How should investors view the trade-offs between AI’s rising consumption of resources and its positive applications? Here we explore AI’s role in enabling climate solutions and how we should approach the challenge of its negative impact on the environment.
- AI is already proving invaluable in enabling climate solutions, for example in energy efficiency, environmental technologies and forecasting.
- However, the significant computational resources required by AI has led to a dramatic increase in energy and water consumption.
- These conflicting risks and opportunities are intertwined with the broader need to decarbonise and managing them effectively will require policies that support the scaling up of renewable energy.
- Future applications could include using AI to integrate renewable energy sources into the grid, the development of multimodal AI, and the use of more efficient language models for specialised tasks
For many investors interested in the transition to a more sustainable economy, their excitement regarding the potential of artificial intelligence (AI) is heavily tempered by a concern about its negative impact on the planet. AI’s rapid growth has come at a high cost in terms of increased electricity and water consumption, which has had a knock-on effect for global aspirations to conserve the earth’s resources. But at the same time AI is also making a significant contribution to the knowledge and tools needed to substantially reduce pollution and waste.
From new materials and technologies to better forecasting, we envision a future where the trade-off between AI’s positive applications and its rising consumption of power creates opportunities for investors, given the necessary support from policy makers.
The scale and scope of the AI revolution
The AI revolution is driven by three key factors: the exponential increase in data availability, advancements in computing power, and recent scientific breakthroughs, particularly in machine learning and AI model architectures.
The explosion of digital data from various sources (social media, the internet of things, and business transactions to name a few) has provided the raw material needed to train complex AI models. According to McKinsey, the amount of digital data in the world doubles every three years, and it’s estimated to total 175 zettabytes by 2025. 1,2
Advances in graphics processing units (GPUs), technology, cloud computing, and specialized AI hardware have made it possible to train large-scale models that were previously unimaginable. GPUs, for example, are becoming ever more efficient and less costly – a 2022 GPU was able to manage approximately 64 times as many calculations per second per dollar as a 2008 model.
Equally, innovations in AI, such as transformers and Large Language Models (LLMs), have drastically enhanced its ability to understand and generate realistic human-like content. At the same time, the rise of reinforcement learning, as demonstrated by DeepMind’s AlphaGo (which outperformed human champions in the complex game of Go) highlight AI’s growing potential across industries such as online education and hiring platforms.
However, significant computational resources required for these developments mean a dramatic increase in energy consumption: a search using AI takes 10 times the computing power, and therefore electricity, as a conventional internet search.3 Add to that the water required to cool the stacks that do those computations, and the environmental toll is high.4
AI’s role in enabling climate solutions
Although as a society we have only scratched the surface of potential use cases for AI, it is already proving invaluable in enabling climate solutions. Below we focus on three examples: energy efficiency, environmental technologies and forecasting. For the first two, AI can help reduce emissions and resource use, while for the latter it can contribute to climate adaptation and resilience, an important subset of AI within our sustainable economy taxonomies.
Energy efficiency
Building energy use is determined by a constantly changing thermal flow, dictated by occupancy and weather. Understanding these dynamics is crucial to the efficient operating of heating, ventilation and air conditioning (HVAC) systems, however this is difficult to observe directly. AI tools can help to simulate how building occupancy, structure, design and the weather interact to affect thermal flow, but also to predict how it may change. For example, the enormous volumes of data generated by networks of connected sensors feed Schneider Electric’s real-world AI solutions. The company claims it has enabled a 15% reduction in electricity use and a 4% reduction in heating energy consumption across 600 school buildings.5
In manufacturing, AI is already used in robotics and predictive maintenance software to reduce operational downtime, maximising resource efficiency. Siemens uses smart monitoring to track systemic indicators, like vibration behaviour and temperature, and offer faster reaction times for manufacturers of products ranging from recycled steel to Lithium-ion batteries.6
Meanwhile, in transport, AI is being used to optimise routing and operate autonomous vehicles. UPS’s Orion system is one example. It uses AI and machine-learning techniques to reduce miles and emissions, enabling its fleet to save approximately 10 million gallons of fuel each year.7
Environmental technologies
AI is also helping to develop more efficient clean technologies, or to reduce the impact of existing technologies.
In renewable energy generation, Vestas has used AI to understand and harness ‘wake steering’ – the process of turning turbine rotors away from oncoming wind to deflect wakes caused by nearby turbines.8
US software company Autodesk worked with Airbus to apply its generative design technology to develop lighter parts for aircraft manufacturing.9 And Google Deepmind has been used to predict structures of new materials, for use in batteries and semiconductors: it claims to have discovered 380,000 stable materials that could power future energy technologies.10
Forecasting
AI is proving particularly powerful in forecasting the behaviour of complex systems. In weather forecasting, AI models have proved more able than supercomputers to accurately predict weather events, and in a fraction of the time.11 Such forecasting has applications across the economy, but particularly in improving agricultural productivity.
It also promises to improve the prediction of extreme weather events, giving advanced warning that can save lives. AI models can also help better predict the long-term impacts of climate change, and so guide decisions on adaptation-related investments. The applications in identifying and mitigating asset-specific risks in insurance and reinsurance are nearly endless.12
Meeting the resource efficiency challenge
Against the benefits promised by AI must be weighed its environmental impacts – specifically the carbon emissions associated with its energy use and the water required for fabrication and cooling.
The International Energy Agency projects that data centre energy consumption is set to double between 2022 and 2026 to 1,000 TWh, roughly equivalent to the entire energy consumption of Japan.13 Morgan Stanley estimates that demand from generative AI will grow at an annual average of 70% to 2027, by which point it will account for approximately 25% of all data centre energy consumption.14
Source: International Energy Agency, 2024
The bar chart shows the estimated electricity consumption by data centres (in Terawatt-hours) by region for 2022, along with the projected consumption for 2026.
Regions include the United States, the European Union and China.
Overall, the chart shows estimates for significant growth in consumption for the period, based on compiled reported and projected numbers from each region.
Presentation
The bar chart shows the estimated electricity consumption by data centres by region for 2022 and projected consumption for 2026.
Source
International Energy Agency
Energy use may rise further with new capabilities and complexities associated with AI models. Forecasts are highly uncertain, however, as improvements in processing and energy efficiency have historically offset increased demand.15
Industry is responding to the resource efficiency challenge by building hyperscale data centres and innovating in chip design. Data centre operators are some of the largest investors in renewable energy: in May, Microsoft signed an agreement with Brookfield Asset Management to develop 10.5 GW of renewables – the largest ever corporate power purchase agreement by a factor of eight.16 Continued innovation in energy efficiency will be critical as AI’s computational demands continue to grow.
On top of its energy consumption, AI is a thirsty business: its data centres need cooling, fabrication sites are water-intensive (often with low recycle rates due to their need for high purity standards or for cost reasons), and many US makers use thermoelectric plants for power, such that the industry’s projected annual water usage could hit 6.6 bn m³ by 2027.17 We expect demand for water-efficient technologies to increase commensurately.18 New cooling technologies, such as liquid cooling or immersion cooling, will be essential to manage the heat generated by dense computing environments, particularly as GPU racks become larger.
Opportunities for economy-wide solutions, with policy support
AI’s positive applications and its rising consumption of power are intertwined with the broader challenge of decarbonising and building out power systems and require policies that support the economy-wide scaling up of renewable energy. For example, policy support is needed to overcome challenges in enabling transmission interconnections. In the US alone, there are nearly 12,000 projects, representing 1,570 GW of generating capacity and 1,030 GW of storage actively seeking interconnection: this capacity is more than double the size of the existing fleet.19
There are also bottlenecks in electrical infrastructure supply chains. Lead times for transformers have increased from 50 weeks in 2021 to 120 weeks in 2024, on average, according to Wood Mackenzie; for large transformers, the wait can be four years.20
In the US in May, the White House launched a federal-state initiative to bolster the US power grid, and the Infrastructure Investment and Jobs Act, signed into law in 2021, provides US$65bn for grid improvements.21, 22 Last November, the EU unveiled a 14-point Action Plan for Grids, designed to help attract the €584 billion of investments needed in the EU’s grid by 2030.23
Policymakers are also mandating greater transparency around data centre resource use and promoting innovation and research into on how AI can address climate change. In the EU, for example, data centres will be required from September 2024 to report key performance indicators, such as around energy use, under revisions to the Energy Efficiency Directive.24 Meanwhile, a recent President Biden Executive Order included the creation of the “National AI Research Resource” – a tool that will provide researchers and students with access to key AI resources and data and expanded grants for AI research including in climate change.25
What may happen next?
We expect several new models and applications will broaden the ways that AI can make businesses more efficient.
AI is increasingly being used to optimize the integration of renewable energy sources into the grid. For instance, AI algorithms can predict energy production from solar and wind sources more accurately, allowing for better load balancing and energy storage management. Research into advanced materials, such as graphene-based cooling systems, could lead to more efficient heat dissipation in data centres, reducing the overall energy footprint of AI operations. As AI and other energy-intensive technologies grow, the existing energy grid will require significant upgrades to handle increased demand. Investments in smart grid technologies, such as AI-powered demand response systems and decentralized energy distribution, will be essential to ensure the grid’s reliability and efficiency in the face of these new challenges.
More broadly, we believe large language models will become more prevalent in real-world applications, ranging from customer service and content creation to more complex tasks in finance, healthcare, and manufacturing.
Emerging models like multimodal AI, which can process and integrate information from different types of data (e.g., text, images, video), will also broaden the technology’s applications, while advances in voice interaction will make AI more accessible and intuitive to use, further integrating it into daily life.
In addition, smaller, more efficient language models can be expected to become more valuable for specialized tasks where large models are not required, or even appropriate. These models can offer more secure cost- and resource-efficient solutions for industries that need to draw from more specific data sets, like healthcare and finance.26
AI will also play an increasingly critical role in robotics, enabling more sophisticated automation in industries like manufacturing, logistics, and healthcare, driving productivity gains and opening new markets.
Conclusion
AI promises to make a significant contribution to developing the materials, technologies and business models we will need to put the global economy on a sustainable footing. It is also likely to help drive innovative resource efficiency, as sector leaders reward chip makers and data centre operators whose products and services use less energy and water.
Ultimately, however, the AI sector will depend upon the decarbonisation of power systems if its relentlessly expanding demand for power is to be met without a major climate impact. It is here that policymakers need to focus their attention, if the benefits of AI are to outweigh its environmental costs.
- McKinsey Global Institute, December 2016: The age of analytics: Competing in a data-driven world ↩︎
- Reinsel, D. et al, November 2018: The Digitization of the World From Edge to Core, IDC. To put that in perspective, one zettabyte is one billion trillion bytes, equivalent to over 12 million 4K videos, or 330 million of the world’s largest hard drive. ↩︎
- Kerr, D., July 2024: AI brings soaring emissions for Google and Microsoft, a major contributor to climate change, NPR ↩︎
- Li, P. Yang, J. et al, October 2023: Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models ↩︎
- Schneider Electric, September 2022: Using AI to optimize HVAC systems in buildings ↩︎
- Siemens, 2024 ↩︎
- UPS, 2020: UPS To Enhance ORION With Continuous Delivery Route Optimization ↩︎
- Choney, S., March 2022: How one of the world’s largest wind companies is using AI to capture more energy, Microsoft ↩︎
- Deplazes, R., November 2019: Autodesk and Airbus Demonstrate the Impact of Generative Design on Making and Building ↩︎
- Merchant, A. and Cubuk, E., November 2023: Millions of new materials discovered with deep learning ↩︎
- World Economic Forum, December 2023: AI can now outperform conventional weather forecasting – in under a minute, too ↩︎
- Genpact, 2020: Insurance in the Age of Instinct ↩︎
- International Energy Agency, January 2024: Electricity 2024: Analysis and forecast to 2026 ↩︎
- Morgan Stanley, March 2024: Powering the AI Revolution ↩︎
- The Economist, Jan 2024: Data centres improved greatly in energy efficiency as they grew massively larger – But can this continue into the age of AI? ↩︎
- Brookfield Renewable Partners, May 2024: Brookfield and Microsoft Collaborating to Deliver Over 10.5 GW of New Renewable Power Capacity Globally ↩︎
- Gordon, C. March 2024: AI Is Accelerating the Loss of Our Scarcest Natural Resource: Water, Forbes ↩︎
- Impax Asset Management, January 2023: Quenching the semiconductor industry’s thirst ↩︎
- Rand, J. et al, April 2024: Queued Up: 2024 Edition Characteristics of Power Plants Seeking Transmission Interconnection As of the End of 2023 ↩︎
- Jacobs, K. et al, April 2024: Supply shortages and an inflexible market give rise to high power transformer lead times, Wood Mackenzie ↩︎
- The White House, May 2024: Federal-State Modern Grid Deployment Initiative ↩︎
- The White House, August 2021: UPDATED FACT SHEET: Bipartisan Infrastructure Investment and Jobs Act ↩︎
- European Commission, November 2023: Commission sets out actions to accelerate the roll-out of electricity grids ↩︎
- European Commission, March 2024: Commission adopts EU-wide scheme for rating sustainability of data centres ↩︎
- The White House, October 2023: FACT SHEET: President Biden Issues Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence ↩︎
- Lee, L., June 2024: Tiny Titans: How Small Language Models Outperform LLMs for Less, Salesforce The 360 Blog