When we talk about the environmental footprint of digital technologies, streaming services and cryptocurrency mining usually dominate the conversation. But a new wave of research is showing that generative AI — particularly video generators — may be demanding far more electricity than expected.
A recent study by Hugging Face researchers revealed that the energy requirements of text-to-video systems scale dramatically as clips get longer. Doubling the video length doesn’t just double the power draw — it quadruples it. A six-second video requires roughly four times the energy of a three-second one. That exponential growth paints a troubling picture for the future of AI-driven media creation.
As the team noted, the structural inefficiency of current video diffusion pipelines is a major factor. Generating images is already energy-intensive, with a single 1,024 x 1,024 output burning through about as much power as running a microwave for five seconds. But scaling to video takes things to another level. Producing just a five-second AI video clip consumes the equivalent of running a microwave for more than an hour. And the longer the output, the worse the curve gets.
These findings, reported in Futurism, suggest “rapidly increasing hardware and environmental costs” as generative video tools become mainstream. With companies like Google boasting tens of millions of AI-generated clips from products such as Veo 3, the aggregate impact is hard to ignore. In its own 2024 environmental report, Google acknowledged that its carbon emissions rose 13 percent year over year, driven in large part by its investment in AI infrastructure — a setback to its pledge of net-zero emissions by 2030.
The problem isn’t without potential mitigations. Techniques like intelligent caching, reusing generations, and pruning inefficient data can cut some of the waste. But whether those optimizations will meaningfully offset the surge in power consumption is still uncertain. Already, AI accounts for about 20 percent of global datacenter energy use, according to recent estimates.
As demand for longer, higher-quality AI-generated content grows, the urgency to redesign these systems around efficiency will only intensify. Without a shift in how these tools are built and deployed, the environmental costs of synthetic media may prove far greater than most of us realize.