Artificial intelligence will produce millions of tons of electronic waste by 2030

According to a new study, the rapid development of the artificial intelligence sector may significantly increase the amount of global electronic waste in the coming years. According to scientists, if the current situation continues, the amount of electronic waste generated by artificial intelligence will increase 1000 times in 2030 compared to today.

Artificial intelligence’s waste problem is growing

It is estimated that the amount of electronic waste generated by artificial intelligence, which was only 2,600 tons in 2023, may reach 2.5 million tons in 2030. This figure is equivalent to each of the world’s estimated population of 8.5 billion throwing away approximately two iPhones in 2030. The study predicts that more than 5 million tons of waste could accumulate in total by 2030. In 2022, 62 million tons of e-waste were produced globally.

According to analysis, this growth requires frequent updating of the hardware used for artificial intelligence applications, and this quickly makes existing devices obsolete. While unused devices contain metals that are harmful to health, such as lead and chromium, there are also precious metals such as gold, silver and platinum that can be recycled. Scientists state that recycling these metals is an effective way to reduce the amount of e-waste.

The weight of the latest Nvidia platform, designed for large language model inference, training and data processing tasks, shows that generative AI is a significantly “material-intensive sector,” the researchers said in the study. In fact, this was also explained by the CEO of Nvidia, which controls 80 percent of the current artificial intelligence chip market. CEO Jensen Huang says that GPU systems, which weighed 32 kg and 35,000 pieces two years ago, have now reached the level of 600,000 pieces and 1,360 kg.

Chip restrictions between US and China could increase e-waste

U.S. restrictions on the sale of advanced GPU chips to China may lead to the use of older model servers in China, reducing processing efficiency. The study states that the US’s advanced chip model, Nvidia H100, contains features that are not allowed to be exported to China, and the less powerful H800 model can be sold instead. Such limitations lead to the need for more physical servers in China, increasing the amount of e-waste.

The study suggests that switching to a circular economy model to reduce e-waste generated by artificial intelligence can reduce the amount of global e-waste by 86 percent. This model includes extending the life of artificial intelligence hardware, reusing obsolete parts in applications requiring low performance, developing more efficient algorithms and increasing the efficiency of chips.