Market Views: How sustainable is AI's escalating energy demand?
Artificial intelligence (AI) has become a huge buzzword around the world.
From finance and healthcare to construction and education, the impact of artificial intelligence is being felt across industries.
Yet this rapid growth comes with its own set of challenges -- the level of energy that AI consumes.
The AI ecosystem's substantial energy consumption includes everything from hardware to training protocols and operational techniques.
The World Economic Forum highlights the startling pace at which AI’s energy demands are escalating, with the computational power required doubling roughly every 100 days.
That poses a challenge for a world also trying to move away from fossil fuels and reduce its carbon footprint.
By 2028, AI could potentially consume more electricity than Iceland did in all of 2021, highlighting a critical need for sustainable practices, according to the WEF.
As investors and stakeholders increasingly prioritise sustainability, there's a growing demand for AI development that aligns with these values.
We asked asset managers and industry experts to assess the sustainability of AI's current development in terms of energy use and to share any strategies their firms are implementing to mitigate environmental impacts.
The following responses have been edited for brevity and clarity.
Tomomi Shimada, lead APAC sustainable investing strategist,
JP Morgan Asset Management
While AI can help resolve sustainability issues and improve efficiency across businesses by using AI technologies, we need to be aware of the potential unintended consequences, particularly the high energy demand.
Generative AI tools uses 6 to 10 times more energy than traditional search engines and this will multiply as the tool goes through version upgrades.
Current estimates show that the power demand of data centres will increase by more than 150% by the end the decade, putting pressure on energy supply.
We believe investments into better infrastructure, smarter energy grids, and, most importantly, renewables are key to ensuring the sustainable development of AI.
Smarter grids provide multi-directional power flow, and the technology helps match the supply and demand of electricity in real-time.
For example, it can enhance energy supply by forecasting weather patterns, predicting solar and wind patterns and allowing for better management of supply and usage.
Improved infrastructure, such as smarter wind turbines, coupled with these predictions, can actively and automatically adjust where wind forces come, capturing as much wind as possible to maximise energy efficiency.
We see abundant investment opportunities in solutions which can allow optimisation of energy usage and efficiency gains.
However, it is also important to note that energy demand is not the only concern, social and governance consequences around workforce and responsible use of AI also require our attention.
Henning Padberg, portfolio manager, global climate and environment strategy
Nordea Asset Management
The rapid advancement of artificial intelligence (AI) significantly impacts our environment, primarily due to the high computational power and vast amounts of electricity required to train and run complex AI models.
Semiconductors are pivotal in tackling this challenge.
As the “brains” behind energy efficiency, silicon chips must evolve to run faster, integrate more capabilities, and process massive data loads reliably, all while consuming less power.
This evolution is crucial for achieving a sustainable digital and green economy.
Companies designing semiconductors play a crucial role in enhancing efficiency.
Modern semiconductor chips, which can contain over a billion circuit elements, require sophisticated automation to manage their complexity, especially as AI technology continues to advance.
In line with its commitment to responsible investment, the Nordea Global Climate and Environment strategy focuses on optimising resource efficiency.
Semiconductor designers, who are key players in this optimisation, form a significant part of our portfolio.
Their advanced, energy-efficient chips are essential in reducing the power consumption of AI systems, supporting our goal of a more sustainable future.
Crystal Geng, Asia ESG research lead
BNP Paribas Asset Management
The rapid development of AI technology has introduced a new set of challenges, and the energy consumption issue is on top of them.
In terms of the environmental impact, data storage and refining AI models and algorithms through thousands of hours of training is energy intensive.
Rising competition over AI among key countries and major market players will likely result in significant greenhouse gas emissions over the coming years.
However, at the same time, upgraded within AI data centres could prompt the same across electrical infrastructure to be more uninterruptible and efficient, and this may also involve upgrading the main electrical service, transformers, switchgear, and distribution systems etc.
According to recent research, GenAI and data centre demand may structurally change the power market dynamics in multiple Asian markets, given that Asia forms more than a third of global data centre capacity.
According to a recent research study, there will be a 24% CAGR in data centre power supply from 2023-28 in Asia.
We will continue to focus on renewable energy and innovative tech investment themes and invest in those contributing to stable and low-carbon electricity system.
We integrate essential indicators into our ESG and net zero AAA frameworks to align our portfolio with the Paris Agreement and actively engage with our investees on AI-related environmental and social risks.
Jerry Goh, investment director of Asian equities
abrdn
Studies have shown that AI chips powering AI technology consume a lot more energy compared to technological software that is not powered by AI.
For instance, a Google search with AI is much more energy intensive versus a normal Google search.
There are several ways in which companies that employ AI can reduce their energy footprint or intensity.
These include exploring the use of renewable energy such as solar, wind, and even hydro.
We have seen this increasingly deployed at some of the global technology firms.
That said, on a broader note, the massive and rapid growth in AI adoption and usage has resulted in a lack of supply of power, particularly in the US and other parts of the world.
To meet the power requirements, governments and companies will need to think about increasing capital expenditure towards raising power generation capacity and to enhance both the hardware and software of state grids to cater to emerging power sources, such as renewables, in addition to traditional sources such as fossil fuels.
Rob Barnett, senior energy analyst
Bloomberg Intelligence
Rising AI spending and the electrification of Europe's energy mix in 2024-30 could revive power demand.
Generative AI hardware, software and other IT investment may grow at a 41% CAGR in 2024-32, according to our global technology team.
That might boost electricity use in Europe, where 10-20% of the investment could take place, led by a proliferation of cloud facilities by Microsoft, Google and others.
Power use for data centres may increase 2-3.5x by 2030, assuming a linear correlation between AI spending and energy demand.
Data centres make up about 2% of European power demand, which sits at about 3,000 terawatt-hours.
While in the US, data-centre electricity use is poised to surge 4-10x by 2030, according to our calculations, and we believe such growth will drive demand for solar plus battery storage and gas generation, given the relatively short time frames for building such technologies versus other sources of power.
Given their focus on ESG and sustainability goals, AI hyperscalers are likely to have a strong preference for lower-emitting sources of power -- especially solar, wind and storage facilities.
The potential surge in AI data centers could further boost gas consumption, based on rising power demand and the fuel's potential uses as onsite backup generation.
Jeffrey Lin, head of thematic equities & Michael van der Meer, head of sustainability
M&G Investments
The rapid rise of generative AI and its computational demands are significantly increasing energy consumption, presenting critical sustainability challenges.
The demand for electricity is set to rise not only from the proliferation of electric vehicles (EVs) but also due to the expanding number of AI-driven data centres.
For example, in the USA, households switching to two electric cars from internal combustion engines could potentially double their electricity usage.
Simultaneously, the significant increase in global renewable power generation appears capable of supporting the electrification trend and the growing electricity demands, including those from AI computations.
In 2023, renewables generated 30% of global electricity, mostly from solar and wind expansions.
The growth in renewable energy, especially solar, expected to continue its upward trajectory, may become the dominant power source by the decade's end, maintaining a compound annual growth rate (CAGR) over 20%.
Additionally, solar projects are now frequently developed with battery storage, allowing for the separation of power supply from generation.
This feature is set to disrupt future energy systems, with stationary batteries anticipated to play a larger role.
The manufacturing sector is already gearing up to meet this surging demand for batteries.
To mitigate the environmental impacts of AI, it is crucial that data centre companies and utilities regulators promote the scaling of renewable energy sources and advanced technologies.