Business

Opinion: As demand for AI data centres grows, we must avoid ‘digital extractivism’

Strategies are needed to prevent a model that serves richer countries at the expense of local needs and ecosystems in the Global South, writes the director of think-tank Transforma
<p>A humanoid robot exhibited at a Chile-Germany innovation summit in Munich, 2024. The growth in artificial intelligence is driving demand for data centres, whose high energy and water needs have caused concern among environmentalists (Image: Marcelo Segura / <a href="https://prensa.presidencia.cl/fotografia.aspx?id=285796">Dirección de Prensa, Presidencia de la República de Chile</a>)</p>

A humanoid robot exhibited at a Chile-Germany innovation summit in Munich, 2024. The growth in artificial intelligence is driving demand for data centres, whose high energy and water needs have caused concern among environmentalists (Image: Marcelo Segura / Dirección de Prensa, Presidencia de la República de Chile)

Artificial intelligence (AI) has been presented by its supporters as the great promise of the 21st century, capable of revolutionising industries – and even helping in the fight against climate change. It is said, for example, that AI will optimise the integration of solar and wind power into “smart grids”, helping to reduce dependence on fossil fuels.

But the other side of this coin is not insignificant: the data centres that provide the computing power to train, deploy and run AI models are energy-greedy, and also generate emissions associated with their operations. The question is inevitable: in the fight against climate change, do the benefits of AI outweigh its negative impact?

Low share of power use today, but with future uncertainty

The data centres that support AI and other digital services consume huge amounts of energy. On a global scale, this currently only accounts for about 1% of electricity demand. But what happens as AI continues to expand? And what could go wrong?

Optimistic scenarios predict that advancements in the energy efficiency associated with software and hardware will keep explosive growth in power demand at bay. One example of this is the world’s largest AI chipmaker Nvidia, which claims that its new Blackwell system, a processor supporting generative AI, is up to 25 times more energy efficient compared to previous models – though, for all its gains in efficiency, each processor may still use significantly more power than its predecessors, drawing up to 1,200 watts per unit.

Other scenarios argue that the sophistication of AI models will come with exponential power consumption: generative AI developed on the basis of large language models (LLM) requires up to 30 times more power during the training phase than a traditional AI model. It is for this reason that a ChatGPT query has been estimated to use up to 10 times more electricity than a Google search.

Investments and risks on the rise

Increasingly complex and intensive uses of AI, such as video creation from text prompts, will require “hyperscale” data centres with power demands of more than 100 megawatts, potentially consuming the equivalent amount of electricity needed to power as many as 400,000 electric vehicles each year. Big technology firms have been duly gearing up: the IEA reports that in 2023, Google, Amazon and Microsoft together made capital investments that were greater than the entire US oil and gas industry – equivalent to 0.5% of the country’s GDP.

The accelerated evolution of AI means that its electricity consumption must be considered at least as a non-negligible risk in the coming years. There are already countries facing strain from the sector’s growth, such as Ireland, where data centres already accounted for over 20% of electricity consumption in 2023, putting pressure on the grid and triggering a moratorium on new data centres in Dublin until 2028.

Santos Dumont supercomputer
The Santos Dumont supercomputer, operated by Brazil’s National Scientific Computing Laboratory, located in the city of Petrópolis. Brazil is said to be one of the Latin American countries with the highest potential in terms of AI infrastructure (Image: Tomaz Silva / Agência Brasil)

The emissions paradox and hidden environmental costs

Paradoxically, AI and its voracious appetite for electricity is growing faster than the transition to fully renewable energy sources and systems. This not only perpetuates the technology’s dependence on fossil fuels, but has in some locations delayed the closure of coal-fired power plants, meaning increased greenhouse gas emissions.

In addition, data centres require large amounts of water for cooling, and critical minerals for the manufacture of their hardware, exacerbating problems of water scarcity and dependence on environmentally questionable supply chains.

The data also reveals a lack of transparency: in 2024, Microsoft and Google not only missed their climate targets, but the emissions associated with their data centres may be underestimated. According to studies, emissions from in-house data centres may be 7.62 times higher than what companies claim.

What about Latin America?

Analysts have highlighted Brazil, Mexico, Chile, Colombia, Peru, Uruguay, Costa Rica and Panama as the countries in Latin America with the highest growth and potential in terms of AI infrastructure. The region, together with Africa and the Middle East, accounts for 5% of the total global electricity consumption by data centres. This brings with it opportunities to strengthen connectivity, foster employment in technology and leverage investments in data centres to boost the integration of renewable energy into the region’s electricity matrix.

We need a strategy to avoid the emergence of a ‘digital extractivism’ that puts data centres at the exclusive service of industrialised countries, at the expense of ecosystems
Felipe Arango García

Indeed, the region is already seeing significant growth in the sector, with countries such as Brazil, Mexico and Chile attracting investors for their robust infrastructure, energy systems and permissive government policies. However, data centres are already putting pressure on electricity and water systems in Mexico, Chile and Uruguay – countries that already suffer from considerable water stress.

Furthermore, it will be necessary to develop a strategy for the harnessing of AI between countries of the Global South, and for its associated infrastructure, to avoid the emergence of a “digital extractivism” that puts data centres at the exclusive service of industrialised countries, at the expense of ecosystems.

The way forward

AI promises climate solutions, but its growing energy demand and environmental impacts create uncertainty. Without measures to ensure efficiency, transparency and sustainability, its development could deepen the crisis it seeks to solve. Latin America has a strategic opportunity, but it must avoid falling into a digital extractivism that aggravates inequalities and puts further pressure on its natural resources.

For this reason, it is essential that Latin American states engage in an open and frank dialogue with academia, the technology industry, think-tanks, and other countries in the Global South, to structure a value chain around the use of AI for the reindustrialisation of our region within the framework of a just energy transition. This involves developing policies, regulations, incentives and safeguards that channel AI towards a use focused both on climate adaptation and mitigation solutions, as well as on economic diversification and tax revenue generation solutions that favour emancipation from fossil fuels.

-->
Cookies Settings

Dialogue Earth uses cookies to provide you with the best user experience possible. Cookie information is stored in your browser. It allows us to recognise you when you return to Dialogue Earth and helps us to understand which sections of the website you find useful.

Required Cookies

Required Cookies should be enabled at all times so that we can save your preferences for cookie settings.

Dialogue Earth - Dialogue Earth is an independent organisation dedicated to promoting a common understanding of the world's urgent environmental challenges. Read our privacy policy.

Cloudflare - Cloudflare is a service used for the purposes of increasing the security and performance of web sites and services. Read Cloudflare's privacy policy and terms of service.

Functional Cookies

Dialogue Earth uses several functional cookies to collect anonymous information such as the number of site visitors and the most popular pages. Keeping these cookies enabled helps us to improve our website.

Google Analytics - The Google Analytics cookies are used to gather anonymous information about how you use our websites. We use this information to improve our sites and report on the reach of our content. Read Google's privacy policy and terms of service.

Advertising Cookies

This website uses the following additional cookies:

Google Inc. - Google operates Google Ads, Display & Video 360, and Google Ad Manager. These services allow advertisers to plan, execute and analyze marketing programs with greater ease and efficiency, while enabling publishers to maximize their returns from online advertising. Note that you may see cookies placed by Google for advertising, including the opt out cookie, under the Google.com or DoubleClick.net domains.

Twitter - Twitter is a real-time information network that connects you to the latest stories, ideas, opinions and news about what you find interesting. Simply find the accounts you find compelling and follow the conversations.

Facebook Inc. - Facebook is an online social networking service. China Dialogue aims to help guide our readers to content that they are interested in, so they can continue to read more of what they enjoy. If you are a social media user, then we are able to do this through a pixel provided by Facebook, which allows Facebook to place cookies on your web browser. For example, when a Facebook user returns to Facebook from our site, Facebook can identify them as part of a group of China Dialogue readers, and deliver them marketing messages from us, i.e. more of our content on biodiversity. Data that can be obtained through this is limited to the URL of the pages that have been visited and the limited information a browser might pass on, such as its IP address. In addition to the cookie controls that we mentioned above, if you are a Facebook user you can opt out by following this link.

Linkedin - LinkedIn is a business- and employment-oriented social networking service that operates via websites and mobile apps.