Big hair, acid-wash denim, dramatic silhouettes and grainy film effects are suddenly taking over social media as users turn their modern photographs into 1980s-style portraits with AI. The trend, which has spread across Instagram and other platforms, follows the earlier wave of Studio Ghibli-style AI images and other viral generative-AI experiments. Users can upload a photograph, describe the desired retro look and receive a transformed image within seconds.
It looks effortless. The infrastructure behind it is anything but.
Your AI Selfie Needs More Than A Prompt
Every AI-generated image is processed by computers in data centres, where powerful chips perform the calculations required to create the final picture. Those facilities need electricity to run the servers and additional resources for cooling. The exact environmental cost of one image varies considerably depending on the model, hardware, resolution and other factors. Research published in 2025 found differences of up to 46 times between image-generation models, highlighting why there is no single reliable number for the footprint of an AI picture.
Research involving Carnegie Mellon University and Hugging Face has also found that generative AI tasks can consume substantially more energy than specialised software performing comparable tasks.
That variation is important. It means claims that every AI image consumes a fixed amount of electricity should be treated cautiously.
The Bigger Problem Is Scale
One retro portrait is unlikely to make a noticeable difference. Millions of users repeatedly generating, editing and regenerating images are a different story. The International Energy Agency estimates that data centres consumed about 415 TWh of electricity globally in 2024, around 1.5% of worldwide electricity consumption. Its base case projects that figure will more than double to approximately 945 TWh by 2030, with AI identified as the biggest driver of the increase.
That does not mean the entire increase can be blamed on Instagram trends. Data centres support everything from cloud storage and streaming to websites, business software and AI services. But the rapid expansion of AI is adding another layer of demand.
AI’s Water Problem Is Even Less Visible
Electricity is only part of the equation. Data centres also need cooling, and water can be involved both directly in cooling systems and indirectly through electricity generation. Research from the University of California, Riverside found that GPT-3 could consume roughly 500 ml of water for around 20 to 50 medium-length queries, depending on where and when the computing took place. The figure relates to an older AI model and should not be treated as the water cost of every modern AI interaction.
The bigger concern is scale and location. Water use that looks small per interaction can become significant when AI services are used billions of times, particularly in regions already facing water stress.
So, Should You Stop Using AI?
Not necessarily. The environmental question is less about one 1980s portrait and more about how casually we use generative AI at scale. AI can be genuinely useful for research, accessibility, productivity and creative work. But generating 20 slightly different versions of the same selfie simply because a trend is everywhere has a different value proposition.
The 1980s trend will eventually disappear from our feeds. The data centres powering it will not. As AI becomes part of everyday digital life, understanding that hidden footprint may be just as important as enjoying the technology itself.
Shiwani works as a Senior Sub Editor at NewsX, covering the entertainment beat. With a strong background in media, she is a true cinema buff who loves keeping up with the latest in pop culture and lifestyle trends. After earning her Bachelor’s degree in Commerce, Shiwani followed her passion and pursued a Master’s in Journalism from IIMC, choosing the road less travelled.
