AI Datacentres Are Surprising Clean Energy Allies

With AI enjoying increased attention this year, the topic of its excess energy usage and greenhouse gas (GHG) emissions has come into the limelight. But despite what some may think, AI datacentres, often with the backing of tech giants, are at the forefront of renewable and clean energy adoption. Investors need not fear AI’s potential energy-intensive needs but should rather consider how this emerging technology may be a tailwind for the renewable energy megatrend.

Key Takeaways

  • AI datacentres, the server farms which train and deploy AI models, are extremely energy intensive, especially with the introduction of large language models.
  • However, datacentres have a consistent track record of improvements in efficiency and have historically managed large increases in demand.
  • Big tech firms are also investing significantly in renewable infrastructure as they offer both an economic and environmental incentive for datacentre operations.

Energy Use – Just How Much?

Recent AI advancements, especially large language AI models such as ChatGPT and Gemini, are trained on variations of graphics cards. These devices are extremely energy hungry, with a single Nvidia H100 chipset (the current industry standard) consuming more energy than the average US household.1 Nvidia sold roughly 1.5 million H100s in 2023 and is expected to ship a further two million units in 2024.2 This implies Nvidia could add 3.5 million household’s worth of energy consumption by the end of the year.

Furthermore, as AI models get bigger, the amount of data they process increases exponentially —which leads to an exponential increase in power consumption as well. GPT 1 was trained on 11,000 unpublished books. GPT 3.5, which was released early last year, used around 175 billion parameters (“parameters” can be anything from books to blogs).3 GPT 4, currently the most advanced iteration, is trained on an estimated 1.8 trillion parameters.4 New AI models will naturally require an increase in GPU power and usage – further contributing to emission fears.

Economics Driving Efficiency: Doing More with Less

The numbers above make it clear that AI pursuits have energy intensive needs, and that concerns around its environmental impact are valid. But tech companies know this, like everyone else, and are working on fixes. This shows up in a number of areas.

First, datacentres overall have had a good track record of improving equipment efficacy. Over the past few decades, increasing internet usage and infrastructure demand invited similar fears of excessive energy consumption. Between 2015 and 2022, internet users almost doubled, and global internet traffic increased by 600%.5 Despite the intense growth in demand, datacentre energy growth remained relatively muted, increasing by a median estimate of only 45% over the period.6

Secondly, the average Power Usage Effectiveness (PUE) (a common datacentre statistic used to measure energy efficiency ) across all datacentres has fallen dramatically. A PUE score of ‘1’ indicates that a datacentre is perfectly efficient with no energy expended outside of system hardware directly driving computation. Over the past 15 years, the average datacentre PUE has fallen by more than 40%, from a score of 2.5 to 1.58 in 2023.7 Today, new datacentres are capable of achieving near perfect efficiency, with examples being Microsoft’s Project Natick – an underwater datacentre which achieved a PUE of 1.07 in 2023, and Facebook’s global datacentres, which average an impressive PUE of 1.11.8

Unlikely Allies in Renewable Energy Adoption

Efficiency is one thing, but growing total energy consumption naturally comes with the threat of increasingly inflated emissions. However, the relationship between energy and emissions can be severed with the use of renewable and clean energy sources.

At first glance, renewable energy does not appear to be the best match for datacentres (AI or not). After all, datacentres are expected to have near 100% uptime to either maintain profitability or fulfil contract requirements. The sun can be hidden by cloud cover, the wind may stop, and nuclear energy is far from widely accessible. So, what’s the benefit of going clean?

It has to do with money. According to CSIRO, renewable energies including wind, hydro, nuclear, and solar, have a levelised cost of electricity (the average net present cost of electricity for a generator over its lifetime) at least 39% cheaper than coal, and 17% cheaper than gas. That figure expands to more than 60% cheaper for coal, and 50% for gas when accounting for carbon capture technology necessary for limiting emissions. More importantly, these figures include ‘firming’ – a technology that helps circumnavigate the issue of renewable energy’s occasional inconsistencies by storing and distributing electricity based on demand fluctuations. So, price is where renewables are showing promise, and is the reason hyperscaler tech companies have invested immense capital into their development.

Examples of big tech going clean have become more and more common. Beyond already being the biggest buyers of green energy Power-Purchase Agreements (PPAs) in the world as of 2023, tech giants have invested heavily in direct renewable energy development projects.9 Google was one of the first big datacentre players to develop onsite solar and wind projects in the late 2000s culminating in a US$2 billion investment in new energy infrastructure in 2019.10 The search engine provider has reportedly been carbon neutral since 2007, and is now aiming to operate on entirely carbon-free energy by 2030.11 Microsoft has committed significant capital to building datacentre batteries to support renewables on the power grid and says the entire company will be carbon-negative by 2030.12 Amazon recently acquired a nuclear-powered datacentre in Pennsylvania for US$650 million as part of its ambition to run all 125+ of its global datacentres on fully renewable energy by as soon as 2025 and reach net-zero emissions by 2040.13,14

AI Could Be A Proponent of Renewable Energy

While the figures of projected emissions and energy consumption by proposed AI datacentres may send alarm bells across the market, it should also be understood that emerging technologies can be drivers of, not detractors from, the global transition towards carbon neutrality. In our view, AI datacentres will not negatively impact the world’s progress towards net-zero by 2050.15 In fact, the AI industry, with its healthy attitude towards energy efficiency and cutting-edge technology, may just become one of the biggest proponents of renewable energies in our clean future.

Related Funds

ATOM: The Global X Uranium ETF (ASX: ATOM) invests in a broad range of companies involved in uranium mining and the production of nuclear components, including those in extraction, refining, exploration, or manufacturing of equipment for the uranium and nuclear industries.

FANG: The Global X FANG+ ETF (ASX: FANG) invests in 10 companies at the leading edge of next-generation technology that includes household names and newcomers.

GXAI: The Global X Artificial Intelligence ETF (ASX: GXAI) invests in global companies involved in AI development, AI-as-a-service, provide AI compute power, or design and manufacture AI hardware.



AI is still an emerging technology, so investing in the space may involve risk. As with all investments, read the relevant PDS and TMD to assess fund-specific risks.

Forecasts are not guaranteed and undue reliance should not be placed on them. This information is based on views held by Global X as at 30/04/2024.

Past performance is not a reliable indicator of future performance

Diversification does not ensure a profit nor guarantee against a loss. Brokerage commissions will reduce returns. This material represents an assessment of the market environment at a specific point in time and is not intended to be a forecast of future events, or a guarantee of future results.