By Justin Heyes. 12th June 2024

In a recent article in the Guardian “The ugly truth behind ChatGPT: AI is guzzling resources at planet-eating rates” the central focus of the piece was the issue surrounding energy consumption within the datacenter industry and the impact this is having upon the environment. Nothing new for those in the industry, but perhaps an eye-opening piece for the general public about an age-old challenge faced throughout the history of the industry.

While the article did a very comprehensive job of explaining the extent to which datacenter advancement and expansions are impacting the climate, what seemed to be absent from the piece was the reason for the expansion, how the dawn of generative AI has set a race against the clock for datacenters to manage to reduce the impact of its energy requirements, or even how the technology itself has now become so ingrained in so many industries that our digital lives would not function without it.

Expanding Capacity Vs. Reducing Impact

There is already a global effort to reduce society’s impact on the environment. Across the world, each country has committed and outlined their ESG targets for the future. In Malaysia, the National Energy Transition Roadmap (NETR) sets ambitious targets for Malaysia, aiming to achieve net-zero emissions by 2050. The plan is comprehensive and outlines a gradual increase in renewable energy shares, targeting 31% by 2025, 40% by 2035, and an impressive 70% by 2050 (MIDA, 2024), and looks at increasing renewable energy capacity, green transportation, bioenergy, carbon capture and utilising hydrogen as a fuel source while implementing energy efficiency standards across industries.

Where this comes into conflict with the digital economy is that recent developments such as generative AI, having seemingly limitless potential in its applications for enhancing productivity, efficiency and allowing for processes that simply were not possible before, has had an unprecedented uptake and consequently caused a significant increase in the need for new datacenters fitted with AI racks to handle the sheer volume of data that needs to be stored and processed.

The issue isn’t necessarily the increase in datacenters, the industry has been continually improving new builds to be more energy efficient, since the lessons learnt from the dot-com boom in the 90’s, the real problem has come from the need now for new datacenters capable of processing a new type of data, which demands a much greater amount of power to process and cool the racks optimally, as was explained in a summary from a report by Newmark this year “Fundamentally supporting accelerating AI/ML adoption requires more power and cooling than much of the existing datacenter inventory can accommodate”.

While new AI racks are capable of truly incredible processing power, previously only dreamt of, the increase in ability has come at a steep cost. Traditional racks typically operate at 10kW per rack, whereas their newer AI counterparts typically operate at 32kW per rack. So when calculating the percentages for ESG targets, the concern then becomes the total power actually needed. While the Datacenter industry has progressed rapidly over the past three years, innovating to reach new heights in digital progression for society, the same cannot be said for the advancement of renewable energies, which by comparison look stagnant.

The Realities Of Renewable Energy

Solar

On the face of it, especially in tropical climates such as those found across Southeast Asia utilising solar farms as a source of renewable energy seems like a viable solution. However, there are factors which, when taken into consideration, may give cause to reconsider.

For one, solar farms have one of the largest footprints of all of the renewable energy sources – while it has been observed that large-scale solar ventures will allow palm oil producers to generate up to 54 times more operating profits per hectare compared to palm oil (The Edge, 2024) – this would also be dependent on the energy produced.

Solar is incredibly susceptible to changes in the weather and other factors, which can make its output highly variable. While studies will tell you solar cell efficiency can reach 42%, this is measured under laboratory conditions, and does not equate to the efficiency of solar panels (modules) as a system, which is often around 15-20%, frequently dropping below this measurement owing to sub optimal tilting, overcast weather, exterior objects blocking sunlight and a number of other conditions (GreenMatch, 2024). 

Wind

Wind as a renewable energy is even more subject to variable output. Owing to the generation of power coming from wind turning the propellor-like blades of a turbine around a rotor, to spin a generator, if the conditions aren’t sufficient to turn the blades, you have a somewhat redundant investment.

While it is estimated by Tenaga Nasional that 500 to 2000 MW worth of electricity could be generated from wind energy in Malaysia, wind farm locations are definitively pre-set and so the capacity for building them is limited to where the wind blows, for want of a better term.

Hydrogen

According to the World Economic Forum, hydrogen could be one of the key drivers in the world’s transition to clean energy.  Hydrogen, the simplest known element in the universe, may very well be an essential source of clean energy in the years ahead. 

But hydrogen isn’t necessarily ‘green’ – in fact, most hydrogen is produced from fossil fuels, in particular, natural gas – green hydrogen currently accounts for less than 1% of total annual hydrogen production. 

Green hydrogen is made from renewables and relies on a process called electrolysis – the process of using electricity to split water into hydrogen and oxygen, but as one can imagine in an industry where water is heavily used for cooling, and not a resource which can be produced, creating the power for datacenters from the same resource could be seen as burning the candle at both ends and is not so efficient to be used broadly as an energy source yet. (Earth.Org, 2024)

Hydroelectric Power

Globally, around half of hydropower’s economically viable potential is untapped. The potential is particularly high in emerging economies and developing economies, reaching almost 60%. Over the life cycle of a power plant, hydropower offers some of the lowest greenhouse gas emissions per unit of energy generated (IEA, 2024).

The catch is in identifying sources suitable for hydropower generation. In Malaysia, there are viable options across Perak which were initially developed under British rule, but there has been hesitancy in updating or restarting these as they would also carry additional investment costs, to send the generated power to where it most needed.

The Nuclear Option

Despite the presumptions many people still have surrounding nuclear energy, the option is not only viable but has been presented as an essential component in the decarbonization strategy of many countries around the world. In China there are currently 150 reactors planned for development in the coming years (Orano, 2024).

Unlike many renewable energy sources, power from nuclear energy can be generated 24 hours a day and isn’t dependent on the weather, like wind and solar power tend to be. Nuclear energy is referred to as a clean energy technology as it produces nearly zero carbon dioxide or other greenhouse gas emissions. Nuclear energy also avoids producing air pollutants that are often associated with burning fossil fuels for energy. (nationalgrid, 2024)

In the case of powering datacenters, particularly in the US, we can see existing models that show a way to effectively supplement the increase in energy consumption, while maintaining ESG targets and benefiting from the ability to be set up and developed wherever needed.

AI As Its Own Solution

AI can be utilised to reduce its own impact and make other areas of society greener. Perhaps one of the most commonly discussed examples of reducing impact, would be Google’s use of DeepMind where the AI was able to consistently achieve a 40% reduction in the amount of energy used for cooling, equating to a 15% reduction in overall PUE overhead after accounting for electrical losses and other non-cooling inefficiencies (Google, 2016). 

Green energy can also benefit from the real-time analytics capabilities of AI. In a recent study it was shown that AI-driven solar energy management systems can increase energy yields by up to 25% while reducing operational costs by 30%. By leveraging AI algorithms, for panel placement, to minimise shading, and ensure optimal tilt angles solar can be optimised for maximum energy capture (GreenMatch, 2024).

And this can be further extended when looking at automating daily life and creating smart cities. For example, in Hong Kong, generative AI uses information from sensors on street lights, to analyse automotive and foot traffic, predicting peak times with an accuracy that allows better management of commuters, resulting in less traffic, and optimisation of trains in service. This kind of data processing can be applied to more efficiently power homes and has more recently been utilised in bakeries to reduce food wastage by accurately predicting peak times for demand and even how many of each item to bake, with remarkable accuracy. 

The Future Will be Green, But Will Take Time

There are challenges ahead not just in Malaysia but globally as the datacenter industry comes to terms with the rapid increase in demand for capacity and processing power needed to enable the use of generative AI. 

But balancing datacenter expansion with the impact it will have on the climate, requires a balanced and nuanced approach. Generative AI is a necessity for the world we live in today, this is an immutable fact. Datacenters are required to enable generative AI, so we cannot simply stop building them or the world as we know it would grind to a halt. 

While the datacenter industry should be open about the extent to which power consumption will increase, and there should be concerted efforts to innovate ways to reduce this, there needs to be a wider conversation between governments, developers and energy providers in investing and advancing renewable energy sources to bring up the available options to fuel the expansion.

However, investing in the furthering of green energies to expand their ability to power datacenters, developing new technology to reduce the impact on resources from datacenters, a wider understanding from the public of the impact of their digital lives, and strategic green targets will all take time and serious investment. 

This needs to be understood, because the alternative “quicker” solution is to reduce the need for datacenters, by reducing the use of generative AI, and the only way to do that is to remove it entirely from public use, and nobody would accept that.