AI boom is reshaping energy needs — and the world needs to be prepared to power it

In the past few years, the amount of computer power used to train artificial intelligence (AI) models has grown by a factor of one trillion. Laura Cozzi, the International Energy Agency’s (IEA’s) director of sustainability, technology and outlooks, notes that more than 40% of online populations in countries like Brazil, India, Indonesia and the US regularly use generative AI (like ChatGPT or Meta AI).

From Face ID on your smartphone to asking ChatGPT for a recipe, from banking apps to real-time traffic predictions on Google Maps, you are using AI.

“So, in many ways, we are actually ushering in an era in which AI is seen as very much a general-purpose technology, much like electricity has been in the 20th century,” she said. But unlike passive infrastructure, AI feeds on data and electricity, consuming staggering amounts of both.

‘There is no AI without electricity’

“AI is one of the biggest stories in the energy world today,” said Fatih Birol, the IEA’s executive director, at the launch of the agency’s special report, Energy and AI, on Friday, 11 April.

“We all know that it is going to affect our economy, industry and all parts of our daily lives substantially. It is already.”

The report is the most comprehensive, data-driven global analysis to date on the growing interdependence between energy and artificial intelligence. At the centre of that relationship, Birol emphasised, is electricity.

“There is no AI without electricity — it will require a substantial amount,” he said. “The energy sector is at the heart of all AI matters — in two ways.”

First, the power needs of data centres — the backbone of AI infrastructure — are surging. These facilities store and process the enormous volumes of data required for machine learning, natural language processing and image recognition.

The report said global electricity demand from data centres was expected to more than double by 2030, reaching around 945 terawatt-hours (TWh) — roughly equivalent to Japan’s total annual electricity consumption. AI will be the main driver of this growth, with AI-optimised data centre demand projected to more than quadruple over the same period.

Globally, data centre electricity consumption has been growing at about 12% annually since 2017, more than four times the overall rate of electricity demand growth. In the US, data centres will account for nearly half of all electricity demand growth by 2030.

“It is huge,” said Birol. “To put it in context: the electricity used by data centres for AI will surpass the combined consumption of the chemical, steel, aluminium and cement industries — all of which are energy-intensive.”

At the same time, AI offers the potential to reshape the energy sector, improving grid management, accelerating clean energy innovation and enhancing energy security by countering cyber threats. However, Birol warned that the energy industry may not be prepared for the scale and speed of AI’s rise. Countries with secure, clean and readily available electricity will have a competitive edge in attracting investment and the AI industry.

Why data centres consume so much energy

Data centres are crucial for AI, processing and storing data for AI applications. Their energy intensity is due to:

• High computational power. AI models require vast computational resources, running thousands of servers around the clock.


• Cooling systems. Servers generate significant heat, necessitating robust cooling systems. Many data centres use water-based cooling systems to dissipate heat.


• Redundancy and reliability. Data centres employ backup power systems and redundancy measures to ensure uninterrupted service, further increasing energy consumption.

SA is African data centre hub

South Africa is a major hub for data centres in Africa, with significant investments and expansions by major players like Amazon Web Services, Microsoft Azure and Teraco.

The IEA’s report projects that by 2030, electricity use per person from data centres in South Africa will be 15 times higher than the rest of Africa.

But Teraco, a leading data centre operator in South Africa, noted that while this ratio is directionally correct, it’s misleading as it doesn’t reflect South Africa’s role in servicing sub-Saharan Africa’s content and cloud requirements — not just local demand.

While South Africa is expected to reach 25 kilowatt-hours (kWh) per person from data centre use by 2030 — compared to just 2kWh across the rest of Africa — this remains modest next to countries like the US, where usage is projected to hit 1,200kWh per person.

“The takeaway is that South African and African data centre electricity consumption massively lags behind that of more industrialised countries,” said Teraco’s CEO, Jan Hnizdo.

Still, with energy-hungry servers and cooling systems, data centres are set to become major electricity users in a country already battling load shedding and emissions targets. Teraco used 231 million kWh in 2023, with demand set to rise as new sites come online.

To meet this growth sustainably, Teraco has pledged to use 100% renewable energy by 2035, aiming for 70% by 2027. A key step is a R3.5-billion, 120MW solar plant under construction, expected to wheel power through Eskom’s grid to Teraco’s facilities. The company has also signed a wind power deal with local aggregator NOA.

Teraco is also improving efficiency, with its newest sites operating at a power usage effectiveness (PUE) of 1.3. Historically, data centres needed one megawatt (MW) of cooling power for every 1MW of computing power, resulting in a PUE greater than 2.00. Modern energy-efficient data centres aim for a PUE of 1.3 or better.

All facilities are being developed with sustainability in mind, including heat recycling, rainwater harvesting and waste diversion, with a goal of zero waste to landfill by 2028.

Congested grids

The IEA’s report includes a geospatial analysis of 10,000 data centres worldwide, finding that data centres tend to be located close to urban areas and cluster together, intensifying local demand.

Addressing this rising demand requires speeding up permission for new power plants, ensuring the energy sector can respond rapidly to demand spikes, and improving the flexibility and efficiency of data centres. Collaboration between governments, utilities and the tech industry was crucial to prevent bottlenecks in power supply, said Cozzi.

Climate change and emissions 

The acceleration of AI and the energy needed to power it has drawn concerns over the amount of emissions it generates. In a worst-case scenario, the IEA estimates that data centres could emit around 500 million tonnes of CO₂ by 2035, challenging global decarbonisation goals.

However, AI’s potential to cut emissions — through smarter grids, industrial automation and traffic optimisation — can outweigh its footprint, provided the electricity that powers it comes from clean sources.

Cozzi highlighted tools like dynamic line rating, which uses real-time data to assess how much electricity transmission lines can safely carry, boosting grid capacity without new infrastructure.

Eskom already uses AI to monitor its national grid. Nick Singh, a smart grid manager at Eskom, explained that Eskom used AI and machine learning to analyse data related to lightning strikes, fires and earth strikes on the national grid to better predict and mitigate infrastructure outages.

The carbon footprint of AI hinges on the electricity mix. Regions dependent on fossil fuels, like South Africa, will face higher emissions. With Eskom’s grid still dominated by coal, its ability to supply clean, reliable power will be critical in limiting AI’s environmental impact.

The IEA’s Birol urged tech companies to prioritise secure, 24/7, low-emission electricity — favouring renewables, batteries, strong grids and, where necessary, dispatchable sources like natural gas or nuclear. DM