Data centres have evolved from the backrooms and basements of IT departments to become the critical infrastructure that drives business operations, innovation and the development of new technologies. They are cornerstones of digital economy. But this growing importance has come at a cost.
Data centre energy expenditure is rising as our dependence on them is increasing. The International Energy Agency says data centres account for 1% of all global energy consumption – more than the United Kingdom and on par with all of France. Research from Schneider Electric anticipates that data centre power demands will almost quadruple by the middle of the 21st century.
Over the last 15 years, the volume of data that has been created, captured, copied, and consumed worldwide has grown from just 2 petabytes in 2010 to about 150 petabytes with the expansion of digital tools permeating almost every aspect of work and leisure. And that trend will continue with the growth in AI adoption – an application that is data and energy hungry.
However, despite this, carbon emissions from data centres remain stable. Data centre operators are at the forefront of boosting energy efficiency. As the thirst for data and processing capability increases, largely driven by new and emerging AI applications and services, data centre operators are simultaneously leveraging AI to better understand and manage their energy use and reduce carbon emissions.
The data centre industry is an essential service that delivers critical societal benefits and recognises that the costs must not outweigh those benefits. It actively supports decarbonisation and the need to continually strive for greater energy efficiency and reduce carbon emissions.
Cloud to edge computing enables the use of sophisticated tools supporting carbon footprint reduction in every area of human activity. Industries such as transport, manufacturing and power generation rely on centralised processing of data to support better decisions and processing at the edge. Devices such as cars, smartphones, and smart home devices can leverage AI to support individual users’ actions towards a net-zero emissions goal.
There are many applications for AI to support data centre operators. AI can be used to choose optimal data centre locations. By understanding how and where data and processing capacity is used, data centres can be located where cooling– which can account for more than half of a data centre’s operating costs – can be provided at a lower cost.
Furthermore, usage data can be used to optimise resource management and to monitor equipment so that preventative action can be taken before a device suffers a failure that leads to increased energy use and repair costs.
Analytics can be for predictive maintenance and to detect issues and dispatch maintenance technicians promptly to the right place without trial/error fault detection and remediation.
Recently observed growth of AI models leads towards a clear goal – the introduction of behavioural evolution and making Green AI a new standard.
Algorithms that are environmentally friendly will solve problems with access to large amounts of data. This can be challenging because code may not yet be optimised in embedded systems or is constrained by limited resources.
Green AI can be achieved in several ways. Strategic selection of data centre location, where machine training takes place, may have a strong impact on the overall carbon footprint of the project. This strategy can benefit mostly time- and latency-insensitive, large workloads. Simpler algorithms can minimise the amount of machine learning models through quantization and knowledge distillation. Models can be partitioned into discrete regions with fewer features.
To read more about the recent findings on AI in ANZ companies, click here