One Day of AI Video Generation Undid a Forest's Work
Summary
OpenAI's Sora, before its March 2026 shutdown, incurred an estimated daily operational cost of ~\$15 million, generating approximately 11.5 million 10-second video clips per day at its peak. This volume translated to over 23 million kilowatt-hours daily, resulting in about 9,200 tonnes of CO2 emissions in a single day, equivalent to the annual carbon absorption of 420,000 trees or 800 hectares of forest. The article contrasts this with more efficient text and image generation, noting Google's 2025 disclosure of 0.24 watt-hours per Gemini text query and Stable Diffusion 3 Medium's 1,141 joules per image. A 2024 HotCarbon study revealed video generation's non-linear energy scaling, where a six-second clip consumes four times the energy of a three-second one, compounded by 3 to 10 iterations per usable clip. The analysis attributes this environmental burden to product design choices that prioritize generation volume and frictionless user experience over efficiency, rather than individual user behavior.
Key takeaway
For AI Product Managers or Directors of AI/ML designing generative AI applications, you must integrate environmental cost visibility and efficiency defaults into your product experience. Prioritize matching model tiers to specific tasks and encourage editing over regeneration to reduce significant CO2 emissions. Your design choices directly impact the environmental footprint, so implement features like prompt caching and rate-limiting to foster restraint and avoid unnecessary compute spend.
Key insights
AI video generation's high energy cost is driven by product design, not user intent, with significant environmental impact.
Principles
- AI product design prioritizes generation volume.
- Environmental costs are often invisible to users.
- Video generation energy scales non-linearly.
In practice
- Match model tier to task complexity.
- Prioritize editing over full regeneration.
- Utilize prompt caching for repeated contexts.
Topics
- AI Environmental Impact
- Generative AI Costs
- AI Video Generation
- Data Center Energy
- Model Efficiency
- Product Design
Best for: AI Engineer, Director of AI/ML, AI Product Manager
Related on AIssential
See Counsel's argued verdicts on the open AI decisions leaders are weighing →
Editorial summary, takeaway, and curation by AIssential. Original article published by HackerNoon.