.jpg)
Your Instagram feed has quietly become a time machine. Somewhere between the morning work updates and the weekend check-ins, there are portraits that look like they were pulled from a 1980s family album , big hair, warm studio lighting, soft film grain and that slightly faded colour that only old photographs have. The person in the photo is someone you recognise. The world around them is four decades away.
This is the 80s AI photo trend and it has taken over Instagram Stories in the past week. People are uploading a current selfie to ChatGPT, describing the era they want, and getting back a version of themselves that never existed but looks entirely convincing.
It is not a filter. That is the part worth understanding. A vintage filter fades your existing photo. This trend asks AI to rebuild the entire image. The hairstyle, the clothing, the jewellery, the background, the lighting , all of it is recreated from scratch. The AI keeps your face and expression roughly intact while replacing everything around it with period detail.
Every one of those images costs something most people do not think about when they hit generate : water.
AI image generation is computationally heavier than a text query. The servers running these models get extremely hot and the most common way to cool them is to pump large volumes of water through the facility. According to a study by the University of California, Riverside, each AI-generated image consumes between 2 and 5 litres of water. That is not the electricity cost. That is the water, evaporated into the air to keep the chips from burning out.
Scale that to a trend with millions of participants and the numbers shift from interesting to uncomfortable.
The Council on Energy, Environment and Water (CEEW) estimated that India's data centres consumed approximately 150 billion litres of water in 2025. That figure is already projected to more than double by 2030. A typical 100 MW hyperscale data centre can consume around 20 lakh litres a day for cooling alone.
The geography makes this harder to dismiss. Three-quarters of India's data centres are concentrated in five states Maharashtra, Tamil Nadu, Karnataka, Telangana, and Uttar Pradesh. Maharashtra alone is expected to account for nearly 45 percent of the country's planned data-centre power capacity, with Mumbai as the central hub. Chennai carries a significant share of the rest. Both cities are already in trouble with water. Meanwhile, 330 million people across India already live under water scarcity conditions.
There is also a less visible layer. Data centres are not only consumers of water on-site. They draw from the same electricity grid as everyone else, and the coal and gas plants generating that electricity are themselves large water users. The indirect water cost of running AI infrastructure can exceed the direct cooling cost at the facility itself.
None of this means you should feel guilty about one 80s portrait. A single image is not the problem. The problem is that the infrastructure making these tools available is growing at a speed that water planning in India has not caught up with, and most of that infrastructure is landing in the cities least equipped to absorb the additional demand.
Sources: Council on Energy, Environment and Water (CEEW), How Is Data Centre Infrastructure in India Shaping Power and Water Use?, 2026. WRI India, More Than Half of India's Data Centres Are in Water-Stressed Regions, 2026. University of California, Riverside, Making AI Less Thirsty, 2023. Observer Research Foundation, The Cloud Runs on Water, 2026. Countercurrents, AI Data Centres vs. Water Crisis, 2026. NewSX, India's AI Data Centre Boom: How Much Water Do They Use?, 2026.
