Portuguese 
English 
Spanish 
5 min of reading 
0 comments 
Study estimates that generative AI could add up to 5 million tons of electronic waste by 2030, with impact concentrated in data centers and discarded hardware.
According to Nature Computational Science, a study published on October 28, 2024 calculated for the first time the volume of electronic waste generated by generative artificial intelligence by 2030. Led by Peng Wang from the Chinese Academy of Sciences, in collaboration with researchers from Reichman University in Israel and institutions in the United Kingdom, the research estimates that the expansion of AI could add between 1.2 and 5 million metric tons of electronic waste to the planet between 2020 and 2030.
According to the study, this calculation only considers the hardware directly involved in AI computing, including servers with GPUs, CPUs, memory and storage modules, communication systems, and power supplies. Within this volume are 1.5 million tons of printed circuit boards and an additional 500,000 tons of data center batteries, with geographic concentration mainly in North America, East Asia, and Europe.
AI electronic waste grows because GPUs become obsolete in about three years
According to Nature Computational Science, the reference lifespan of three years for AI servers was not chosen by chance. The issue is not that a data center GPU stops functioning quickly, but that it becomes economically obsolete long before the end of its physical durability.
-
11 African countries are racing against the advance of the Sahara with an 8,000 km green wall, billions of dollars in international support, and the nearly impossible mission of restoring 100 million hectares before the desert turns fertile lands into dead zones.
-
It seemed like science fiction, but it is already being tested: scientists want to spray salt into the sky to turn clouds into giant mirrors, reflect sunlight, and try to cool the planet before global warming gets even more out of control.
-
Australia removes 1 million tons of rock under Sydney, weight compared to almost 200,000 school buses, to open a tunnel under the harbor and create a new underground crossing in one of the most famous cities in the world.
-
In the waters of the Coral Sea, Australia, researchers have described a new species of ghost shark, a fish with a prehistoric appearance that lives in the depths.
A high-performance GPU can continue operating for 10 to 15 years if well maintained. But the semiconductor industry releases new generations every two or three years, with significant performance gains per watt. For companies like Microsoft, Google, Meta, and Amazon, keeping older chips means spending more electricity to do the same work.
In a sector where electricity can represent between 40% and 60% of the operational cost of a data center, early replacement makes financial sense even when the hardware continues to function. The result is an accelerated disposal treadmill fueled by the very race for efficiency in AI.
AI Hardware is Difficult to Recycle and Concentrates Toxic and Strategic Materials
According to Nature Computational Science, an AI server is not comparable to a regular laptop. It brings together highly specialized components designed for maximum performance, not for simple disassembly or efficient repurposing.
The printed circuit boards concentrate materials such as copper, tin, lead, gold, palladium, platinum, beryllium, and indium compounds, combined in complex structures of fiberglass and epoxy resin. These materials have real economic value, but separating them requires expensive, complex, and potentially hazardous metallurgical and chemical processes.
According to the research, large-scale GPUs further elevate this challenge. Advanced chips like those used in AI data centers employ 3D packages, advanced metal alloys, and encapsulation materials that the global electronic recycling chain has not yet been prepared to process on a large scale.
78% of Global Electronic Waste Already Goes to Landfills or Informal Recycling
According to Nature Computational Science, AI waste does not arise in isolation but within a global electronic waste crisis that was already out of control before the explosion of generative artificial intelligence. In 2022, the world generated 62 million tons of electronic waste, an increase of 82% since 2010.
The growth rate of the problem is five times greater than the installed capacity of formal recycling. According to data cited by the research, about 78% of the world’s electronic waste goes to common landfills or informal recycling, often in locations in Africa and Asia, where workers dismantle equipment without adequate protection and release heavy metals and toxic compounds into the environment.
In this scenario, AI waste tends to follow the same path, with the aggravating factor of containing strategic materials for the semiconductor industry itself, such as indium, gallium, and germanium, which may end up buried or burned while the global chain pays billions to extract them in other regions.
Circular Economy Could Cut AI Electronic Waste by Up to 86%
According to Nature Computational Science, the study not only quantified the problem but also simulated solutions. The most important result is that a combination of circular economy strategies could reduce AI electronic waste generation by up to 86%.
The first strategy is to extend the lifespan of hardware in its initial use. Instead of replacing servers every three years, companies could redeploy them to less intensive tasks within the same data center. A system that is no longer suitable for training large models can still perform well in inference, a less demanding phase of AI usage.
The second is the remanufacturing and reuse of components, such as memory and storage, which alone could reduce waste by 42%. The third is design for recycling, designing boards and components from the start to facilitate disassembly, material recovery, and industrial reuse.
AI industry still doesn’t solve the problem due to lack of regulation and excess competitive pressure
According to Nature Computational Science, companies like Microsoft and Google have already announced public goals of zero net waste and zero net emissions by 2030, but electronic waste from data centers rarely appears in reports with the same detail given to carbon and energy.
In the United States, Senator Ed Markey introduced in February 2024 the Artificial Intelligence Environmental Impacts Act, a bill that sought to require federal agencies to study the environmental impacts of AI, including electronic waste. The proposal only provided for voluntary reporting and did not advance in the Senate.
The absence of rules creates a direct incentive problem. A company that replaces its servers more slowly to reduce waste risks operating at a higher cost than a competitor that replaces its equipment more aggressively to maintain the best efficiency. Without common regulation, competitive pressure pushes the sector in the opposite direction of the solution.
Explosion of generative AI turns discarded hardware into a new front of the environmental crisis
According to Nature Computational Science, the baseline in 2023 was only 2,600 tons of electronic waste per year specifically linked to AI hardware. In just seven years, this number could grow up to a thousand times in the most aggressive adoption scenario.
This means that generative artificial intelligence not only brings an increase in electricity consumption and carbon emissions. It also creates a new front of environmental pressure based on discarded GPUs, data center batteries, printed circuit boards, and toxic or critical materials for the tech industry.
The central warning of the study is clear. The AI race is being discussed as a competition for computing power, productivity, and technological leadership, but it is already beginning to produce a mountain of electronic waste that the global recycling infrastructure does not yet know how to absorb.
Be the first to react!