The sustainability scene was recently awash with news that Google had reported a shocking 48% increase in their global carbon emissions over the last five years, in stark contrast to their sustainability goals. In their latest environmental report, the tech giant stated their total 2024 GHG emissions as 14.3 million tCO2e compared to 9.7 million tCO2e in 2019. This surge has been primarily driven by increase in Scope 2 and Scope 3 emissions, highlighting energy consumption in their data centres and supply chain emissions. Though the company reiterated their commitment to achieving Net Zero by 2030, they acknowledged that this would be a challenging feat due to their ever-increasing investment in AI infrastructure.
In a similar vein, Microsoft’s 2024 environmental report also noted a substantial increase in GHG emissions. While their Scope 1 and 2 emissions decreased by 6.3% compared to their 2020 results, Scope 3 emissions increased by over 30%. The company attributed this rise to the construction of more data centres to accommodate their expanded AI research as well as their increased consumption of associated hardware components like servers and semiconductors. This article gives a further breakdown of the difference between Scope 1, 2 and 3 emissions. So, what is the correlation between the growth of AI and climate change and how concerned should we be?
The Energy Demands of AI
The rapid growth of AI is accompanied by an escalating demand for energy to power its operations. This surge in energy consumption poses significant challenges for the global transition towards sustainable AI and net-zero carbon emissions.
Electricity Consumption
According to estimates, AI could consume up to 51% of global electricity by 2030 if electricity efficiency improvements are not achieved. The primary sources driving AI's electricity consumption are consumer devices, communication networks and data centres. Consumer devices like smartphones, laptops, and smart TVs require electricity for operation and charging. Communication networks that enable data transmission consume power for base stations, routers and antennas. However, data centres that store and process vast amounts of data for AI applications are the most energy-intensive, accounting for a significant portion of AI's electricity usage.
Impact on Global Grids
The escalating electricity demand from AI data centres is expected to strain existing power grids worldwide. In the United States alone, data centres could account for 20% of the country's electric demand by 2030, with AI data centres contributing three-quarters of this demand. Similarly, it is estimated that Europe’s power grid will require significant investments, estimated at nearly €800 billion ($861 billion), in transmission and distribution over the coming decade to accommodate the growing energy demands of data centres. These numbers contrast with the existing Information and Communication Technology (ICT) sector, which does not involve AI applications, and accounted for between 5% and 9% of total global electricity consumption in 2018.
The rapid expansion of AI has the potential to revolutionise various industries and sectors, but it also presents a formidable challenge in terms of energy consumption and environmental impact. Addressing this challenge will require concerted efforts from stakeholders, including the development of more energy-efficient AI technologies, increased investment in renewable energy sources and the implementation of sustainable practices throughout the AI lifecycle.
AI Carbon Emissions
The carbon emissions associated with AI operations can be categorised into different scopes. Scope 1 emissions refer to direct emissions from sources owned or controlled by an organisation. Scope 2 emissions are indirect emissions from the generation of purchased electricity, heat or steam. Scope 3 emissions are all other indirect emissions that occur in an organisation's value chain, including e.g., emissions from business travel, employee commuting and the use of sold products.
The training and deployment of AI models require massive computational power, leading to significant energy consumption. For instance, training GPT-3, one of the largest AI models, consumed about 1287 MWh of electricity, resulting in 502 metric tons of CO2 emissions, which is equivalent to the annual carbon footprint of 56 average American homes. Moreover, once these models are trained, the process of inference (where the AI makes predictions or provides responses) can consume even more energy. Google estimates that inference accounts for 60% of the total energy usage in AI.
Increased Water Consumption
Thanks to their powerful computing operations, data centres generate a lot of heat and need equally powerful cooling systems. This puts the local ecosystem and communities where centres are hosted at higher risk of droughts. Large tech companies have implemented initiatives to combat this, ranging from monitoring systems to setting ambitious targets aimed at water neutrality by 2030. Microsoft states, "being net water positive means we will reduce water consumption across our global operations, replenish more water than we use, provide people across the globe with access to water and sanitation services, drive innovation and engage in water policy". Big techs have started replenishing watersheds to offset their cooling water consumption and achieve 'water positive by 2030' for their data centres.
Working Towards Sustainable Solutions
In their environmental reports, these companies have stated their awareness of the impact of AI on the environment. To address these concerns, major tech companies are taking steps to reduce the carbon footprint of their AI operations. Microsoft, for instance, has committed to running its data centres on 100% renewable energy by 2025. Similarly, Google’s data centres already operate on 100% renewable energy, indicating their commitment to sustainable AI operations.
Experts emphasise the need for more energy-efficient AI models and smarter deployment strategies. Running data centres more efficiently and using renewable energy sources are critical steps. Furthermore, researchers suggest that not all AI applications require large, energy-intensive models; smaller, more efficient models can often perform tasks with sufficient accuracy, thus reducing the overall environmental footprint.
The Bottom Line
With the rapid developments and adoption of AI technologies in everyday use, it is clear that AI is here to stay and there is a direct influence between AI and climate change. The dual challenges of AI's escalating energy demands and its substantial environmental footprint, ranging from voracious energy consumption to considerable carbon emissions, highlight the urgent need for sustainable AI practices. By spotlighting these critical issues, we underscore the necessity for collective action towards developing and implementing sustainable AI systems, alongside broader investment in renewable energy sources and energy-efficient practices within the tech industry.
Businesses that incorporate AI in their operations can still achieve Net Zero with a well-thought-out strategy that considers their emissions across Scopes 1, 2 and 3, supply chain output and future business strategy.
