Electricity demand in the US has been steady since the mid‑2000s, growing at about half a percent per year on average. This is down to several factors, including the growth of less energy‑intensive sectors (such as financial services), offshoring manufacturing to places like China and Southeast Asia, and better energy efficiency.
Can we deliver the energy we need to power AI?
As we shift towards electrification and build more data centre infrastructure, electricity demand is expected to change significantly. Energy demand estimates vary widely, but key variables include the pace of electrification, onshoring, and data centres. We expect that US electricity demand could grow 2%–4% per year over the next 10–15 years.[1]
Markets have been factoring a sustained uptick in electricity demand due to electrification, onshoring, and data centres, but the debut of ChatGPT in November 2022 marked a turning point for data centre electricity demand expectations. With the rise of generative artificial intelligence (AI), we expect data centre growth to be much higher, potentially increasing from 2%–4% of US electricity consumption[2] to around 8% by 2035.[1]
Generative AI and data centre growth are pushing the US on this broader challenge of how to meet the needs of an electrifying and growing economy with clean electricity, while reducing greenhouse gas emissions. Concerns about electricity scarcity amid technological advancements are not new; we saw similar concerns in the early 2000s and with the emergence of cloud computing, for example. Over the past two decades, energy efficiency has outpaced electricity demand growth in advanced economies (including digitalisation, heat, and mobility).
According to the International Energy Agency (IEA), without efficiency improvements, electricity demand growth would have been 1.6% instead of 0.3%.[3] Since 2010, the number of internet users worldwide has more than doubled, and global internet traffic has expanded 25‑fold.[4] Rapid improvements in energy efficiency have, however, helped moderate grow thin energy demand from data centres and data transmission networks.
To achieve similar results with AI, there are several levers throughout the AI supply chain that could reduce electricity consumption while still meeting demand. These include continued improvements in AI training (as seen recently with Chinese AI company DeepSeek) and AI algorithms, the energy efficiency of information technology equipment, liquid cooling technology, data centre design and modularisation, and clean power adoption. However, it will take time for these levers to have an effect, so we expect AI‑related energy demand will likely have a negative impact on decarbonisation in the near term.
Will AI derail decarbonisation?
While virtually all the largest hyperscalers[5] in the US have longstanding decarbonisation targets in place, they’re now planning for notably higher energy requirements. Data centres require firm and reliable electricity, meaning intermittent power sources like renewables won’t be enough. While demand for renewables and storage are likely to keep rising, we also expect increase demand for natural gas‑fired generation, and the scheduled shutdowns of coal‑fired power plants may face delays. Wind and solar energy will need battery storage and/or natural gas peaking capacity to provide round‑the‑clock reliability.
The US power sector has shown a steady decline in carbon (CO2) emissions for two decades[6] thanks to renewables and natural gas growing their share of the generation fuel mix. Solar deals have dropped from $5/megawatt to $1/megawatt, and the cost of capital for both solar and wind have declined meaningfully, which has made these technologies more attractive investments.[7]
How can AI help address US energy constraints?
AI offers transformative solutions to tackle energy challenges in the US by enhancing grid efficiency, improving renewable energy integration, and optimising energy storage systems.
AI algorithms can improve smart grid management by predicting power demand fluctuations, enabling utilities to adjust supply proactively and prevent blackouts, and accessing surplus energy between energy grids. By analysing vast amounts of data from smart meters and sensors, AI helps balance load distribution and manage energy storage effectively.
Renewable energy forecasting can be improved through AI, enhancing the predictability of renewable energy sources by analysing weather patterns and historical data—allowing for better scheduling and dispatch of power. This ideally leads to more efficient use of renewables and reduces reliance on fossil fuels.
Energy storage can also be improved through AI optimising charging and discharging cycles of energy storage systems, ensuring that excess renewable energy is stored efficiently and released when needed.
Green(er) data centres: Improving energy efficiency
Companies throughout the AI supply chain are hyper focused on improving energy efficiency, which is why we believe the AI sector’s carbon intensity could reduce over time. We expect emissions from the data centre industry to increase in the short term (5–10years) and begin to decrease in the medium term (10–20 years). This is because the advanced clean technologies being invested in today (such as nuclear, geothermal carbon capture and storage, and long duration storage) are expected to be at scale between the mid‑2030s and 2040s.
Ultimately, there are many ways data centres can reduce energy consumption through improvements in energy efficiency. These include hardware components, software and management systems, cooling infrastructure, and power infrastructure.
[1] Analysis by T. Rowe Price.
[2] Source: International Energy Agency, Electricity Mid‑Year Update—July 2024.
[3] Source: International Energy Agency (IEA), “The mysterious case of disappearing electricity demand,” February 14, 2019.
[4] Source: IEA,iea.org/energy-system/buildings/data-centres-and-data-transmission-networks, July 2023.
[5] Hyperscaler is the term usually given to mega‑cap technology companies that occupy AI data centers.
[6] statista.com/statistics/204879/us-carbon-dioxide-emissions-by-sector-since-1950/
[7] Source: BNEF (bnef.com/interactive-datasets/2d5d59acd9000006);commercialsolarguy.com/cost-of-solar-power-capital-down-69-strong-economy-and-learning-curves-abound/
<small> Disclaimer: The views and opinions expressed in this article are solely those of the author(s) and do not necessarily reflect the view or position of the Responsible Investment Association Australasia (RIAA). This article is intended as general information and should not be considered investment advice. It is recommended to seek appropriate professional advice before making any investment decisions.
About the contributors
About the speakers
.jpg)
.jpg)
Ashley Hogan
Associate Analyst, Responsible Investing
-
T.RowePrice
Ashley Hogan is an associate analyst for the Socially Responsible Investing team in the Global Equity division. The Socially Responsible Investing team serves as specialists for incorporating environmental and social considerations into the firm's research process.
.jpg)
.jpg)
Vineet Khanna
Investment Analyst, Utilities
-
T.RowePrice
Vineet Khanna is an investment analyst covering utility companies as part of the natural resources team in the Global Equity Division. He is a vice president and Investment Advisory Committee member of the Global Real Estate Equity, Global Natural Resources Equity, and US Value Equity Strategies.
Electricity demand in the US has been steady since the mid‑2000s, growing at about half a percent per year on average. This is down to several factors, including the growth of less energy‑intensive sectors (such as financial services), offshoring manufacturing to places like China and Southeast Asia, and better energy efficiency.
Can we deliver the energy we need to power AI?
As we shift towards electrification and build more data centre infrastructure, electricity demand is expected to change significantly. Energy demand estimates vary widely, but key variables include the pace of electrification, onshoring, and data centres. We expect that US electricity demand could grow 2%–4% per year over the next 10–15 years.[1]
Markets have been factoring a sustained uptick in electricity demand due to electrification, onshoring, and data centres, but the debut of ChatGPT in November 2022 marked a turning point for data centre electricity demand expectations. With the rise of generative artificial intelligence (AI), we expect data centre growth to be much higher, potentially increasing from 2%–4% of US electricity consumption[2] to around 8% by 2035.[1]
Generative AI and data centre growth are pushing the US on this broader challenge of how to meet the needs of an electrifying and growing economy with clean electricity, while reducing greenhouse gas emissions. Concerns about electricity scarcity amid technological advancements are not new; we saw similar concerns in the early 2000s and with the emergence of cloud computing, for example. Over the past two decades, energy efficiency has outpaced electricity demand growth in advanced economies (including digitalisation, heat, and mobility).
According to the International Energy Agency (IEA), without efficiency improvements, electricity demand growth would have been 1.6% instead of 0.3%.[3] Since 2010, the number of internet users worldwide has more than doubled, and global internet traffic has expanded 25‑fold.[4] Rapid improvements in energy efficiency have, however, helped moderate grow thin energy demand from data centres and data transmission networks.
To achieve similar results with AI, there are several levers throughout the AI supply chain that could reduce electricity consumption while still meeting demand. These include continued improvements in AI training (as seen recently with Chinese AI company DeepSeek) and AI algorithms, the energy efficiency of information technology equipment, liquid cooling technology, data centre design and modularisation, and clean power adoption. However, it will take time for these levers to have an effect, so we expect AI‑related energy demand will likely have a negative impact on decarbonisation in the near term.
Will AI derail decarbonisation?
While virtually all the largest hyperscalers[5] in the US have longstanding decarbonisation targets in place, they’re now planning for notably higher energy requirements. Data centres require firm and reliable electricity, meaning intermittent power sources like renewables won’t be enough. While demand for renewables and storage are likely to keep rising, we also expect increase demand for natural gas‑fired generation, and the scheduled shutdowns of coal‑fired power plants may face delays. Wind and solar energy will need battery storage and/or natural gas peaking capacity to provide round‑the‑clock reliability.
The US power sector has shown a steady decline in carbon (CO2) emissions for two decades[6] thanks to renewables and natural gas growing their share of the generation fuel mix. Solar deals have dropped from $5/megawatt to $1/megawatt, and the cost of capital for both solar and wind have declined meaningfully, which has made these technologies more attractive investments.[7]
How can AI help address US energy constraints?
AI offers transformative solutions to tackle energy challenges in the US by enhancing grid efficiency, improving renewable energy integration, and optimising energy storage systems.
AI algorithms can improve smart grid management by predicting power demand fluctuations, enabling utilities to adjust supply proactively and prevent blackouts, and accessing surplus energy between energy grids. By analysing vast amounts of data from smart meters and sensors, AI helps balance load distribution and manage energy storage effectively.
Renewable energy forecasting can be improved through AI, enhancing the predictability of renewable energy sources by analysing weather patterns and historical data—allowing for better scheduling and dispatch of power. This ideally leads to more efficient use of renewables and reduces reliance on fossil fuels.
Energy storage can also be improved through AI optimising charging and discharging cycles of energy storage systems, ensuring that excess renewable energy is stored efficiently and released when needed.
Green(er) data centres: Improving energy efficiency
Companies throughout the AI supply chain are hyper focused on improving energy efficiency, which is why we believe the AI sector’s carbon intensity could reduce over time. We expect emissions from the data centre industry to increase in the short term (5–10years) and begin to decrease in the medium term (10–20 years). This is because the advanced clean technologies being invested in today (such as nuclear, geothermal carbon capture and storage, and long duration storage) are expected to be at scale between the mid‑2030s and 2040s.
Ultimately, there are many ways data centres can reduce energy consumption through improvements in energy efficiency. These include hardware components, software and management systems, cooling infrastructure, and power infrastructure.
[1] Analysis by T. Rowe Price.
[2] Source: International Energy Agency, Electricity Mid‑Year Update—July 2024.
[3] Source: International Energy Agency (IEA), “The mysterious case of disappearing electricity demand,” February 14, 2019.
[4] Source: IEA,iea.org/energy-system/buildings/data-centres-and-data-transmission-networks, July 2023.
[5] Hyperscaler is the term usually given to mega‑cap technology companies that occupy AI data centers.
[6] statista.com/statistics/204879/us-carbon-dioxide-emissions-by-sector-since-1950/
[7] Source: BNEF (bnef.com/interactive-datasets/2d5d59acd9000006);commercialsolarguy.com/cost-of-solar-power-capital-down-69-strong-economy-and-learning-curves-abound/
<small> Disclaimer: The views and opinions expressed in this article are solely those of the author(s) and do not necessarily reflect the view or position of the Responsible Investment Association Australasia (RIAA). This article is intended as general information and should not be considered investment advice. It is recommended to seek appropriate professional advice before making any investment decisions.