What awaits us: 1,000 times more e-waste by AI

The artificial intelligence It has arrived to completely revolutionise our lives. The fact is, it is not simply a piece of technology that brings us new features. Far from it: AI has a number of implications, including the environmental impact. A recent study published in the journal Nature Computational Science points out that The amount of electronic waste it generates will increase a thousandfold in a decade.

The reality is that major technology companies such as Google, Microsoft o Goal are locked in a race to claim the throne of generative artificial intelligence, rolling out new tools at breakneck speed whilst improving existing ones. There is no end in sight to this race, so, if this estimate is correct, we will have a huge environmental problem to manage.

Are you interested in this topic? Let’s have a closer look!

Between 1.2 and 5 million tonnes of electronic waste

We already know that the the environmental footprint of generative AI contains a high energy consumption, carbon emissions and a excessive water consumption. Added to all these consequences is the huge amount of electronic waste that AI is expected to generate. It is important to bear in mind that the equipment used in data centres needs to be replaced on an ongoing basis. This results in digital waste with compounds such as lead or chromium.

For this reason, the journal *Nature Computational Science* is clear on this point: if nothing is done to reduce electronic waste, will be multiplied by 1,000 by 2030, arriving between 1.2 and 5 million tonnes. How can we reduce all the waste generated by large language models (LLMs) such as ChatGPT, Copilot, Gemini, Claude Or is it ‘Llama’? Researchers draw on the principles of circular economy.

What did this study on electronic waste generated by AI involve?

For this report, the following was used as the basis for the calculations: the an Nvidia DGX H100 server with eight GPUs, in 2023, the most popular hardware for AI processing. The researchers therefore developed four future scenarios to estimate the growth of AI and the demand for data processing between 2020 and 2030. Let’s take a look at them:

• The first scenario assumes that there will be limited growth in the chip industry, putting it at a 41 %. To arrive at this figure, they based their calculations on the growth they had already achieved between 2022 and 2023, concluding that it is impossible for growth to be any lower.
• The second scenario takes us to a conservative scenario, in which growth stands at 85 %. In this case, they are based on the growth figures achieved by voice assistants such as Alexa.
• The third scenario was labelled “moderate”, inspired by TikTok’s figures. Growth, in this case, amounted to as much as a 115 %.
• The fourth scenario points to aggressive growth of a 136 %, Given that AI is set to become a companion to people in their daily lives. In this context, Facebook’s growth rate was used as a benchmark.

How much electronic waste is being generated?

We now have the figures for 2023, a year in which a total of 2,600 tonnes of electronic waste. According to this study, by 2030, This figure will amount to between 0.4 and 2.5 million tonnes. The conclusion is that there will be 1,000 times more waste, which would be equivalent to discarding between 2,100 and 12,300 million iPhone 15 Pro models.

We are talking here about electronic waste generated by generative AI. However, whilst all technological waste in 2002 totalled 4.6 million tonnes – including laptops and tablets – by 2030 this figure could reach 43.3 million tonnes. Obviously, this study does not take into account the implementation of measures to reduce digital waste during the current decade.

Nor did this study include the restrictions on semiconductor imports, a key product. Why is it important? Because it could lead to a 14 % increase in the number of generative AI servers reaching the end of their useful life, which would leave us with an additional 5.7 million tonnes of waste,

Finally, it is worth noting that Digital waste would be concentrated in Europe (14 %), East Asia (25 %) and North America (58%). Despite these figures, virtually all of the waste would end up being sent to countries in Africa and Asia.

What can be done to reduce this electronic waste?

As we mentioned at the start of this post, this study is based on the the circular economy to limit the waste generation. At present, the most effective approach is to extend the lifespan of the hardware; it is estimated that 62 % of the AI servers that are discarded each year could be kept in service for a further year. As a general rule, their lifespan is three years, so this would be extended to four.

Another factor that could reduce the amount of electronic waste is the reuse of certain parts of GPUs, such as communication, memory and battery modules. This approach would reduce electronic waste by 42 %, although it may not be entirely feasible given the difficulty involved in recycling these types of components.

Another significant finding to emerge from the study is that, although the equipment contains toxic materials such as lead, chromium and acrylonitrile, it also contains precious metals such as gold, silver, platinum, nickel and palladium. If these were recycled properly, they could be worth between 14,000 and 28,000 million dollars.

So, what do you think the situation will be like in 2030? We look forward to hearing from you on our social media!