By- Dr. Kammula Sunil Kumar
Assistant Professor, Department of Computer Science & Engineering, SRM University – AP (Amaravati)
It takes about ten seconds: upload a selfie, ask for an AI tool to make it look like 1986, and watch teased hair, neon lighting, and a mall-photo-studio backdrop appear around a face that is unmistakably yours. Since entering India through ChatGPT in September 2026, after first circulating in China, the trend has spread with unusual speed, Google Trends recording over five lakh searches for the specific prompt inside India alone, a growth rate of roughly 1,000 percent in a matter of days. Actors, politicians, and ordinary users alike joined in, turning a niche AI feature into one of the year’s most visible pieces of shared culture. What almost none of those ten-second uploads carried was any sense that each one drew, however briefly, on real electricity and real water somewhere in a data centre most users will never see or think about.
The psychology behind the trend’s speed is not hard to explain. Researchers who study nostalgia have found that a striking share of Gen Z, roughly two-thirds by some estimates, feel genuinely drawn to eras they never lived through, a longing for a version of the past assembled entirely from films, photographs, and other people’s memories. AI image tools offered a way to briefly inhabit that imagined past personally, not as a costume or a filter, but as a photograph that looks convincingly real. That emotional pull, rather than any deliberate marketing push, is largely what carried the trend from a handful of viral posts in China to a Google Trends spike across India within days. The resource question sits underneath that emotional appeal, and it deserves a more careful accounting than the alarming round numbers that tend to circulate online. Widely shared claims that a single AI image consumes something like ten gallons of water have been shown, on closer examination of the underlying research, to be exaggerated by a hundred times or more. The more defensible figures, drawn from peer-reviewed and industry research through 2026, put a standard AI-generated image at roughly 2.9 watt-hours of electricity and somewhere between five and thirty millilitres of water, once the water used to generate that electricity is factored in, an amount closer to a couple of tablespoons than a bottle, let alone ten gallons of it. On its own, that is a genuinely small number, smaller than the electricity a phone draws just charging for a few minutes.
The honest concern, then, is not the individual image but the arithmetic of repetition at scale. A user rarely accepts the first output; five or ten regenerations, adjusting the hairstyle, the backdrop, the lighting, are common before someone is satisfied enough to post, and that pattern repeats across a user base too large to count precisely. Stability AI alone has reported that its models generated more than 12.7 billion images in a single year, a scale at which even a few thousandths of a kilowatt-hour per image adds up to the annual electricity use of thousands of homes. The United Nations University has separately estimated that everyday use of AI systems, rather than the one-time cost of training them, now accounts for roughly 80 to 90 percent of their total energy demand, a reminder that the infrastructure strain from a viral trend is not a one-off spike but a preview of AI’s ordinary, permanent running cost.
Placed against the industry’s broader trajectory, these numbers stop looking like a curiosity confined to one viral week. Research published this year projects that AI’s global water footprint could reach between 4.2 and 6.6 billion cubic metres a year by 2027 if current growth continues, drawing on water used for cooling data centres, water embedded in the electricity that powers them, and water consumed in manufacturing the chips themselves. None of this is unique to a nostalgia trend; the same infrastructure processes medical research, weather forecasting, and financial systems that societies clearly consider worth the resource cost. What a moment like this does, precisely because the underlying task is so trivial, is make the resource cost visible in a way that genuinely useful AI applications rarely do, since nobody pauses to calculate the water footprint of a life-saving diagnosis.
That asymmetry, between how casually AI is used for entertainment and how invisibly its running costs are absorbed, is arguably the more interesting story than the trend itself. Few people would decline a fun photo because it costs a spoonful of water elsewhere in the world; that math genuinely favours indulgence. But a habit of treating AI compute as a free, weightless resource, formed while playing with retro filters, tends to carry over into how casually the same compute gets used for far less meaningful things later, generating ten versions of an email or a slide deck nobody needed refined that many times. The 1980s photo trend will fade the way viral trends always do. What it leaves behind, if anyone chooses to notice it, is a rare glimpse of the electricity bill and water bill sitting quietly behind every AI request, whether or not the person making it ever looks.
