Topic: The 5 million tons only account for the "essence" of AI e-waste (Read 17 times)

When it comes to AI e-waste, many people's first reaction is the pollution caused by internet information, rather than the physical waste itself. However, after seeing a set of statistics, I realized that what I thought was a virtual existence is actually responsible for generating such a massive amount of waste.

The CEO of a leading chip company publicly stated that the GPU platform produced two years ago weighed 31.7 kilograms and contained 35,000 parts. In contrast, the GPU platform produced in 2024 consists of 600,000 parts, weighs 1.36 tons, and is roughly the same weight as a carbon fiber Ferrari. The foundation of generative AI is a range of resource-intensive supercomputers and data centers. This study focuses on the core hardware of AI centers, including the motherboard, graphics processing unit (GPU), central processing unit (CPU), internet communication modules, and uninterruptible power supplies (UPS). Industry reports indicate that the lifespan of these hardware components is typically around three years, and the constant upgrading and replacement result in a significant amount of e-waste. The study does not include peripheral devices from data centers, such as cooling equipment, communication units, or power cables, as these auxiliary parts are difficult to estimate in terms of lifespan. Some can be reused, unlike core hardware which is more likely to be discarded. Moreover, for the sake of rigor, it is challenging to classify these components as e-waste directly caused by generative AI.

While much attention is given to AI's energy and water consumption, the "material consumption" behind it is often overlooked.