A new assessment by Basel Action Network estimates that 395 million to 617 million tonnes of AI-related electronic equipment could be retired between 2025 and 2050.
By 2030, around 8.6 million to 13.1 million tonnes of AI-linked equipment could become waste every year, according to the assessment.
BAN says earlier estimates focused mainly on servers and GPUs, while its assessment includes the wider physical infrastructure of AI data centres, including power, cooling and networking systems.
The report raises concerns over hazardous e-waste, shortened equipment lifespans, data security risks and the possibility of discarded AI equipment moving to countries with weaker recycling oversight.
The rapid expansion of artificial intelligence, or AI, could create a new e-waste challenge for the world. Servers, GPUs, networking equipment, power systems and cooling infrastructure used in AI data centres could turn into large quantities of discarded equipment in the coming years.
According to a new assessment by the environmental organisation Basel Action Network (BAN), between 2025 and 2050, 395 million to 617 million metric tonnes of electronic equipment associated with AI could be retired.
BAN says this estimate is considerably larger than previous estimates. One reason is that earlier assessments focused mainly on equipment such as AI servers and GPUs. The new assessment includes the entire infrastructure of AI data centres.
According to BAN, by 2030, around 8.6 million to 13.1 million tonnes of AI-related electronic equipment could become waste every year.
Discussion of the environmental impact of AI is often limited to electricity and water consumption or the number of servers and GPUs. But BAN's assessment suggests that the picture is much larger.
In an AI data centre, servers and GPUs, or accelerator equipment, account for only around 13 per cent of the total equipment by weight. The rest includes networking equipment, power supply and distribution systems, cooling systems and backup power infrastructure.
This means that as AI capacity expands, it will not be only computing chips that eventually become waste. The entire physical infrastructure required to operate them could gradually become part of the e-waste stream.
BAN believes that the rapidly changing technology of AI equipment could further increase the problem. When newer and more powerful chips become available, older equipment may not necessarily be defective, but it could become less useful for larger AI systems.
The race to increase computing capacity in the AI sector is bringing new equipment to the market at a rapid pace. As a result, older equipment may need to be replaced before it reaches the end of its functional life.
This could increase the volume of e-waste and also raise an important question: what happens to the old equipment?
Reusing or repairing the equipment, or recovering valuable materials from it, could reduce the demand for new resources. But if the equipment is simply discarded as waste, the pressure on the environment could increase.
E-waste can contain lead, mercury and other hazardous substances. If improperly handled, these materials can pose risks to soil, water and human health.
The problem of old equipment generated by AI has another dimension: data security.
Servers, storage equipment and networking systems removed from data centres may contain confidential corporate information, details about system architecture, login information and other sensitive data.
Therefore, removing a piece of equipment from a data centre is not the end of the security process.
According to Linda Lee, Chief Strategy Officer at IT asset disposition company Re-Teck, every piece of equipment entering the reverse supply chain should first be viewed from a data-security perspective.
She says companies need to ensure that equipment remains under secure supervision, that data stored on it is erased through proper procedures, and that records of the entire process are maintained.
As the volume of old equipment generated by AI increases, the question will not only be how it is recycled. It will also be important to know where this equipment ultimately ends up and how it is handled there.
BAN has long monitored the international movement of electronic waste and its improper disposal. The organisation has also conducted GPS-based studies to track the movement of old electronic equipment.
This issue could become even more important in the case of old AI equipment. If large quantities of old servers and other equipment end up in countries where recycling oversight is weak, the risks to both the environment and human health could increase.
In addition, protecting the data contained in such equipment could remain a challenge.
India is already witnessing a rapid increase in the volume of e-waste. As AI data centres and related infrastructure expand in the country, AI-related e-waste could become an important issue in the future.
It will be particularly important to see what happens to old servers, GPUs, networking equipment and other electronic goods after they are taken out of service.
Are they reused? Are their valuable components recovered? Are they safely recycled? Or do they end up in waste markets where adequate safeguards for the environment and workers are absent?
The answers to these questions will be important for understanding the actual environmental impact of AI.
So far, discussions about the environmental impact of AI have focused mainly on the electricity and water requirements of data centres. But BAN's new assessment shows that it is necessary to look at the entire life cycle of AI.
The manufacture of equipment required for AI involves the use of raw materials and energy. The equipment is then installed in data centres and, after a few years, replaced as newer technology becomes available. It then goes through reuse, repair, recycling or disposal.
If any stage of this chain is overlooked, the environmental footprint of AI cannot be assessed accurately.
BAN's estimate suggests that between 2025 and 2050, hundreds of millions of tonnes of AI-related equipment could be retired. The scale is so large that it is necessary to start thinking now about how this waste will be managed safely.
