Power Hungry AI
The real challenge is not just innovation itself, but managing the change it brings about.
Peter Drucker
Generative artificial intelligence (‘Gen AI’) has been the dominant theme in markets this year, driving innovation, investment, and new opportunities. Many are excited about the transformative potential of AI, but there lies an inconvenient challenge beneath this excitement: the increased energy needed to sustain these technological advancements. The International Energy Agency (IEA) predicts that rising Gen AI workloads will double global data centre electricity demand by 2026. This is equivalent to adding the energy needs of Sweden.
As our major tech holdings—Microsoft, Alphabet, and Meta Platforms—increase their AI investments, we’ve taken a step back to assess the broader business implications and the amount of energy required to power their data centres. The world is complicated, and innovation is often messy. The future of AI will undoubtedly require the careful management of the operational and environmental challenges these innovations bring. As long-term investors, we must understand the implications.
Why all this Energy?
AI’s energy usage can be divided into two primary stages: training and inference. During training, AI models learn from large datasets, requiring intensive computational power to process data, refine algorithms, and enhance their understanding. While OpenAI has not publicly disclosed the exact energy usage for training its first ChatGPT model, estimates based on the number of parameters and the complexity of the model suggest the amount of energy is roughly equivalent to the power consumption of ~150 US homes for an entire year.
Inference, the second stage, refers to how AI models apply their training to new data, making predictions or generating responses. This process is less energy-consumptive than training but will become more significant as AI tools are adopted widely across industries. Generating a response to a ChatGPT prompt is approximately 10x more energy intensive than a standard search query on Google. While training is a substantial upfront investment in energy, inference becomes a recurring demand as AI is integrated into more business processes and everyday tasks, see Figure 1.
Figure 1: Inference will account for a growing proportion of AI power demand
Source: Wells Fargo, 2024
The Art of Balance: Weighing Opportunity and Risk
The value proposition of Gen AI lies in its potential to drive higher levels of efficiency, innovation, and personalised experiences, while transforming industries and solving problems previously beyond our reach.
However, realising this potential requires balancing AI’s benefits with its societal and ethical implications, as well as establishing the right regulatory framework to ensure responsible use. We believe that Microsoft, Alphabet, and Meta recognise this. Their continued licence to operate relies on their ability to navigate these challenges responsibly.
The energy challenge is one we’ve encountered before. We don’t have to look far back into history to see the same pattern play out – no civilisation has ever managed to decouple economic growth from increased energy consumption.
Energy Bottlenecks and Grid Electrification
The rapid growth of Gen AI has exposed vulnerabilities in current energy infrastructure. With energy markets already forecast to be tight over the next 5-10 years, growth in AI could exacerbate supply challenges. This is particularly relevant in regions like Ireland and parts of the United States, where the concentration of data centres has already strained electricity grids.
The expansion of AI has necessitated a parallel increase in energy supply and updates to grid infrastructure. Take the US as an example, it is home to 33% of the world’s data centres which are expected to account for 6% of the country’s total electricity consumption by 2026. Many parts of the US grid that are still in use today were constructed in the mid-20th century. Components—such as transmission lines, transformers, and substations—are nearing or have surpassed their expected lifespans. The grid in its current form is also not fit for electrification, an essential aspect of the energy transition from fossil fuels to cleaner sources of energy.
We’ve long known that the grid has been underinvested in, but it seems that only now, with the alarmist headlines surrounding Gen AI and its energy consumption, people are finally starting to pay attention to the issue. The grid needs to be upgraded with modern technology so that it is fit for renewable energy transmission. This takes time and, most importantly, investment.
Reasons to be Hopeful.
In the middle of difficulty lies opportunity.
Albert Einstein
We’re at a crucial turning point when it comes to the energy transition, and we simply can’t afford to waste any time. Just as Gen AI has demonstrated the remarkable capabilities of human ingenuity, addressing the energy dilemma highlights a similar need for creative solutions.
Big tech has a vested interest in solving this problem too. Some of the most innovative minds and companies in the world are stepping up to tackle it. Importantly, they’re not just bringing their expertise—they’re also backing it with significant financial investment. Microsoft, Alphabet, and Meta Platforms have collectively spent $100 billion in capital expenditure in 2023, primarily on data centre infrastructure and resource optimisation, including power.
These companies have also set ambitious renewable energy goals, including Microsoft’s target of 100% renewable energy by 2025, and Alphabet’s goal of 24/7 carbon-free energy by 2030. To achieve this, many are using Power Purchase Agreements (PPAs) to secure financing for new wind, solar, and nuclear projects, driving the expansion of global renewable energy. In 2023, PPA demand surged, largely driven by tech leaders like Microsoft and Alphabet.
They’re also investing in next-generation nuclear technology, including small modular reactors (SMRs), which offer low-emission power and flexibility. Microsoft recently signed the world’s first fusion energy purchase agreement with Helion Energy, aiming to receive electricity from its first fusion power plant by 2028. While nuclear isn’t a short-term solution due to its lengthy development timelines, it presents a promising way to balance AI’s growing energy demands with sustainability goals in the long run.
In addition to the development of cleaner energy sources to power data centres, hardware is also expected to become more energy-efficient and cost-effective. NVIDIA’s newest Blackwell platform enabling large language models to run at almost 25x less cost and energy consumption using its GB200 chip versus its predecessor, the H100.
Short-term Challenges, Long-term Vision
In the immediate future, the path to net zero for major tech companies is unlikely to be straight. For example, Microsoft and Alphabet have recently missed some of their emissions reduction targets, citing challenges such as the carbon embedded in building new data centres and the energy demands of advanced hardware.
Alphabet’s total greenhouse gas (GHG) emissions in 2023 were 14.3 million tCO2e, representing a 13% year-over-year increase and a 48% increase compared to their 2019 target base year. Similarly, Microsoft’s emissions in 2023 were 17.2 million tCO2e, a 4% year-over-year increase and a 40% increase from 2020 levels. Despite rising emissions, their assurances of long-term decarbonisation plans and ongoing investments in climate action mean that engagement with them on the matter is not yet necessary.
We believe that the short-term difficulties associated with AI’s energy needs should not overshadow the long-term potential of these investments. It’s essential to recognise that while AI is energy-intensive, it can help direct the necessary capital investments into long-term, secure, clean energy sources. This in turn will drive decarbonisation in other sectors.
Gen AI may also have substitution effects, where digital processes replace more carbon-intensive activities. Alphabet has leveraged AI to optimise data centres’ energy efficiency through AI-driven cooling systems, generative AI for predictive maintenance, and AI-enhanced load balancing. These innovations improve resource allocation, streamline the distribution of computing loads across servers, and significantly reduce power consumption.
The many potential applications of AI are yet to be explored and as the technology continues to mature, its potential to enhance productivity and resource efficiency across industries will become a key factor in driving sustainable economic growth.
Asking the Hard Questions
Pragmatism and hope are not opposites, they are companions. Real hope sees the difficulties ahead but moves forward with practical steps.
Unknown
Our companies have made significant commitments to sustainability, backed by real financial investment. We take these commitments at face value because their track record demonstrates that they have a strong grasp of the fact that meeting both societal and customer needs is essential for their enduring success.
As shareholders, it’s our duty to ensure they stay on course. We will continue to closely monitor how they respond to these challenges and make sure their actions align with sound business principles. If we detect any breakdown in their commitment to corporate integrity and responsible business practices, it will lead us to question management’s effectiveness and the long-term sustainability of profits.
Progress is about balancing innovation with responsibility and managing externalities along the way. We believe that with the right mix of capital, patience, and a thoughtful regulatory framework, the challenges related to AI energy intensity can be solved.
It is only by asking these hard questions that we can fully appreciate the true complexity of how businesses operate and navigate the world around them. It is precisely by observing how our investee companies tackle these challenges that we gain valuable insights, which in turn informs our investment conviction.
Further information relating to how ESG integration is applied to the fund can be found in the fund prospectus and investor disclosure document. For further information relating to Troy’s approach to company voting and engagement, please see Troy’s Responsible Investment and Stewardship Policy available at www.taml.co.uk. The document has been provided for information purposes only. Neither the views nor the information contained within this document constitute investment advice or an offer to invest or to provide discretionary investment management services and should not be used as the basis of any investment decision. The document does not have regard to the investment objectives, financial situation or particular needs of any particular person. Although Troy Asset Management Limited considers the information included in this document to be reliable, no warranty is given as to its accuracy or completeness. The views expressed reflect the views of Troy Asset Management Limited at the date of this document;
however, the views are not guarantees, should not be relied upon and may be subject to change without notice. No warranty is given as to the accuracy or completeness of the information included or provided by a third party in this document. Third party data may belong to a third party. Past performance is not a guide to future performance. All references to benchmarks are for comparative purposes only. Overseas investments may be affected by movements in currency exchange rates. The value of an investment and any income from it may fall as well as rise and investors may get back less than they invested. The investment policy and process of the may not be suitable for all investors. Tax legislation and the levels of relief from taxation can change at any time. References to specific securities are included for the purposes of illustration only and should not be construed as a recommendation to buy or sell these securities. Although Troy’s information providers, including without limitation, MSCI ESG Research LLC and its affiliates (the “ESG Parties”), obtain information from sources they consider reliable, none of the ESG Parties warrants or guarantees the originality, accuracy and/or completeness of any data herein. None of the ESG Parties makes any express or implied warranties of any kind, and the ESG Parties hereby expressly disclaim all warranties of merchantability and fitness for a particular purpose, with respect to any data herein. None of the ESG Parties shall have any liability for any errors or omissions in connection with any
data herein. Further, without limiting any of the foregoing, in no event shall any of the ESG Parties have any liability for any direct, indirect, special, punitive, consequential or any other damages (including lost profits) even if notified of the possibility of such damages. All reference to FTSE indices or data used in this presentation is © FTSE International Limited (“FTSE”) 2024. ‘FTSE ®’ is a trademark of the London Stock Exchange Group companies and is used by FTSE under licence. Issued by Troy Asset Management Limited (registered in England & Wales No. 3930846). Registered office: 33 Davies Street, London W1K 4BP. Authorised and regulated by the Financial Conduct Authority (FRN: 195764) and registered with the U.S. Securities and Exchange Commission (“SEC”) as an Investment Adviser (CRD: 319174). Registration with the SEC does not imply a certain level of skill or training.
© Troy Asset Management Limited 2024