Sora Environmental Impact
10-second video generation — extremely energy intensive
- Architecture
- Diffusion Transformer (spatiotemporal)
- Provider
- OpenAI
Energy per query
500.0 Wh
1667x more than a Google search (0.3 Wh)
CO2 per query
225.0 g
US East (Virginia) grid (450 gCO₂/kWh)
Water per query
1.85 L
~1 queries to fill 1 litre
Processing location
Azure US East / Sweden
Provider
OpenAI
Category
Video Generation
Grid carbon intensity
450 g CO2/kWh (25% renewable)
How does Sora compare?
Detailed Breakdown
Energy Consumption
Sora consumes approximately 500 Wh per 10-second video clip — our mid-range estimate between the WSJ figure of 20-100 Wh and the Alexander (2025) upper bound of 1,000 Wh (1 kWh). This is roughly 1,650x more energy than a standard text query. Video generation requires running a diffusion transformer model across both spatial and temporal dimensions, generating coherent frames at high resolution. Each video may require multiple NVIDIA H100 GPUs running for several minutes. Energy use quadruples when video length doubles.
Power Source & Carbon
Sora runs on Microsoft Azure's GPU clusters, likely requiring dedicated high-density compute nodes with multiple H100 GPUs per video. The extreme power density required for video generation means these workloads are concentrated in Azure's largest data centers. Microsoft's overall electricity consumption nearly tripled between 2020 and 2024, reaching 29.8 million MWh.
Water Usage
At approximately 1,850 mL (~1.85 liters) per 10-second video, Sora's water consumption is substantial — equivalent to a large bottle of water per clip. This is a direct consequence of the extreme GPU compute time and the resulting heat that must be dissipated through cooling systems. Generating just 10 short videos would consume over 18 liters of water.
About Sora
Sora is the extreme end of AI energy consumption — and a preview of where the industry is heading. A single 10-second video clip consumes an estimated 500 Wh, roughly the same as running a refrigerator for half a day. That figure is inherently uncertain (estimates range from 20 Wh to 1 kWh) because OpenAI has disclosed nothing about Sora's inference costs, but even the low-end estimate makes it over 50x more intensive than generating an image. Video generation is not an incremental step from text or image AI — it is a different order of magnitude, and its growing popularity has serious implications for data centre energy demand.
These figures are estimates derived from hardware specifications and API benchmarks — OpenAI has not published official energy data for Sora. Actual consumption may vary significantly depending on batching, quantisation, and infrastructure optimisations that we cannot observe from outside.
Sora in Context
The video generation premium
One Sora generation uses 500.0 Wh — the equivalent of 42 full smartphone charges. Video generation requires running a diffusion model across both spatial and temporal dimensions, making it fundamentally more compute-intensive than text or image tasks. This is not an efficiency problem to be solved; it is an inherent characteristic of the task.
What does your Sora usage cost the planet?
Use our calculator to estimate your personal environmental footprint based on how often you use Sora.
Calculate My ComputeFrequently Asked Questions
How much energy does Sora use per query?
Each Sora query consumes approximately 500.0 Wh of energy. This is 1667x more than a traditional Google search (~0.3 Wh).
What is Sora's carbon footprint?
Based on the carbon intensity of Azure US East / Sweden, each query produces approximately 225.0 g of CO2. The grid in this region has a carbon intensity of 450 g CO2/kWh with 25% renewable energy.
How much water does Sora use?
Each query consumes approximately 1.85 L of water, primarily used for cooling the data centers that process the request.
How does Sora compare to a Google search?
A Sora query uses 1667x more than a Google search in terms of energy. A Google search uses approximately 0.3 Wh, while Sora uses 500.0 Wh.
Technical Details
Architecture
Diffusion Transformer (spatiotemporal)
Release date
2024-12-09
Open source
No