Studies show the advancement of generative AI will put a strain on the waste stream.

Researchers from Cambridge University and the Chinese Academy of Sciences have published a paper in the journal Nature warning that the rapidly evolving technology of generative artificial intelligence (GAI) can lead to compounded amounts of e-waste equivalent to more than 10 billion iPhones per year by 2030.

In 2022, 62 million metric tons of e-waste were produced worldwide, according to the Global E-waste Monitor, a waste stream growing 5X faster than recycling programs designed to deal with it. Meanwhile, according to Deloitte, 80% of private companies expect AI to drive their business in the next 3 years.

The paper considers only large language models (LLMs) – which are a type of program used to generate and interpret human language – and not other forms of GAI such as visual arts tools and image generators or industry-specific solutions.

Study coauthor Asaf Tzachor, a sustainability and climate researcher at Reichman University in Israel, explained that the team focused on LLMs because they're among the most computationally intensive, "Including other forms of AI would increase the projected e-waste figures."

Artificial intelligence – from streamlining administrative tasks and equipment automation to predictive medicine and cybersecurity – is an increasingly valuable tool affecting a growing number of industries, and while its effects are only beginning to be demonstrated, the lesser-explored ideas of its cost in energy and materials have begun to come into focus.

Beyond the power and computing requirements for generative AI, specifically in software producing chats, conversations and written material, this rapidly expanding industry could produce an unmanageable e-waste stream through the constant replacement of servers, hardware upgrades, temperature-controlled building construction and enough wiring to lasso the moon – all necessary physical materials involved in the life cycle of this new and exciting field.

Most research on sustainability in the field of AI has focused on energy, resource use, and carbon emissions. Tzachor's study is intended to provide an estimate of the potential scale of the problem and spur companies to adopt more sustainable practices rather than dissuade them from using the technology.

Research suggests that adoption of large language models (LLMs) alone will generate 2.5 million tons of e-waste per year by 2030. The study aims to provide initial estimates that highlight the potential scales of the forthcoming challenge and explore potential solutions for a more circular economy.

The study details four potential scenarios for generative AI adoption ranging from limited to aggressive expansion and projects potential e-waste expansion from a 2023 starting metric of 2.6 kilotons per year. However, much of today's AI computing infrastructure has rapidly deployed over the last two years, demonstrating the approximate e-waste kilotons were mostly generated before the AI boom took off.

Limited expansion of AI is estimated to generate a total of 1.2 million tons of e-waste by 2030; aggressive use would result in a total of 5 million tons over the same period. Tzachor notes that given current trends, the aggressive scenario is more likely. The world will see a sharp uptick in e-waste figures when this first outlay of infrastructure reaches end-of-life.

"AI doesn't exist in a vacuum; it relies on substantial hardware resources that have tangible environmental footprints," said Tzachor. "Awareness of the e-waste issue is crucial for developing strategies that mitigate negative environmental impacts while allowing us to reap the benefits of AI advancements."

Mitigation

Big-name tech companies, including Amazon, Google, and Meta, have announced sustainability goals that focus on carbon footprints. Microsoft has specifically pledged to reduce e-waste from its data centers. However, Tzachor encouraged regulation to ensure adherence to the best practices around AI e-waste, saying, "Companies should have incentives to adopt these strategies," he says.

Keeping up with the latest advancements by way of upgrading GPUs, CPUs, and other electronic equipment in data centers as newer, more advanced chips become available, researchers project, will lead to an explosion in the production of electronic waste. But Tzachor isn't against advancement, focusing instead on how the waste stream is managed.

More advanced technology, such as advanced computer chip design, as opposed to the cash-grab of planned obsolescence, can help server farms do more and generate less waste.

However, each upgrade should be made thoughtfully, as it will result in a net increase in the waste stream. Given current trade restrictions on semiconductors, upgrading is not always an option. A one-year delay in upgrading to the latest chips will result in a 14% increase in e-waste, according to the study.

Mitigation of the electronic waste stream includes reuse; outdated servers can be repurposed for hosting websites or doing more basic data processing tasks, or they can be donated to educational institutions, which could reduce the waste load anywhere from 16 to 86%.

And finally, appropriate recycling infrastructure needs to match the exponential growth of the e-waste stream. Reports like this are necessary to alert policymakers to the dangers of sitting on their hands when it comes to implementing a circular economy, which includes incorporating waste streams back into supply chains.

With luck and likely the help of AI itself, solutions will be forthcoming.