Is AI slowing down the energy transition?

Executive summary:

• AI (artificial intelligence) is an extremely energy-intensive technology. As its usage becomes increasingly widespread around the globe, energy consumption is soaring, along with a demand for additional power.
• The surge in energy demand is leading to higher carbon emissions among tech companies, which are heavy users of AI. This is a problem, as many of these companies committed to net-zero carbon emissions targets before the AI boom set in. Currently, many companies are grappling with this dilemma by purchasing carbon offsets. 
• Generative AI and large language models are primarily responsible for the surge in energy demand, but other emerging AI technologies have the potential to mitigate carbon emissions by improving energy efficiencies.
• AI presents a wide range of opportunities and risks for investors. Utility companies, data centers, and software companies are among those that could benefit the most from the technology.

Artificial intelligence (AI) is revolutionising the global economy, driving operational efficiencies, automating processes, analysing and interpreting vast amounts of data, and now even generating content and images.  To some, AI may seem like something out of a Harry Potter movie—a powerful, almost magical tool with endless possibilities.

However, CEOs across the globe are warning that energy, not chips, will be the limiting factor for the growth of this technology. As power demand surges, companies are being forced to confront the reality of potentially resorting to less sustainable energy sources to meet their immediate needs. This begs the question: are the net-zero targets set by companies in recent years now at risk of becoming obsolete?

While we may marvel at all the novel uses for AI, it’s important for investors to consider a practical question: what is the impact of AI on power grids and the energy transition, and what are the corresponding investment implications? This article seeks to explore these crucial topics.

Part 1: Current pre-pandemic trends in energy consumption vs. energy supply

To better understand the extent to which AI may exacerbate energy shortages, it’s worthwhile to first analyse the current state of the energy markets. This way, we can see the current strains the power grids face.

While we could choose any number of metrics to gauge the current stresses on the power grids, in this report we’ll focus on three key measures:

• Consumption relative to theoretical full capacity
• Consumption relative to actual production
• Growth in consumption relative to growth in capacity

Exhibit 1: Summary of current electrical grid dynamics by country/region 

Source: EMBER. Data through 2023. 

First, let’s examine the total electricity consumption in each country/region as a percentage of full theoretical capacity. In this context, we consider full theoretical capacity to be the total amount of electricity that the power grids could generate in a year if all power plants were operating at full capacity each day.

On the surface, these calculations may suggest that the power grids have ample excess capacity, with none of the countries/regions consuming more than 50% of full theoretical capacity in a given year. However, it’s important to keep in mind that power plants may not be able to continuously produce power. According to the Department of Energy, nuclear plants had an average capacity factor of 93% in 2021, meaning they could reliably produce 93% of full theoretical capacity three years ago. But solar could only produce 25%. These dynamics suggest that there may not be as much excess capacity as one might think.

We can also look at the ratio of electricity consumption to electricity generation. Here the U.S. stands out as an interesting datapoint: electricity consumption was actually slightly higher than electricity production, meaning that the U.S. had to rely in part on imported electricity. While it is true that there are proposals (such as the Champlain Hudson Power Express) for more grid interconnections between Canada and the U.S., lengthy and complex permitting processes mean that it may take time for more of these interconnections to become available.

Meanwhile, in 2023, Europe and Asia consumed all of the power that was produced in their respective continents. So overall, on this metric, the electrical grid may be somewhat strained across regions.

One final metric to consider: the growth in energy consumption vs. energy capacity over the past 10 years. If energy consumption is growing slower than energy capacity, then over time this might help ease pressures on the electricity grid, assuming that the new power plants have a similar average capacity factor as the old power plants.

In the U.S., Europe, and Asia, the cumulative average growth rate in electricity consumption was slightly slower than the cumulative average growth rate in electricity capacity, suggesting that perhaps there may have been a narrow reprieve with the passage of time. But nevertheless, the totality of the three metrics suggests that the overall electrical grid may still be somewhat fragile.

Part 2: AI’s incremental impact on energy consumption and carbon emissions

As AI becomes increasingly integrated in our daily lives and its capabilities continue to advance, energy consumption has seen a significant surge. A recent study found that generating a single image using generative AI requires as much energy as fully charging a smartphone.¹ Moreover, data centers use up 10 times more power for these AI-related tasks than they do for a normal search. The computational power required to develop machine learning models has doubled every five to six months since 2010.² This dynamic presents a significant challenge for policymakers and tech leaders alike: balancing the need for AI-driven growth with the imperative to reduce emissions and meet global climate targets.

In 2022, the International Energy Agency (IEA) estimated that data centers, cryptocurrencies and AI consumed almost 2% of the total global electricity demand.³ Exhibit 2 shows the IEA’s projections through 2026, which predict that global electricity consumption of these technologies will range between 620 to 1,050 terawatt hours (TWh), up from 460 TWh in 2022. To put this into perspective, this increase is roughly equivalent to adding at least one Sweden or, at most, one Germany’s entire electricity consumption to the grid, according to the IEA.

Exhibit 2: Global electricity demand from data centers, AI and cryptocurrencies, 2019-2026

Source: Chart from IEA “Electricity 2024” report.

Such trends are raising concerns about whether companies will be able to meet their climate targets set before the AI boom. The Magnificent Seven (Mag 7) companies, for example, are finding it increasingly difficult to reduce their emissions. The energy-intensive nature of AI is forcing them to explore alternative energy sources, such as onsite power generation and even nuclear energy, to maintain operations without further escalating their carbon emissions. Google’s recent sustainability report revealed that they experienced a 48% increase in total greenhouse gas emissions from 2019 to 2023 and a 37% rise in Scope 2⁷ emissions between 2022 to 2023 alone. This is a trend we are seeing in other technology companies, whereby companies are seeing sharp increases in their carbon footprints rather than the opposite.

As the environmental impact becomes more prominent, some tech companies are attempting to meet their net-zero commitments through offsets, such as investing in clean energy projects. This approach is often chosen because current sustainability reporting practices allow for carbon-emission reductions to be accounted for through global investments rather than solely through direct reductions in emissions. For example, one company is matching 100% of its electricity consumption with renewable energy purchases. This has brought media attention to the issue, with some individuals worried that these companies may be too reliant on creative accounting to deal with rising carbon emissions —rather than confronting the problem with technological and market improvements. Notably, these offsets do not eliminate the companies’ underlying emissions, which will continue to grow as long as data centers rely on fossil fuels for power generation. While offsets can help balance a company’s carbon ledger, using them ultimately just means that as a company’s emissions increase, their investments grow in tandem—without reducing the company’s overall carbon footprint.

Other AI companies are opting to set operations in regions with abundant energy supplies—such as Iceland—in an attempt to mitigate their environmental impact without making substantial changes to their operations.⁸ However, accessing renewable-heavy grids presents challenges. For instance, more than half of the new utility-scale solar capacity in the U.S. is planned for just three states: Texas (35%), California (10%), and Florida (6%)⁹, highlighting the potential for difficulties for companies not located in or near these regions seeking to access renewable energy. This limitation forces companies to either relocate their operations—which is a costly and logistically complex process—or continue relying on fossil fuels, which undermines their sustainability goals.

Part 3: Offsets and Efficiency

While AI’s increasing energy demands are a concern, the technology also offers potential to support the energy transition. Much of the recent discussion in the media concerning AI has been centered around generative AI and large language models, but AI is a broad term that encompasses a myriad of technologies, some of which can be incredibly powerful in reducing energy usage. In a recent interview in London, Bill Gates suggested that AI could drive more than a 6% reduction in overall energy consumption.¹⁰

A recent report by Boston Consulting Group (BCG) supports this view, predicting that AI could help mitigate 5% to 10% of global GHG (greenhouse gas) emissions by 2030. AI’s ability to optimise operations will be central for this to be the case, particularly in the energy, transportation and agriculture sectors, which are major sources of emissions. For instance, AI can improve energy use in buildings, reduce waste in supply chains, and enhance the efficiency of renewable energy sources.¹¹

AI can also enhance grid management by better predicting energy supply and demand and enabling real-time monitoring, predictive analytics and dynamic control of energy storage assets.  This helps minimise emissions while maximising the value of energy, making it a more attractive option. For example, machine learning models can improve the accuracy of renewable energy forecasts, helping companies better align energy supply with demand. This also enables load shifting, where energy use is timed to match peak renewable energy output, cutting costs and encouraging more investment in renewables.¹² AI can also be used to optimise the use of battery storage systems by predicting when to store excess renewable energy and when to release it, ensuring a consistent energy supply when the sun isn’t shining, or the wind isn’t blowing.

¹ Carnegie Mellon University & Hugging Face 2024 - https://arxiv.org/pdf/2311.16863

² IEA, 2023 - https://www.iea.org/commentaries/why-ai-and-energy-are-the-new-power-couple

³ IEA , 2024 - https://iea.blob.core.windows.net/assets/18f3ed24-4b26-4c83-a3d2-8a1be51c8cc8/Electricity2024-Analysisandforecastto2026.pdf

⁴ Goldman Sachs, 2024 - https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand

⁵ Patterson et al, 2021 - https://arxiv.org/pdf/2104.10350

⁶ A typical coal-fueled power plant uses 1.14 pounds per kilowatt hour (KWh), according to the US Energy Information Administration.

⁷ Scope 2 emissions are indirect GHG emissions associated with the purchase of electricity, steam, heat or cooling.

⁸ Nature, 2024 - https://www.nature.com/articles/d41586-024-01137-x

⁹ EIA, 2024 - https://www.eia.gov/todayinenergy/detail.php?id=61424

¹⁰ Techspot, 2024 - https://www.techspot.com/news/103617-bill-gates-dont-have-worry-about-ai-energy.html

¹¹ BCG, 2023 - https://www.bcg.com/publications/2023/how-ai-can-speedup-climate-action#:~:text=1.,related%20adaptation%20and%20resilience%20initiatives.

¹² IEA, 2023 - https://www.iea.org/commentaries/why-ai-and-energy-are-the-new-power-couple