Since the emergence of ChatGPT at the latest, it has become evident that artificial intelligence (AI) will shape our future in every conceivable way. One hope is that these digital brains will spearhead the transition to carbon neutrality. With their assistance, for instance, millions of electric vehicles can be charged in a grid-friendly manner when green electricity is abundantly available and then discharged when it is scarce. Buildings can be heated or cooled in an energy-efficient way. An increasingly complex electricity grid can be maintained cost-effectively. These are just a few examples.
By 2030, optimistic forecasts project that AI will help to reduce global greenhouse gas emissions by an amount equivalent to what the countries of the European Union produce today.
Yet, as it goes in life, there's a catch. AI not only has the potential to conserve resources: it also consumes vast amounts of electricity and water. Within a few years, AI data centers alone are expected to consume as much electricity as Argentina does in a year. The energy appetite for developing large language models, which are a prerequisite for ChatGPT & Co., is not even included in this estimate. And as long as the necessary electricity is not produced emission-free, obviously corresponding CO2 emissions occur.
AI is neither a panacea nor a boondoggle
This is hardly conducive to the sustainability goals of tech giants. Google reported a 13 percent increase in its greenhouse gas emissions for 2023. Compared to 2019, they have grown by almost half. The main reason: new data centers required for AI and cloud computing. At Microsoft, the increase in 2023 was even 29 percent. By 2030, these corporations aim to be carbon-neutral. At the moment, they are getting further away from this objective. Moreover, the U.S., the world's second-largest CO2 emitter, might miss its goal of halving emissions by 2030, partly due to the data centers, despite the successes attributed to the IRA.
It is estimated that the cooling of AI data centers worldwide will soon require as much water as United Kingdom uses in a year. An average Google data center consumes three Olympic swimming pools full of water per day. No wonder resistance is growing in some U.S. regions, particularly those affected by drought, against the construction of new data centers.
The march of AI will not be halted by such collateral damage. States around the world are already competing to position themselves as locations for data centers. At the same time, data centers cannot simply be built where water and electricity are cheap and plentiful. The facilities are too critical for data security. All of this presents a challenge.
Savings Must Exceed Consumption
However, this also acts as an incentive to advance the technology with as much energy efficiency as possible. And to simultaneously promote the expansion of renewable energies and energy infrastructure. Everywhere. In the hope that AI will one day indeed save significantly more resources than it consumes.
Interestingly, ChatGPT 4.0 was still unable to generate this “To the point” column upon request even, with a precise description of the desired outcome. The AI managed to spit out a few sterile statements of common knowledge. No more. Yet those ultimately futile 28 prompts consumed some 80 watt-hours of electricity (see figure). I could have used that to make myself a cup of tea and bake a few waffles. Which would have been cozier.
Estimated energy consumption per request (in Wh)
This publication is addressed exclusively at recipients in the EU, Switzerland, Liechtenstein and the United Kingdom. This report is not being distributed by LBBW to any person in the United States and LBBW does not intend to solicit any person in the United States. LBBW is under the supervision of the European Central Bank (ECB), Sonnemannstraße 22, 60314 Frankfurt/Main (Ger many) and the German Federal Financial Supervisory Authority (BaFin), Graurheindorfer Str. 108, 53117 Bonn (Ger many) / Marie-Curie-Str. 24-28, 60439 Frankfurt/Main (Germany). This publication is based on generally available sources which we are not able to verify but which we believe to be reliable. Nevertheless, we assume no liability for the accuracy and completeness of this publication. It conveys our non-binding opinion of the market and the products at the time of the editorial deadline, irrespective of any own holdings in these products. This publication does not replace individual advice. It serves only for informational purposes and should not be seen as an offer or request for a purchase or sale. For additional, more timely in-formation on concrete investment options and for individual investment advice, please contact your investment advisor. We retain the right to change the opinions expressed herein at any time and without prior notice. Moreover, we retain the right not to update this information or to stop such updates entirely without prior notice. Past performance, simulations and forecasts shown or described in this publication do not constitute a reliable indicator of future performance. The acceptance of provided research services by a securities services company can qualify as a benefit in supervisory law terms. In these cases LBBW assumes that the benefit is intended to improve the quality of the relevant service for the customer of the benefit recipient. Additional Disclaimer for recipients in the United Kingdom: Authorised and regulated by the European Central Bank (ECB), Sonnemannstraße 22, 60314 Frankfurt/Main (Germany) and the German Federal Financial Supervisory Authority (BaFin), Graurheindorfer Str. 108, 53117 Bonn (Germany) / Marie-Curie-Str. 24-28, 60439 Frankfurt/Main (Germany). Deemed authorised by the Prudential Regulation Authority. Subject to regulation by the Financial Conduct Authority and limited regulation by the Prudential Regulation Authority. Details of the Temporary Permissions Regime, which allows EEA-based firms to operate in the UK for a limited period while seeking full authorisation, are available on the Financial Conduct Authority’s website.
