The Ghibli-style image trend recently took over social media. Everyone was captivated by those doe-eyed, whimsical versions of themselves created by AI—perfect for the dopamine rush of likes and shares.
It was cool, right? Well, not literally. Imagine clicking “generate” for your Ghibli-inspired image, expecting a magical masterpiece. In milliseconds, voilà! But behind that charm, there’s a sweltering story—GPUs overworking, battling heat, and data centres buzzing like bees. Creating one image consumes more energy than sending a dozen emails or making a quick Google search.
The demand was so high that even OpenAI CEO Sam Altman had to ask people to chill, as their GPUs were melting. It’s fascinating how quickly trends spread, but behind the fun lies a hidden cost—carbon footprints piling up like invisible scars on our planet.
Power Hungry AI
AI chips, such as GPUs, use more energy and produce more heat than CPUs. Each time you click “generate,” you’re not just creating art; you’re working thousands of processors hard, causing them to heat up like they’re in a sauna.
A study by AI start-up Hugging Face and Carnegie Mellon University found that generating one AI image uses as much energy as fully charging your smartphone, while 1,000 text generations use just 16% of a full charge. Just imagine the impact from millions of people together generating Ghibli-styled images just for fun or trends!
A report by the Internal Energy Agency predicts that energy consumption associated with AI, data centres, and cryptocurrency will be equivalent to the total energy usage of entire Japan. Generative AI alone will consume 10 times more energy in 2026 than it did in 2023.
Researchers say that a single query on ChatGPT uses 5 times more electricity than a simple web search. Google’s BERT energy usage for training was equivalent to a round trip of transcontinental flight. Even post-training these AI models consumes a lot of power.
Massive Carbon Footprint
Image Credit: Frontiers of Environmental Science & Engineering
AI has a huge carbon footprint due to its high energy consumption. Data centres which operate 24/7, rely heavily on fossil fuels, contributing to 2.5 to 3.7% of global greenhouse gas emissions—even surpassing the emissions of the aviation industry.
In a 2019 study by researchers at the University of Massachusetts, Amherst, conducted a life cycle assessment of training large AI models, focusing on natural-language processing (NLP). They found that training a single deep learning model emits around 283 tons of CO₂—equivalent to 300 round-trip flights between New York and San Francisco or the lifetime carbon footprint of five cars.
Researchers say creating ChatGPT-3 consumed 1,287 megawatt-hours of electricity and generated 552 tons of CO₂, roughly the same as 123 petroleum-powered cars driven for a year. The energy needed to train advanced AI models doubles every 3.4 months, leading to exponential increase in power usage and carbon emissions.
Researchers from Zhejiang and Nankai Universities in China conducted a study on carbon emissions from AI systems, analysing 79 major AI systems from 2020 to 2024. They found these systems could collectively emit over 102 million tons of CO₂ per year. The study calls for regulations and standardisation to reduce carbon emissions from AI systems.
AI’s Unquenchable Thirst
Cooling AI servers requires massive amounts of fresh water. Do you know that generating 20 ChatGPT responses uses water as much as a 500ml bottle! As of October 2024, OpenAI uses over 2 litres of water per 50 queries to keep its data centre cool.
Researchers at the University of California, Riverside, found that Microsoft used nearly 700,000 litters of fresh water to cool its data centres during GPT-3’s training—enough to produce over 370 BMWs or 320 Teslas.
Many AI-focused data centres rely on liquid cooling. Though it lowers power consumption but uses large amounts of water. A survey revealed that liquid cooling in data centres has increased significantly from 7% in 2021 to 22% in 2024.
To reduce water consumption, some major companies are building data centres in colder regions for natural cooling. For example, Google established a data centre in Hamina, Finland, in 2009, and Meta (Facebook) followed with one in Luleå, Sweden, in 2011.
Companies like Microsoft, Google, and Interxion have also started using saltwater for cooling to lessen the strain on freshwater supplies. However, the adoption rate is still low. Plus, there’s a limit to how much saltwater can be used before we’re literally draining the oceans—all just to make ourselves look cuter with Ghibli-styled images!
Mining Rare Earth Elements
Rare earth elements are a group of 17 metallic elements with magnetic, luminescent, and electrochemical properties. They are vital for data storage and powering the processors that drive complex learning systems in AI. Microchips for AI models are crafted using these elements, which come from mining processes that are harmful to environment.
According to the Digital Economy Report 2024, producing a 2 kg computer requires extracting 800 kg of raw materials. Extraction of raw materials for manufacturing GPUs has serious environmental consequences. It involves dirty mining practices and use of toxic chemicals for processing. The damage doesn’t stop there—mining disturbs large land areas, causing erosion and habitat loss, while the toxic chemicals used in processing pollute water sources, creating long-term ecological damage.
Growing Electronic Waste
Every time you throw an old gadget; it becomes part of the 70% of the world’s toxic e-waste—and only 12.5% of it gets recycled. The rest? It leaks hazardous lead and mercury into the environment, turning landfills into ticking time bombs for ecosystems and health.
Generative AI isn’t just creating art and text—it’s also crafting a mountain of e-waste. By 2030, this could reach 1.2 to 5 million metric tons, about 1,000 times more than in 2023, according to research in Nature Computational Science. Why? Because AI models like LLMs need constant hardware upgrades to keep up with demand. Every new version means old servers become electronic junk, adding to the growing e-waste pile.
The World Economic Forum projects that by 2050, e-waste will surpass 120 million metric tons! So, while your AI-generated image might look stunning, it’s also leaving behind a not-so-pretty digital footprint.
Lack of Transparency and Accountability
The realm of AI development and usage lacks transparency and accountability regarding its environmental impact. Most companies prioritise financial gains and competitive advantages over addressing the potential negative effects of AI technologies on the environment.
Users struggle to understand the environmental footprint of AI due to the complexity of these systems.
Green AI: A Sustainable Tech Future
As the environmental impact of generative AI grows at an alarming rate, a new approach known as “Green AI” has emerged. Tech companies are now seeking ways to make AI more sustainable. Strategies like developing energy-efficient AI models, using renewable energy sources and improving cooling systems in data centres can help reduce its environmental impact. The key challenge lies in balancing the user benefits of AI with the need to minimize its ecological footprint.
Here are some effective ways to make AI greener:
- Energy-Efficient Algorithms –AI systems can significantly reduce energy consumption by designing energy-efficient algorithms. This includes developing streamlined network architectures and optimizing existing algorithms for lower energy use.
- Sustainable Energy Sources: Powering data centres with renewable energy sources like solar or wind can greatly diminish the environmental impact of AI operations.
- Improved Cooling in Data Centres: Implementing more energy-efficient cooling technologies can reduce the energy required to maintain optimal temperatures. For instance, Microsoft is exploring underwater data centre systems to enhance efficiency.
- E-Waste Management and Recycling: Proper management and recycling of e-waste are crucial to prevent environmental damage. Stricter regulations and ethical disposal practices are needed to handle AI-related e-waste responsibly.
- Increase Transparency and Accountability: AI researchers should disclose not only performance and accuracy metrics but also the energy consumption involved in AI model development. Greater transparency and regulations can ensure that AI development aligns with environmental goals.
- Energy Star Rating for AI: If an appliance can have an Energy Star rating, why can’t the AI model have it? Energy Star ratings for AI models would help consumers identify environmentally sustainable options and raise awareness about AI’s energy impact.
How Can You Help?
While systemic changes are crucial, as consumers, we can also play a role in reducing the carbon footprint of our internet usage. Here are some strategies to consider:
- Digital Sobriety: Embrace digital sobriety by reducing your overall technology usage. Less screen time means lower energy consumption, contributing to a more sustainable digital environment.
- Use Less Energy-Intensive AI: Opt for AI tools and services that are less energy-demanding. This means prioritizing text-based interactions over video or generative AI, which require more computational power.
- Enable Energy-Saving Settings: Adjust energy-saving settings on your devices. Lower your screen brightness, switch to dark mode, and activate power-saving features to reduce energy use.
- Prefer Wi-Fi Over Mobile Data: Use Wi-Fi instead of 4G or 5G whenever possible, as Wi-Fi generally consumes less energy, especially for data-intensive activities.
- Extend the Life of Your Devices: Prolong the use of your current hardware to delay the need for new equipment. This helps reduce e-waste and the environmental impact associated with manufacturing new devices.
Melting Our Planet, unknowingly!
Every time we create AI-generated content like the Ghibli-style images or any other art, the fun might be short-term but the environmental cost lingers long after.
Our small actions collectively have a bigger ripple effect on the environment. Remember we are not just melting those GPUs but, unknowingly, Melting Our Planet!
So next time you hit “generate,” ask yourself: Is this the image of me or the image of our planet, melting under the pressure?
——————————————————————————————————————–
