Globally, we are at the convergence of three global mega-trends—the rise of artificial intelligence (AI), the electrification of everything, and the shift to a low-carbon future.
In recent years, AI and large language models (LLMs) have witnessed a period of rapid expansion and extensive large-scale application. Global tech companies keep finding new ways to bring AI into every facet of our lives. Virtual assistants, chat bots, and LLMs are taking over how we go about our daily lives, and they certainly do bring efficiencies. But AI models need training, with huge amounts of data, housed in massive data warehouses, all powered by electricity, which under our current energy system is primarily generated by the burning of fossil fuels, like coal, oil and gas.
How can we continue the AI expansion, while simultaneously reducing carbon emissions? In some ways, the demand for electricity from AI is no different from the rising demand from the electrification of everything—from electric vehicles (EVs), heat pumps and industry. It’s how we meet the demand that matters most. If we build more fossil fuel plants or extend their useful lives to meet our growing electricity demand, it will come with negative consequences for climate.
But if we use our insatiable appetite for electricity to lean harder into renewables and other low-carbon power sources (perhaps nuclear energy in some global regions) and leverage AI to become more efficient, optimising more and using less, then we can lower carbon emissions, even as AI continues its unstoppable march into our lives.
In some ways, the demand for electricity from AI is no different from the rising demand from the electrification of everything—from EVs, heat pumps and industry. It’s how we meet the demand that matters most.
As the use of AI and related technologies becomes more widespread, there is growing concern over the energy consumption required to power these tools and, therefore, the subsequent environmental impact associated with these advancements.
AI refers to a range of technologies that enable machines to exhibit intelligent behaviour. Generative AI is used for creating new content—i.e., text, images, or videos. Examples include ChatGPT, which generates text, and DALL-E, which can take text and create images. To generate its answers, these tools need to be trained on huge datasets using computer power from thousands of servers that are housed in data centres. Data centres need enormous amounts of electricity to meet that demand. According to a report by Goldman Sachs, a ChatGPT query needs nearly 10 times as much electricity as a Google search. In Goldman Sachs’ view, this difference lies in a coming sea change for how the US, Europe, and the rest of the world will generate power. This is because under the current system, the majority of that energy comes from burning fossil fuels, like coal, oil and gas, the primary drivers of greenhouse gas emissions (GHGs).
And AI is predicated to become even more sophisticated, requiring even more energy. The amount of data required for each upgrade of Chat GPT is requiring more and more data inputs. GPT 1 was trained on a dataset of 11,000 books, GPT 2, was trained using 1.5bn parameters, GPT 3.5, around 175 billion parameters, and Chat GPT 4, the most advanced version, has been trained on an estimated 1.8 trillion parameters (Invgate, February 2024). Every upgrade to AI models and an increase in the size of their datasets, will command more electricity than its predecessor and potentially increase carbon emissions even further.
The explosion in demand for data centres has attracted investor attention in the past few years. They are an attractive investment opportunity due to their utility-like cash flows and risk-adjusted yields.
The explosion in demand for data centres has attracted investor attention in the past few years. They are an attractive investment opportunity due to their utility-like cash flows and risk-adjusted yields (McKinsey & Co, January 2023). However, pressure to make data centres sustainable is high. Not only are their emissions from electricity a concern, but the massive amounts of water required for cooling the servers and the cement and steel required in their build out will add further strain to environmental resources.
Some regulators and governments are imposing sustainability standards on all new builds. The Inflation Reduction Act provides tax credits and production tax credits for renewable energy generation and storage. This development and the surge in demand for AI and its capabilities, and the net -zero pledges of datacentres’ biggest users gives investors opportunities to help data centres secure carbon-free energy.
Datacentres will though, continue with technological advancements and improvements themselves. As Jensen Huang, CEO Nvidia said recently “you can’t just assume that companies will just buy more computers…you also have to assume that those computers are going to get better and faster”.
In recent years, data centre consumption has been a relatively stable 1-2% of global energy consumption, but this is forecast to rise towards 5% by the end of the decade, on par with the aviation industry.
There is no doubt that global tech companies are full steam ahead in the race for AI domination. Global tech giants, Alphabet and Google, have been influenced by the successful launch of OpenAI’s ChatGPT (the now well-known generative AI chatbot that reached 100 million users in just two months). They have, subsequently, launched their own chatbots, Bing Chat and Bard respectively (Vrjie, Uni of Amsterdam, 2023). In the race to produce the best AI systems Google is pouring an enormous amount of resources and funding into training AI systems and building bigger and bigger data centres, all of which require tremendous amounts of electricity and, under the current system, increased CO2 emissions. Google said it spent USD 12 billion on capex in Q1 2023, driven overwhelmingly by investments in data centres to fuel its AI endeavours. The company said it expects that it will keep up that same level of spending year-on-year.
Meta, Microsoft, Amazon, and Google are all users of datacentres in the race for AI domination. All these companies also have net- zero targets.
In our view, the only way AI and net zero can co-exist is if the aggressive lean into AI technologies is matched with an equally aggressive lean into renewables and other low-carbon sources of energy.
Meta, Microsoft, Amazon, and Google are all users of datacentres in the race for AI domination. All these companies also have net-zero targets. They and other hyperscalers have committed to using only carbon-free energy by 2030. Microsoft and Google have pledged to reach 100% clean energy usage for their data centres by 2030 (International Energy Agency, February 2024). Amazon is targeting a much more ambitious target by 2025. These three companies alone currently account for more than 50% of the world’s large-scale data centre capacity (CRN, January 2021).
But just recently, Google released its sustainability report, where it announced that its GHG emissions rose last year by 48% (compared to its baseline year in 2019). It admits “as we further integrate AI into our products, reducing emissions may be challenging”.
Microsoft has taken its climate pledge one step further than Google, saying it intends to be carbon negative by 2030. But just recently, Brad Smith of Microsoft said, “In 2020, we unveiled what we called our carbon moonshot, but that was before the explosion in Artificial intelligence’. So how do the largest consumers of datacentres, and the most aggressive in the AI gold rush, intend to power their ambitious plans?
In our view, the only way AI and net zero can co-exist is if the aggressive lean into AI technologies is matched with an equally aggressive lean into renewables and other low- carbon sources of energy. Whilst wind and solar will be crucial for sustainable AI development, they do face intermittency issues. We see the role of nuclear power in some regions as a key player in optimising performance and providing stability to the grid.
Increasingly small modular reactors (SMRs), a newer form of nuclear power technology, is also showing promise as a source of carbon-free energy. The push to win the AI arms race is swelling the electricity demands of the global tech giants who are all looking for carbon- free energy. The use of renewable energy is already a critical component in all their carbon reduction strategies.
Microsoft, Amazon (AWS), and Google are all working to design green data centres, which use 100% renewable energy. Hyperscalers are also starting to fund the building of renewable energy plants in the face of soaring prices caused by supply-chain shortages. In the UK, for example, Amazon has supported Scottish Power’s wind farm and is purchasing its entire 50-megawatt output. For co-location companies, to help their tenants reach their carbon-free targets, they are signing power purchase agreements (PPAs) with suppliers of renewable energy.
Yet, such moves will not suffice if only renewable energy is involved. The first problem is intermittency. Solar power is generated only in the daytime, while wind power is obviously reliant on the wind. So, under the current system, fossil fuels are often the supplementary source of power from renewable PPAs. It is expected that nuclear energy may also fill part of the gap, and in March this year, Talen Energy announced a USD 650 million deal with Amazon to sell a data centre powered by one of the largest US nuclear plants. Nuclear energy is carbon -free, but does have a long development time. It is also cost-intensive, so it won’t be able to solve everything.
Microsoft, Amazon, and Google are all working to design green data centres, which use 100% renewable energy. Hyperscalers are also starting to fund the building of renewable energy plants in the face of soaring prices caused by supply-chain shortages.
All of this presents opportunities for investors. Not all data centre providers have the scale to procure renewable power either through PPAs or investments in renewable power plants. The pressure to decarbonise data centres also presents opportunities—not only in infrastructure, but in renewables needed to power them, as well as cooling and energy efficiency technologies.
Whilst the market has been quick to embrace the AI narrative, it has not been so kind to the renewable energy sector particularly since it is a key factor with respect to how widely and how quickly AI is adopted. There has been a significant divergence of AI relative to the S&P Global Clean Energy Index (see the chart below).
The explosion in demand for datacentre capacity and allocations has attracted the attention of all types of investors—growth capital, buyout, real estate and, increasingly, infrastructure investors. But there are other opportunities up and down the supply chain, within capital structures, and across asset classes (e.g., listed equity, commodity exposures, alternatives, as well as energy providers). Commitments by the hyperscalers to be net zero make the case for clean energy a promising one. Further, some of the key inputs into datacentres, including copper and uranium, also make these commodities an attractive investment opportunity, given constrained supply backdrops.
The pressure to decarbonise data centres also presents opportunities – not only in infrastructure, but in renewables needed to power them, as well as cooling, and energy- efficiency technologies.
This document has been prepared by LGT Crestone Wealth Management Limited (ABN 50 005 311 937, AFS Licence No. 231127) (LGT Crestone Wealth Management). The information contained in this document is of a general nature and is provided for information purposes only. It is not intended to constitute advice, nor to influence a person in making a decision in relation to any financial product. To the extent that advice is provided in this document, it is general advice only and has been prepared without taking into account your objectives, financial situation or needs (your Personal Circumstances). Before acting on any such general advice, we recommend that you obtain professional advice and consider the appropriateness of the advice having regard to your Personal Circumstances. If the advice relates to the acquisition, or possible acquisition of a financial product, you should obtain and consider a Product Disclosure Statement (PDS) or other disclosure document relating to the financial product before making any decision about whether to acquire it.
Although the information and opinions contained in this document are based on sources we believe to be reliable, to the extent permitted by law, LGT Crestone Wealth Management and its associated entities do not warrant, represent or guarantee, expressly or impliedly, that the information contained in this document is accurate, complete, reliable or current. The information is subject to change without notice and we are under no obligation to update it. Past performance is not a reliable indicator of future performance. If you intend to rely on the information, you should independently verify and assess the accuracy and completeness and obtain professional advice regarding its suitability for your Personal Circumstances.
LGT Crestone Wealth Management, its associated entities, and any of its or their officers, employees and agents (LGT Crestone Group) may receive commissions and distribution fees relating to any financial products referred to in this document. The LGT Crestone Group may also hold, or have held, interests in any such financial products and may at any time make purchases or sales in them as principal or agent. The LGT Crestone Group may have, or may have had in the past, a relationship with the issuers of financial products referred to in this document. To the extent possible, the LGT Crestone Group accepts no liability for any loss or damage relating to any use or reliance on the information in this document.
This document has been authorised for distribution in Australia only. It is intended for the use of LGT Crestone Wealth Management clients and may not be distributed or reproduced without consent. © LGT Crestone Wealth Management Limited 2024.
