In a study published in Resources, Conservation and Recycling, Alex de Vries-Gao estimates that AI servers could produce between 131 and 224.8 kilotons of electronic waste per year by 2030. This volume is comparable to the total annual e-waste generated by countries such as Denmark, Norway or Austria.

The research builds on De Vries-Gao’s earlier work examining the energy, carbon and water footprint of AI. Previous studies suggested that AI systems might generate up to five million tons of e-waste by 2030. His new estimate is significantly lower.

According to De Vries-Gao, the difference stems from revised assumptions. Rather than assuming unconstrained growth, he estimated how many AI servers can realistically be produced based on available manufacturing capacity and market data. He also suggests that the commonly assumed three-year lifespan for such equipment is likely too pessimistic, as servers may remain operational for longer in practice. This reduces projected annual waste volumes.

Despite the lower estimate, De Vries-Gao cautions against complacency. The absolute volumes remain considerable. He emphasises the importance of extending hardware lifespans, investing in high-quality recycling and improving data collection on equipment production and disposal.

He also argues that large technology companies operating data centres have a responsibility to increase transparency. Without more open reporting, the true scale and impact of AI-related e-waste may continue to be underestimated.