Every time we ask an AI to write an email, generate a piece of art, or analyse a complex dataset, we are engaging with one of the most sophisticated supply chains ever created. While AI feels “weightless”—existing in a cloud we cannot see—its foundation is built on silicon, electricity, and massive amounts of physical hardware. As we rush toward an AI-driven future, a critical question looms: can the planet afford the hardware required to power our intelligence?
For several years, the organisation Interface has provided a vital lens into this hidden world through its “Semiconductor Emissions Explorer.” This tool allows us to peer inside the factories of the world’s leading chipmakers to assess their true carbon footprints. For a moment, the outlook was bright. Between 2021 and 2023, direct “Scope 1” emissions (the pollution created during the manufacturing process itself) fell significantly. As researcher Julia Hess explained to us in May 2025, the industry seemed to be mastering the art of “green” electronics.
However, a newly published study by Hess and Maria Nowicka suggests that this era of progress has hit a sudden, sharp turning point. The ongoing AI boom has triggered a manufacturing frenzy that threatens to erase years of sustainability gains.
The researchers point to two main culprits: the high-powered Graphics Processing Units (GPUs) that act as AI’s “brain,” and High-Bandwidth Memory (HBM), the specialised tech that allows data to move at lightning speeds. While HBM was a niche product in 2024—accounting for only five per cent of the total memory (DRAM) market—it is becoming the new industry standard.
The data from 2025 and 2026, which has yet to be fully released by manufacturers, is expected to confirm a sobering reality: our demand for smarter software is resulting in a much heavier environmental footprint.
While the hardware itself is the starting point, its final home—the data centre—is where the impact multiplies. In our special feature on the sustainable development of AI, we look at why these facilities are so different from the internet infrastructure of the past. Unlike traditional cloud computing, AI data centres are characterised by extreme energy requirements, an insatiable thirst for fresh water to cool overheating chips, and a much faster rate of component wear and tear.
Semiconductor Emissions Explorer now on new microsite
Hess also pointed us towards a new microsite that simplifies access to the Semiconductor Emissions Explorer. The tool provides both an overview of aggregated data and specific datasets for individual semiconductor manufacturers.
We will return to this tool next year at the latest. As soon as the CSR (Corporate Social Responsibility) data from the major manufacturers becomes available, we will be able to clearly see the link between the AI boom, HBM memory, and the increased emissions within the semiconductor industry.
Original article from 21st of May 2025
The coronavirus pandemic brought semiconductors into focus. Supply chains halted, and demand surged. This led to a shortage of these vital components. Europe and the USA faced a critical lack of chips.
Semiconductors power all our tech devices. Companies like TSMC, Samsung, and UMC produce them. Most of this production occurs in South Korea and Taiwan.
The industry has growing geopolitical importance. Yet, its environmental cost is often overlooked. Semiconductor production uses immense energy and requires vast amounts of water. High-purity materials are essential. Their extraction is costly and harms the environment.
Interface, formerly Stiftung Neue Verantwortung, investigate the industry’s environmental impact. With the Semiconductor Emissions Explorer, Interface has, for several years, offered a comparison tool for the carbon footprint of various semiconductor manufacturers. In a newly published study, the Interface team is now investigating how the ongoing AI boom could affect the industry.
A tool that compensates for insufficient data
Julia and her team developed the Emission Explorer after encountering a problem during their research into the environmental impact of semiconductor production. “A great deal of sensitive data and business secrets are not disclosed to the outside world. The information we were able to work with actually only comes from annual reports and social responsibility reports,” Julia told us during our video call (via semiconductors!). “All other information is difficult to access in expensive datasets.”
The team tackled this shortcoming with a pioneering spirit: “We wanted to think about what we could do with the data we have.” The “Semiconductor Emission Explorer” has resulted in a tool that gives politicians, researchers and companies more decision-making power. Julia says:
“The Explorer can be used to see how high the emissions [… of various semiconductor manufacturers …] are and can be used to create a political framework. It also helps companies in Asia to assess what it would be like to relocate to Europe or cooler countries.” This tool also benefits scientists. It highlights the energy and water consumption of different semiconductor types, as well as their resource intensity.
AI processors, memory chips and more—what’s this all about?
To understand why a central dataset is crucial, we first need to look at semiconductor production. Technically speaking, semiconductors are materials whose conductivity falls between an insulator and a conductor. Through complicated processing of materials like silicon or germanium, their conductivity can be controlled by changing temperature or electrical fields. This property is vital for modern chipsets in devices such as computers and smartphones.
This simplified explanation contains two important details for the topic of sustainability. First, we need materials like silicon and germanium for their production. Second, we have to adapt them in a technically complicated way to make them usable for modern chipsets.
What is a wafer disc?
A brief digression for nerds:To produce modern chipsets in large quantities, semiconductor manufacturers cut high-purity silicon ingots into thin slices and then polish them.In the front end, processes such as cleaning with ultra-pure water, doping, photolithography and etching can be used to create complex structures and circuits on these extremely smooth surfaces.The wafer slices are then cut up and other components are soldered onto the silicon plates – until a finished computer chip is finally created.
Image source: Intel Corporation
Julia Hess describes processors as highly complicated semiconductors. These are found in every smartphone, notebook and modern car. They also exist in data centres on a massive scale. “The problem is that the energy consumption during production is very high. The production of a modern semiconductor can take up to three months and consists of cooling processes and automated procedures. A wafer disc also has to be repeatedly cleaned with ultra-pure water, which also leads to increased water consumption.”
How can we manufacture semiconductors more sustainably?
Although the production of semiconductors is complicated, the demand for chipsets continues to rise, according to the results of Julia and her team’s study. “On the one hand, this is due to the AI boom of recent years,” says the researcher. “But there are also other influencing factors. Autonomous weapon systems require powerful processors, as do drones, the field of robotics and the defence industry in general. They don’t always need the latest chips, but demand is increasing. Developments such as the electrification of mobility are also dependent on semiconductors.”
However, the research team’s study revealed a critical point: “69 percent of energy consumption is attributable to five companies that manufacture logic and memory chips for AI and consumer electronics.” This finding underscores the urgent need to produce computer chips more sustainably.
Emissions from the semiconductor industry fell sharply after 2021. However, it will be interesting to see what impact the AI boom has had since 2023. Emissions from the semiconductor industry have fallen sharply since 2021. However, it will be interesting to see what impact the AI boom has had since 2023.
Hess believes that a new miracle solution using artificial intelligence, as predicted by some experts, is unlikely. However, as in other sectors, sustainable practices in semiconductor production have long been known:
“We need better recycling, a functioning circular economy, the use of renewable energies, the recycling of water and a consistent consideration of how we can reuse waste from production in other industries.” None of this is being done with the necessary seriousness—and companies currently have too few incentives to do so.
“Companies that operate in Asia hardly get any pressure. The entire industry is aware of this and yet the focus is on short-term solutions and less on long-term changes.”
Julia Hess developed the Semiconductor Emission Explorer with her team. Thank you for the interview! The “Semiconductor Emission Explorer” is therefore a guide for precisely this purpose. With transparent data and better comparability, companies and politicians should be able to make more sustainable decisions. According to Julia Hess, a current example shows that this is urgently needed.
AI Continent Plan emphasises the need for reliable data
With its “AI Continent Plan”, the European Commission is striving to become a global leader in the field of artificial intelligence. The strategy, which was only presented in April 2025, creates economic and technological framework conditions for the development and long-term application of European AI.
Part of this plan is the establishment of five AI gigafactories, which will provide developers with the technical basis for the development of a European AI or a European LLM. Each of these data centres will contain around 100,000 AI chips. Julia Hess cites this plan as an example of the lack of a long-term strategy with a sense of proportion in technology policy:
“In many places, there is a lack of coordination of strategies in the semiconductor and AI sectors to align objectives and create synergies. Perhaps we should think about this first: Why do we need gigafactories?” On the one hand, building them would swallow up a lot of money, which in turn would not be available for sustainability projects. On the other hand, we would need semiconductors that are currently not produced sustainably and locally – and which may already be obsolete after a few years.
How can we ensure a green digital future?
Growing e-waste, carbon emissions from AI, data centre water usage—is rampant digitalisation compatible with a healthy planet? Our latest project explores how digital tools and services can be developed with sustainability in mind.
