apple or wherever you get your podcasts.
